docs: make Chinese README the default

CMakeLists.txt

@@ -10,6 +10,7 @@ add_library(vosk
   src/recognizer.cc
   src/spk_model.cc
   src/vosk_api.cc
+  src/postprocessor.cc
 )
 
 find_package(kaldi REQUIRED)

README.en.md

@@ -0,0 +1,27 @@
+# Vosk Speech Recognition Toolkit
+
+Vosk is an offline open source speech recognition toolkit. It enables
+speech recognition for 20+ languages and dialects - English, Indian
+English, German, French, Spanish, Portuguese, Chinese, Russian, Turkish,
+Vietnamese, Italian, Dutch, Catalan, Arabic, Greek, Farsi, Filipino,
+Ukrainian, Kazakh, Swedish, Japanese, Esperanto, Hindi, Czech, Polish.
+More to come.
+
+Vosk models are small (50 Mb) but provide continuous large vocabulary
+transcription, zero-latency response with streaming API, reconfigurable
+vocabulary and speaker identification.
+
+Speech recognition bindings implemented for various programming languages
+like Python, Java, Node.JS, C#, C++, Rust, Go and others.
+
+Vosk supplies speech recognition for chatbots, smart home appliances,
+virtual assistants. It can also create subtitles for movies,
+transcription for lectures and interviews.
+
+Vosk scales from small devices like Raspberry Pi or Android smartphone to
+big clusters.
+
+# Documentation
+
+For installation instructions, examples and documentation visit [Vosk
+Website](https://alphacephei.com/vosk).

README.md

@@ -1,27 +1,21 @@
+<!-- WEHUB_ZH_README -->
+> [!NOTE]
+> 本文档由 WeHub 基于上游 README 翻译整理，属于社区翻译，非官方中文文档。  
+> [English](./README.en.md) · [原始项目](https://github.com/alphacep/vosk-api) · [上游 README](https://github.com/alphacep/vosk-api/blob/HEAD/README.md)  
+> 原作者、版权与许可证归属以原始项目及本仓库 LICENSE 文件为准。
+
 # Vosk Speech Recognition Toolkit
 
-Vosk is an offline open source speech recognition toolkit. It enables
-speech recognition for 20+ languages and dialects - English, Indian
-English, German, French, Spanish, Portuguese, Chinese, Russian, Turkish,
-Vietnamese, Italian, Dutch, Catalan, Arabic, Greek, Farsi, Filipino,
-Ukrainian, Kazakh, Swedish, Japanese, Esperanto, Hindi, Czech, Polish.
-More to come.
+Vosk 是一款离线开源语音识别工具包。它支持 20 多种语言及方言的语音识别——英语、印度英语、德语、法语、西班牙语、葡萄牙语、中文、俄语、土耳其语、越南语、意大利语、荷兰语、加泰罗尼亚语、阿拉伯语、希腊语、波斯语、菲律宾语、乌克兰语、哈萨克语、瑞典语、日语、世界语、印地语、捷克语、波兰语。更多语言即将推出。
 
-Vosk models are small (50 Mb) but provide continuous large vocabulary
-transcription, zero-latency response with streaming API, reconfigurable
-vocabulary and speaker identification.
+Vosk 模型体积小（50 Mb），但可提供连续大词汇量转写、通过流式 API（streaming API）实现零延迟响应、可重新配置的词汇表以及说话人识别。
 
-Speech recognition bindings implemented for various programming languages
-like Python, Java, Node.JS, C#, C++, Rust, Go and others.
+已为 Python、Java、Node.JS、C#、C++、Rust、Go 等多种编程语言实现语音识别绑定。
 
-Vosk supplies speech recognition for chatbots, smart home appliances,
-virtual assistants. It can also create subtitles for movies,
-transcription for lectures and interviews.
+Vosk 为聊天机器人、智能家居设备、虚拟助手提供语音识别，还可为电影生成字幕，为讲座和访谈提供转写。
 
-Vosk scales from small devices like Raspberry Pi or Android smartphone to
-big clusters.
+Vosk 的部署规模可从小型设备（如 Raspberry Pi 或 Android 智能手机）扩展到大型集群。
 
-# Documentation
+# 文档
 
-For installation instructions, examples and documentation visit [Vosk
-Website](https://alphacephei.com/vosk).
+有关安装说明、示例和文档，请访问 [Vosk 网站](https://alphacephei.com/vosk).

README.wehub.md

@@ -0,0 +1,7 @@
+# WeHub 来源说明
+
+- 原始项目：`alphacep/vosk-api`
+- 原始仓库：https://github.com/alphacep/vosk-api
+- 导入方式：上游默认分支的最新快照
+- 原作者、版权和许可证信息以原始仓库及本仓库 LICENSE 为准
+- 本文件仅用于记录来源，不代表 WeHub 是原项目作者

android/build.gradle

@@ -4,62 +4,50 @@ buildscript {
         mavenCentral()
     }
     dependencies {
-        classpath 'com.android.tools.build:gradle:4.2.0'
-        classpath 'com.vanniktech:gradle-maven-publish-plugin:0.18.0'
+        classpath 'com.android.tools.build:gradle:8.13.0'
+        classpath 'com.vanniktech:gradle-maven-publish-plugin:0.34.0'
     }
 }
 
 allprojects {
-    version = '0.3.45'
+    version = '0.3.75'
 }
 
 subprojects {
 
     apply plugin: 'com.android.library'
-    apply plugin: 'maven-publish'
     apply plugin: 'com.vanniktech.maven.publish'
 
-    plugins.withId('com.vanniktech.maven.publish') {
-        mavenPublish {
-            group = 'com.alphacephei'
-            version = version
-            sonatypeHost = 's01'
-            androidVariantToPublish = 'release'
-        }
-    }
-
     repositories {
         google()
         mavenCentral()
     }
 
-    publishing {
-        publications {
-            aar(MavenPublication) {
-                groupId 'com.alphacephei'
-                version version
-                pom {
-                    url = 'http://www.alphacephei.com.com/vosk/'
-                    licenses {
-                        license {
-                            name = 'The Apache License, Version 2.0'
-                            url = 'http://www.apache.org/licenses/LICENSE-2.0.txt'
-                        }
-                    }
-                    developers {
-                        developer {
-                            id = 'com.alphacephei'
-                            name = 'Alpha Cephei Inc'
-                            email = 'contact@alphacephei.com'
-                        }
-                    }
-                    scm {
-                        connection = 'scm:git:git://github.com/alphacep/vosk-api.git'
-                        url = 'https://github.com/alphacep/vosk-api/'
-                    }
+    mavenPublishing {
+        publishToMavenCentral()
+        signAllPublications()
+    }
+
+    mavenPublishing {
+        pom {
+            url = 'http://www.alphacephei.com.com/vosk/'
+            licenses {
+                license {
+                    name = 'The Apache License, Version 2.0'
+                    url = 'http://www.apache.org/licenses/LICENSE-2.0.txt'
                 }
             }
+            developers {
+                developer {
+                    id = 'com.alphacephei'
+                    name = 'Alpha Cephei Inc'
+                    email = 'contact@alphacephei.com'
+                }
+            }
+            scm {
+                connection = 'scm:git:git://github.com/alphacep/vosk-api.git'
+                url = 'https://github.com/alphacep/vosk-api/'
+            }
         }
     }
-
 }

android/lib/build-vosk.sh

@@ -29,7 +29,7 @@ set -x
 OS_NAME=`echo $(uname -s) | tr '[:upper:]' '[:lower:]'`
 ANDROID_TOOLCHAIN_PATH=$ANDROID_NDK_HOME/toolchains/llvm/prebuilt/${OS_NAME}-x86_64
 WORKDIR_BASE=`pwd`/build
-PATH=$ANDROID_NDK_HOME/toolchains/llvm/prebuilt/${OS_NAME}-x86_64/bin:$PATH
+PATH=$ANDROID_TOOLCHAIN_PATH/bin:$PATH
 OPENFST_VERSION=1.8.0
 
 for arch in armeabi-v7a arm64-v8a x86_64 x86; do
@@ -45,6 +45,7 @@ case $arch in
           CC=armv7a-linux-androideabi21-clang
           CXX=armv7a-linux-androideabi21-clang++
           ARCHFLAGS="-mfloat-abi=softfp -mfpu=neon"
+          PAGESIZE_LDFLAGS=""
           ;;
     arm64-v8a)
           BLAS_ARCH=ARMV8
@@ -54,6 +55,8 @@ case $arch in
           CC=aarch64-linux-android21-clang
           CXX=aarch64-linux-android21-clang++
           ARCHFLAGS=""
+          # Ensure compatibility with 16KiB page size devices
+          PAGESIZE_LDFLAGS="-Wl,-z,common-page-size=4096 -Wl,-z,max-page-size=16384"
           ;;
     x86_64)
           BLAS_ARCH=ATOM
@@ -63,6 +66,7 @@ case $arch in
           CC=x86_64-linux-android21-clang
           CXX=x86_64-linux-android21-clang++
           ARCHFLAGS=""
+          PAGESIZE_LDFLAGS=""
           ;;
     x86)
           BLAS_ARCH=ATOM
@@ -72,6 +76,7 @@ case $arch in
           CC=i686-linux-android21-clang
           CXX=i686-linux-android21-clang++
           ARCHFLAGS=""
+          PAGESIZE_LDFLAGS=""
           ;;
 esac
 
@@ -79,16 +84,16 @@ mkdir -p $WORKDIR/local/lib
 
 # openblas first
 cd $WORKDIR
-git clone -b v0.3.13 --single-branch https://github.com/xianyi/OpenBLAS
-make -C OpenBLAS TARGET=$BLAS_ARCH ONLY_CBLAS=1 AR=$AR CC=$CC HOSTCC=gcc ARM_SOFTFP_ABI=1 USE_THREAD=0 NUM_THREADS=1 -j4
+git clone -b v0.3.20 --single-branch https://github.com/xianyi/OpenBLAS
+make -C OpenBLAS TARGET=$BLAS_ARCH ONLY_CBLAS=1 AR=$AR CC=$CC HOSTCC=gcc ARM_SOFTFP_ABI=1 USE_THREAD=0 NUM_THREADS=1 -j 8
 make -C OpenBLAS install PREFIX=$WORKDIR/local
 
 # CLAPACK
 cd $WORKDIR
 git clone -b v3.2.1  --single-branch https://github.com/alphacep/clapack
 mkdir -p clapack/BUILD && cd clapack/BUILD
-cmake -DCMAKE_C_FLAGS=$ARCHFLAGS -DCMAKE_C_COMPILER_TARGET=$HOST \
-    -DCMAKE_C_COMPILER=$CC -DCMAKE_SYSTEM_NAME=Generic -DCMAKE_AR=$ANDROID_NDK_HOME/toolchains/llvm/prebuilt/${OS_NAME}-x86_64/bin/$AR \
+cmake -DCMAKE_C_FLAGS="$ARCHFLAGS" -DCMAKE_C_COMPILER_TARGET=$HOST \
+    -DCMAKE_C_COMPILER=$CC -DCMAKE_SYSTEM_NAME=Generic -DCMAKE_AR=$ANDROID_TOOLCHAIN_PATH/bin/$AR \
     -DCMAKE_TRY_COMPILE_TARGET_TYPE=STATIC_LIBRARY \
     -DCMAKE_CROSSCOMPILING=True ..
 make -j 8 -C F2CLIBS/libf2c
@@ -118,7 +123,7 @@ CXX=$CXX AR=$AR RANLIB=$RANLIB CXXFLAGS="$ARCHFLAGS -O3 -DFST_NO_DYNAMIC_LINKING
     --fst-root=${WORKDIR}/local --fst-version=${OPENFST_VERSION}
 make -j 8 depend
 cd $WORKDIR/kaldi/src
-make -j 8 online2 lm rnnlm
+make -j 8 online2 rnnlm
 
 # Vosk-api
 cd $WORKDIR
@@ -129,7 +134,7 @@ make -j 8 -C ${WORKDIR_BASE}/../../../src \
     OPENFST_ROOT=${WORKDIR}/local \
     OPENBLAS_ROOT=${WORKDIR}/local \
     CXX=$CXX \
-    EXTRA_LDFLAGS="-llog -static-libstdc++ -Wl,-soname,libvosk.so"
+    EXTRA_LDFLAGS="-llog -static-libstdc++ -Wl,-soname,libvosk.so ${PAGESIZE_LDFLAGS}"
 cp $WORKDIR/vosk/libvosk.so $WORKDIR/../../src/main/jniLibs/$arch/libvosk.so
 
 done

android/lib/build.gradle

@@ -3,14 +3,15 @@ def pomName = "Vosk Android"
 def pomDescription = "Vosk speech recognition library for Android"
 
 android {
-    compileSdkVersion 29
+    namespace 'org.vosk'
+    compileSdkVersion 36
     defaultConfig {
         minSdkVersion 21
-        targetSdkVersion 29
-        versionCode 6
+        targetSdkVersion 36
+        versionCode 10
         versionName = version
         archivesBaseName = archiveName
-        ndkVersion = "22.1.7171670"
+        ndkVersion = "28.2.13676358"
     }
     compileOptions {
         sourceCompatibility JavaVersion.VERSION_1_8
@@ -24,32 +25,15 @@ task buildVosk(type: Exec) {
 }
 
 dependencies {
-    api 'net.java.dev.jna:jna:4.4.0@aar'
+    api 'net.java.dev.jna:jna:5.18.1@aar'
 }
 
 //preBuild.dependsOn buildVosk
 
-publishing {
-    publications {
-        aar(MavenPublication) {
-            artifactId = archiveName
-            artifact("$buildDir/outputs/aar/$archiveName-release.aar")
-            pom {
-                name = pomName
-                description = pomDescription
-            }
-            //generate pom nodes for dependencies
-            pom.withXml {
-                def dependenciesNode = asNode().appendNode('dependencies')
-                configurations.implementation.allDependencies.each { dependency ->
-                    if (dependency.name != 'unspecified') {
-                        def dependencyNode = dependenciesNode.appendNode('dependency')
-                        dependencyNode.appendNode('groupId', dependency.group)
-                        dependencyNode.appendNode('artifactId', dependency.name)
-                        dependencyNode.appendNode('version', dependency.version)
-                    }
-                }
-            }
-        }
+mavenPublishing {
+    coordinates("com.alphacephei", archiveName, version)
+    pom {
+        name = pomName
+        description = pomDescription
     }
 }

android/lib/src/main/AndroidManifest.xml

@@ -1,3 +1,3 @@
 <?xml version="1.0" encoding="utf-8"?>
-<manifest xmlns:android="http://schemas.android.com/apk/res/android" package="org.vosk">
+<manifest>
 </manifest>

android/lib/src/main/java/org/vosk/LibVosk.java

@@ -52,10 +52,30 @@ public class LibVosk {
 
     public static native String vosk_recognizer_partial_result(Pointer recognizer);
 
+    public static native void vosk_recognizer_set_grm(Pointer recognizer, String grammar);
+
     public static native void vosk_recognizer_reset(Pointer recognizer);
 
+    public static native void vosk_recognizer_set_endpointer_mode(Pointer recognizer, int mode);
+
+    public static native void vosk_recognizer_set_endpointer_delays(Pointer recognizer, float t_start_max, float t_end, float t_max);
+
     public static native void vosk_recognizer_free(Pointer recognizer);
 
+    public static native Pointer vosk_text_processor_new(String verbalizer, String tagger);
+
+    public static native void vosk_text_processor_free(Pointer processor);
+
+    public static native String vosk_text_processor_itn(Pointer processor, String input);
+
+    /**
+     * Set log level for Kaldi messages.
+     *
+     *  @param loglevel the level
+     *     0 - default value to print info and error messages but no debug
+     *     less than 0 - don't print info messages
+     *     greater than 0 - more verbose mode
+     */
     public static void setLogLevel(LogLevel loglevel) {
         vosk_set_log_level(loglevel.getValue());
     }

android/lib/src/main/java/org/vosk/Model.java

@@ -1,13 +1,18 @@
 package org.vosk;
 
+import java.io.IOException;
 import com.sun.jna.PointerType;
 
 public class Model extends PointerType implements AutoCloseable {
     public Model() {
     }
 
-    public Model(String path) {
+    public Model(String path) throws IOException {
         super(LibVosk.vosk_model_new(path));
+
+        if (getPointer() == null) {
+            throw new IOException("Failed to create a model");
+        }
     }
 
     @Override

android/lib/src/main/java/org/vosk/Recognizer.java

@@ -1,36 +1,163 @@
 package org.vosk;
 
 import com.sun.jna.PointerType;
+import java.io.IOException;
 
 public class Recognizer extends PointerType implements AutoCloseable {
-    public Recognizer(Model model, float sampleRate) {
+    /**
+     * Creates the recognizer object.
+     *
+     * The recognizers process the speech and return text using shared model data
+     * @param model       VoskModel containing static data for recognizer. Model can be
+     *                    shared across recognizers, even running in different threads.
+     * @param sampleRate The sample rate of the audio you are going to feed into the recognizer.
+     *                    Make sure this rate matches the audio content, it is a common
+     *                    issue causing accuracy problems.
+     * @throws IOException if the recognizer could not be created
+     */
+    public Recognizer(Model model, float sampleRate) throws IOException {
         super(LibVosk.vosk_recognizer_new(model, sampleRate));
+
+        if (getPointer() == null) {
+            throw new IOException("Failed to create a recognizer");
+        }
     }
 
-    public Recognizer(Model model, float sampleRate, SpeakerModel spkModel) {
+    /**
+     * Creates the recognizer object with speaker recognition.
+     *
+     *  With the speaker recognition mode the recognizer not just recognize
+     *  text but also return speaker vectors one can use for speaker identification
+     *
+     *  @param model       VoskModel containing static data for recognizer. Model can be
+     *                     shared across recognizers, even running in different threads.
+     *  @param sampleRate The sample rate of the audio you are going to feed into the recognizer.
+     *                     Make sure this rate matches the audio content, it is a common
+     *                     issue causing accuracy problems.
+     *  @param spkModel speaker model for speaker identification
+     *  @throws IOException if the recognizer could not be created
+     */
+    public Recognizer(Model model, float sampleRate, SpeakerModel spkModel) throws IOException {
         super(LibVosk.vosk_recognizer_new_spk(model.getPointer(), sampleRate, spkModel.getPointer()));
+
+        if (getPointer() == null) {
+            throw new IOException("Failed to create a recognizer");
+        }
     }
 
-    public Recognizer(Model model, float sampleRate, String grammar) {
+    /**
+     * Creates the recognizer object with the phrase list.
+     *
+     *  Sometimes when you want to improve recognition accuracy and when you don't need
+     *  to recognize large vocabulary you can specify a list of phrases to recognize. This
+     *  will improve recognizer speed and accuracy but might return [unk] if user said
+     *  something different.
+     *
+     *  Only recognizers with lookahead models support this type of quick configuration.
+     *  Precompiled HCLG graph models are not supported.
+     *
+     *  @param model       VoskModel containing static data for recognizer. Model can be
+     *                     shared across recognizers, even running in different threads.
+     *  @param sampleRate The sample rate of the audio you are going to feed into the recognizer.
+     *                     Make sure this rate matches the audio content, it is a common
+     *                     issue causing accuracy problems.
+     *  @param grammar The string with the list of phrases to recognize as JSON array of strings,
+     *                 for example "["one two three four five", "[unk]"]".
+     *  @throws IOException if the recognizer could not be created
+     */
+    public Recognizer(Model model, float sampleRate, String grammar) throws IOException {
         super(LibVosk.vosk_recognizer_new_grm(model.getPointer(), sampleRate, grammar));
+
+        if (getPointer() == null) {
+            throw new IOException("Failed to create a recognizer");
+        }
     }
 
+    /**
+     * Configures recognizer to output n-best results.
+     *
+     * <pre>
+     *   {
+     *      "alternatives": [
+     *          { "text": "one two three four five", "confidence": 0.97 },
+     *          { "text": "one two three for five", "confidence": 0.03 },
+     *      ]
+     *   }
+     * </pre>
+     *
+     * @param maxAlternatives - maximum alternatives to return from recognition results
+     */
     public void setMaxAlternatives(int maxAlternatives) {
         LibVosk.vosk_recognizer_set_max_alternatives(this.getPointer(), maxAlternatives);
     }
 
+    /** Enables words with times in the output
+     *
+     * <pre>
+     *   "result" : [{
+     *       "conf" : 1.000000,
+     *       "end" : 1.110000,
+     *       "start" : 0.870000,
+     *       "word" : "what"
+     *     }, {
+     *       "conf" : 1.000000,
+     *       "end" : 1.530000,
+     *       "start" : 1.110000,
+     *       "word" : "zero"
+     *     }, {
+     *       "conf" : 1.000000,
+     *       "end" : 1.950000,
+     *       "start" : 1.530000,
+     *       "word" : "zero"
+     *     }, {
+     *       "conf" : 1.000000,
+     *       "end" : 2.340000,
+     *       "start" : 1.950000,
+     *       "word" : "zero"
+     *     }, {
+     *       "conf" : 1.000000,
+     *       "end" : 2.610000,
+     *       "start" : 2.340000,
+     *       "word" : "one"
+     *     }],
+     * </pre>
+     *
+     * @param words - boolean value
+     */
     public void setWords(boolean words) {
         LibVosk.vosk_recognizer_set_words(this.getPointer(), words);
     }
 
+    /**
+     * Like above return words and confidences in partial results.
+     *
+     * @param partial_words - boolean value
+     */
     public void setPartialWords(boolean partial_words) {
         LibVosk.vosk_recognizer_set_partial_words(this.getPointer(), partial_words);
     }
 
+    /**
+     * Adds speaker model to already initialized recognizer.
+     *
+     * Can add speaker recognition model to already created recognizer.
+     * Helps to initialize speaker recognition for grammar-based recognizer.
+     *
+     * @param spkModel Speaker recognition model
+     */
     public void setSpeakerModel(SpeakerModel spkModel) {
         LibVosk.vosk_recognizer_set_spk_model(this.getPointer(), spkModel.getPointer());
     }
 
+    /**
+     * Accept and process new chunk of voice data.
+     *
+     *  @param data - audio data in PCM 16-bit mono format
+     *  @param len - length of the audio data
+     *  @return 1 if silence is occurred and you can retrieve a new utterance with result method
+     *           0 if decoding continues
+     *           -1 if exception occurred
+     */
     public boolean acceptWaveForm(byte[] data, int len) {
         return LibVosk.vosk_recognizer_accept_waveform(this.getPointer(), data, len);
     }
@@ -43,22 +170,104 @@ public class Recognizer extends PointerType implements AutoCloseable {
         return LibVosk.vosk_recognizer_accept_waveform_f(this.getPointer(), data, len);
     }
 
+    /**
+     * Returns speech recognition result
+     *
+     * @return the result in JSON format which contains decoded line, decoded
+     *          words, times in seconds and confidences. You can parse this result
+     *          with any json parser
+     *
+     * <pre>
+     *  {
+     *    "text" : "what zero zero zero one"
+     *  }
+     * </pre>
+     *
+     * If alternatives enabled it returns result with alternatives, see also #setMaxAlternatives().
+     *
+     * If word times enabled returns word time, see also #setWordTimes().
+     */
     public String getResult() {
         return LibVosk.vosk_recognizer_result(this.getPointer());
     }
 
+    /**
+     * Returns partial speech recognition.
+     *
+     * @return partial speech recognition text which is not yet finalized.
+     *          result may change as recognizer process more data.
+     *
+     * <pre>
+     * {
+     *    "partial" : "cyril one eight zero"
+     * }
+     * </pre>
+     */
     public String getPartialResult() {
         return LibVosk.vosk_recognizer_partial_result(this.getPointer());
     }
 
+    /**
+     * Returns speech recognition result. Same as result, but doesn't wait for silence.
+     *  You usually call it in the end of the stream to get final bits of audio. It
+     *  flushes the feature pipeline, so all remaining audio chunks got processed.
+     *
+     *  @return speech result in JSON format.
+     */
     public String getFinalResult() {
         return LibVosk.vosk_recognizer_final_result(this.getPointer());
     }
 
+    /**
+     * Reconfigures recognizer to use grammar.
+     *
+     * @param grammar      Set of phrases in JSON array of strings or "[]" to use default model graph.
+     * @see #Recognizer(Model, float, String)
+     */
+    public void setGrammar(String grammar) {
+        LibVosk.vosk_recognizer_set_grm(this.getPointer(), grammar);
+    }
+
+    /**
+     * Resets the recognizer.
+     * Resets current results so the recognition can continue from scratch.
+     */
     public void reset() {
         LibVosk.vosk_recognizer_reset(this.getPointer());
     }
 
+    /**
+     * Endpointer delay mode
+     */
+    public class EndpointerMode {
+        public static final int DEFAULT = 0;
+        public static final int SHORT = 1;
+        public static final int LONG = 2;
+        public static final int VERY_LONG = 3;
+    }
+
+    /**
+     * Configures endpointer mode for recognizer
+     */
+    public void setEndpointerMode(int mode) {
+        LibVosk.vosk_recognizer_set_endpointer_mode(this.getPointer(), mode);
+    }
+
+    /**
+     * Set endpointer delays
+     *
+     * @param t_start_max     timeout for stopping recognition in case of initial silence (usually around 5.0)
+     * @param t_end           timeout for stopping recognition in milliseconds after we recognized something (usually around 0.5 - 1.0)
+     * @param t_max           timeout for forcing utterance end in milliseconds (usually around 20-30)
+     **/
+    public void setEndpointerDelays(float t_start_max, float t_end, float t_max) {
+        LibVosk.vosk_recognizer_set_endpointer_delays(this.getPointer(), t_start_max, t_end, t_max);
+    }
+
+    /**
+     * Releases recognizer object.
+     * Underlying model is also unreferenced and if needed, released.
+     */
     @Override
     public void close() {
         LibVosk.vosk_recognizer_free(this.getPointer());

android/lib/src/main/java/org/vosk/SpeakerModel.java

@@ -1,13 +1,36 @@
 package org.vosk;
 
 import com.sun.jna.PointerType;
+import java.io.IOException;
 
+/**
+ * Helps to initialize speaker recognition for grammar-based recognizer.
+ */
 public class SpeakerModel extends PointerType implements AutoCloseable {
     public SpeakerModel() {
     }
 
-    public SpeakerModel(String path) {
+    /**
+     * Loads speaker model data from the file.
+     *
+     * The path must contain:
+     * - a config file: mfcc.conf
+     * - kaldi nnet: final.ext.raw
+     * - mean.vec
+     * - transform.mat
+     *
+     * @param path the path of the model on the filesystem
+     * @throws IOException if the model could not be created
+     *
+     * @see <a href="http://kaldi-asr.org/doc/structkaldi_1_1MfccOptions.html">Kaldi MfccOptions</a>
+     * @see <a href="http://kaldi-asr.org/doc/classkaldi_1_1nnet3_1_1Nnet.html">Kaldi Nnet</a>
+     */
+    public SpeakerModel(String path) throws IOException {
         super(LibVosk.vosk_spk_model_new(path));
+
+        if (getPointer() == null) {
+            throw new IOException("Failed to create a speaker model");
+        }
     }
 
     @Override

android/lib/src/main/java/org/vosk/TextProcessor.java

@@ -0,0 +1,21 @@
+package org.vosk;
+
+import com.sun.jna.PointerType;
+
+public class TextProcessor extends PointerType implements AutoCloseable {
+    public TextProcessor() {
+    }
+
+    public TextProcessor(String verbalizer, String tagger) {
+        super(LibVosk.vosk_text_processor_new(verbalizer, tagger));
+    }
+
+    @Override
+    public void close() {
+        LibVosk.vosk_text_processor_free(this.getPointer());
+    }
+
+    public String itn(String input) {
+        return LibVosk.vosk_text_processor_itn(this.getPointer(), input);
+    }
+}

android/lib/src/main/java/org/vosk/android/SpeechService.java

@@ -19,9 +19,9 @@ import android.media.AudioRecord;
 import android.media.MediaRecorder.AudioSource;
 import android.os.Handler;
 import android.os.Looper;
+import android.annotation.SuppressLint;
 
 import org.vosk.Recognizer;
-
 import java.io.IOException;
 
 /**
@@ -48,6 +48,7 @@ public class SpeechService {
      *
      * @throws IOException thrown if audio recorder can not be created for some reason.
      */
+    @SuppressLint("MissingPermission")
     public SpeechService(Recognizer recognizer, float sampleRate) throws IOException {
         this.recognizer = recognizer;
         this.sampleRate = (int) sampleRate;
@@ -65,6 +66,51 @@ public class SpeechService {
         }
     }
 
+    /**
+     * Creates speech service with a caller-supplied {@link AudioRecord}.
+     * <p>
+     * Use this when you need to control the audio input device - for example,
+     * to pin recording to the built-in microphone when an external USB device
+     * without a microphone is present:
+     * <pre>
+     * AudioRecord recorder = new AudioRecord.Builder()
+     *         .setAudioSource(MediaRecorder.AudioSource.VOICE_RECOGNITION)
+     *         .setAudioFormat(format)
+     *         .build();
+     * if (Build.VERSION.SDK_INT >= 28) {
+     *     AudioManager am = (AudioManager) context.getSystemService(Context.AUDIO_SERVICE);
+     *     for (AudioDeviceInfo d : am.getDevices(AudioManager.GET_DEVICES_INPUTS)) {
+     *         if (d.getType() == AudioDeviceInfo.TYPE_BUILTIN_MIC) {
+     *             recorder.setPreferredDevice(d);
+     *             break;
+     *         }
+     *     }
+     * }
+     * SpeechService service = new SpeechService(recognizer, 16000f, recorder);
+     * </pre>
+     * <p>
+     * The caller retains ownership of {@code recorder}: if this constructor
+     * throws, the recorder is <em>not</em> released. Call
+     * {@link AudioRecord#release()} yourself in that case.
+     *
+     * @param recognizer the Vosk recognizer
+     * @param sampleRate sample rate in Hz; must match {@code recorder}'s configuration
+     * @param recorder   a fully-initialised {@link AudioRecord}
+     * @throws IOException if {@code recorder} is in STATE_UNINITIALIZED
+     */
+    public SpeechService(Recognizer recognizer, float sampleRate, AudioRecord recorder)
+            throws IOException {
+        this.recognizer = recognizer;
+        this.sampleRate = (int) sampleRate;
+        this.recorder   = recorder;
+
+        bufferSize = Math.round(this.sampleRate * BUFFER_SIZE_SECONDS);
+
+        if (recorder.getState() == AudioRecord.STATE_UNINITIALIZED) {
+            throw new IOException(
+                    "Failed to initialize recorder. Microphone might be already in use.");
+        }
+    }
 
     /**
      * Starts recognition. Does nothing if recognition is active.
@@ -136,6 +182,19 @@ public class SpeechService {
         return stopRecognizerThread();
     }
 
+
+    /**
+     * Returns the audio session ID of the underlying {@link AudioRecord}.
+     * <p>
+     * The session ID can be used to attach audio effects such as
+     * {@link android.media.audiofx.NoiseSuppressor} to the recording session.
+     *
+     * @return audio session ID, or {@link AudioRecord#ERROR} if unavailable
+     */
+    public int getAudioSessionId() {
+        return recorder.getAudioSessionId();
+    }
+
     /**
      * Shutdown the recognizer and release the recorder
      */

android/model-en/build.gradle

@@ -3,11 +3,12 @@ def pomName = "Vosk English Model"
 def pomDescription = "Small English model for Android"
 
 android {
-    compileSdkVersion 29
+    namespace "org.vosk"
+    compileSdkVersion 36
     defaultConfig {
         minSdkVersion 21
-        targetSdkVersion 29
-        versionCode 6
+        targetSdkVersion 36
+        versionCode 10
         versionName = version
         archivesBaseName = archiveName
     }
@@ -33,15 +34,10 @@ tasks.register('genUUID') {
 
 preBuild.dependsOn(genUUID)
 
-publishing {
-    publications {
-        aar(MavenPublication) {
-            artifactId = archiveName
-            artifact("$buildDir/outputs/aar/$archiveName-release.aar")
-            pom {
-                name = pomName
-                description = pomDescription
-            }
-        }
+mavenPublishing {
+    coordinates("com.alphacephei", archiveName, version)
+    pom {
+        name = pomName
+        description = pomDescription
     }
 }

android/model-en/src/main/AndroidManifest.xml

@@ -1,3 +1,2 @@
-<manifest xmlns:android="http://schemas.android.com/apk/res/android"
-    package="org.vosk.model.en">
+<manifest>
 </manifest>

android/model-en/src/main/assets/model-en-us/README

@@ -1,7 +0,0 @@
-US English model for mobile Vosk applications
-
-Copyright 2020 Alpha Cephei Inc
-
-Accuracy: 10.38 (tedlium test) 9.85 (librispeech test-clean)
-Speed: 0.11xRT (desktop)
-Latency: 0.15s (right context)

csharp/demo/VoskDemo.cs

@@ -28,6 +28,9 @@ public class VoskDemo
    {
         // Demo float array
         VoskRecognizer rec = new VoskRecognizer(model, 16000.0f);
+
+        rec.SetEndpointerMode(EndpointerMode.LONG);
+
         using(Stream source = File.OpenRead("test.wav")) {
             byte[] buffer = new byte[4096];
             int bytesRead;

csharp/demo/VoskDemo.csproj

@@ -2,7 +2,7 @@
 
   <PropertyGroup>
     <OutputType>Exe</OutputType>
-    <TargetFramework>net5.0</TargetFramework>
+    <TargetFramework>net8.0</TargetFramework>
     <RootNamespace>VoskDemo</RootNamespace>
   </PropertyGroup>
 
@@ -11,7 +11,7 @@
   </PropertyGroup>
 
   <ItemGroup>
-    <PackageReference Include="Vosk" Version="0.3.45" />
+    <PackageReference Include="Vosk" Version="0.3.75" />
   </ItemGroup>
 
 </Project>

csharp/nuget/Vosk.csproj

@@ -0,0 +1,17 @@
+<Project Sdk="Microsoft.NET.Sdk">
+
+  <PropertyGroup>
+    <TargetFramework>net8.0</TargetFramework>
+    <ImplicitUsings>enable</ImplicitUsings>
+    <Nullable>enable</Nullable>
+    <PackageId>Vosk</PackageId>
+    <Version>0.3.75</Version>
+    <authors>Alpha Cephei Inc</authors>
+    <owners>Alpha Cephei Inc</owners>
+  </PropertyGroup>
+
+  <Target Name="CopyFiles" AfterTargets="Build">
+    <Copy SourceFiles="bin/Release/net8.0/Vosk.dll" DestinationFolder="lib/net8.0" />
+  </Target>
+
+</Project>

csharp/nuget/Vosk.nuspec

@@ -2,7 +2,7 @@
 <package>
   <metadata>
     <id>Vosk</id>
-    <version>0.3.45</version>
+    <version>0.3.75</version>
     <authors>Alpha Cephei Inc</authors>
     <owners>Alpha Cephei Inc</owners>
     <license type="expression">Apache-2.0</license>
@@ -23,10 +23,10 @@ Vosk scales from small devices like Raspberry Pi or Android smartphone to big cl
     <copyright>Copyright 2020-2050 Alpha Cephei Inc</copyright>
     <tags>speech recognition voice stt asr speech-to-text ai offline privacy</tags>
     <dependencies>
-      <group targetFramework=".NETStandard2.0"/>
+      <group targetFramework="net8.0"/>
     </dependencies>
   </metadata>
   <files>
-    <file src="**" exclude="src/*.cs;build.sh;**/.keep-me;*.nupkg" />
+    <file src="**" exclude="bin/**;obj/**;build.sh;src/*.cs;*.nupkg;**/.keep-me" />
   </files>
 </package>

csharp/nuget/build.sh

@@ -1,2 +1,2 @@
-mcs -out:lib/netstandard2.0/Vosk.dll -target:library src/*.cs
-nuget pack
+rm -rf bin lib obj
+/home/shmyrev/local/dotnet/dotnet pack Vosk.csproj -p:NuspecFile=Vosk.nuspec -o .

csharp/nuget/src/BatchModel.cs

@@ -0,0 +1,55 @@
+using System;
+
+namespace Vosk
+{
+    public class BatchModel : global::System.IDisposable
+    {
+        private global::System.Runtime.InteropServices.HandleRef handle;
+
+        internal BatchModel(global::System.IntPtr cPtr)
+        {
+            handle = new global::System.Runtime.InteropServices.HandleRef(this, cPtr);
+        }
+
+        internal static global::System.Runtime.InteropServices.HandleRef getCPtr(BatchModel obj)
+        {
+            return (obj == null) ? new global::System.Runtime.InteropServices.HandleRef(null, global::System.IntPtr.Zero) : obj.handle;
+        }
+
+        ~BatchModel()
+        {
+            Dispose(false);
+        }
+
+        public void Dispose()
+        {
+            Dispose(true);
+            global::System.GC.SuppressFinalize(this);
+        }
+
+        protected virtual void Dispose(bool disposing)
+        {
+            lock (this)
+            {
+                if (handle.Handle != global::System.IntPtr.Zero)
+                {
+                    VoskPINVOKE.delete_BatchModel(handle);
+                    handle = new global::System.Runtime.InteropServices.HandleRef(null, global::System.IntPtr.Zero);
+                }
+            }
+        }
+
+        public BatchModel(string model_path) : this(VoskPINVOKE.new_BatchModel(model_path))
+        {
+        }
+
+        public void WaitForCompletion()
+        {
+            if (handle.Handle != global::System.IntPtr.Zero)
+            {
+                VoskPINVOKE.wait_BatchModel(handle);
+            }
+        }
+    }
+
+}

csharp/nuget/src/VoskBatchRecognizer.cs

@@ -0,0 +1,88 @@
+using System;
+
+namespace Vosk
+{
+    public class VoskBatchRecognizer : System.IDisposable
+    {
+        private System.Runtime.InteropServices.HandleRef handle;
+
+        internal VoskBatchRecognizer(System.IntPtr cPtr)
+        {
+            handle = new System.Runtime.InteropServices.HandleRef(this, cPtr);
+        }
+
+        internal static System.Runtime.InteropServices.HandleRef getCPtr(VoskBatchRecognizer obj)
+        {
+            return (obj == null) ? new System.Runtime.InteropServices.HandleRef(null, System.IntPtr.Zero) : obj.handle;
+        }
+
+        ~VoskBatchRecognizer()
+        {
+            Dispose(false);
+        }
+
+        public void Dispose()
+        {
+            Dispose(true);
+            System.GC.SuppressFinalize(this);
+        }
+
+        protected virtual void Dispose(bool disposing)
+        {
+            lock (this)
+            {
+                if (handle.Handle != System.IntPtr.Zero)
+                {
+                    VoskPINVOKE.delete_VoskBatchRecognizer(handle);
+                    handle = new System.Runtime.InteropServices.HandleRef(null, System.IntPtr.Zero);
+                }
+            }
+        }
+
+        public VoskBatchRecognizer(BatchModel model, float sample_rate) : this(VoskPINVOKE.new_VoskBatchRecognizer(BatchModel.getCPtr(model), sample_rate))
+        {
+        }
+
+        public bool AcceptWaveform(byte[] data, int len)
+        {
+            return VoskPINVOKE.VoskBatchRecognizer_AcceptWaveform(handle, data, len);
+        }
+
+        private static string PtrToStringUTF8(System.IntPtr ptr)
+        {
+            int len = 0;
+            while (System.Runtime.InteropServices.Marshal.ReadByte(ptr, len) != 0)
+                len++;
+            byte[] array = new byte[len];
+            System.Runtime.InteropServices.Marshal.Copy(ptr, array, 0, len);
+            return System.Text.Encoding.UTF8.GetString(array);
+        }
+
+        public string FrontResult()
+        {
+            return PtrToStringUTF8(VoskPINVOKE.VoskBatchRecognizer_FrontResult(handle));
+        }
+
+        public string Result()
+        {
+            string result = PtrToStringUTF8(VoskPINVOKE.VoskBatchRecognizer_FrontResult(handle));
+            VoskPINVOKE.VoskBatchRecognizer_Pop(handle);
+            return result;
+        }
+
+        public int GetNumPendingChunks()
+        {
+            return VoskPINVOKE.VoskBatchRecognizer_GetPendingChunks(handle);
+        }
+
+        public void FinishStream()
+        {
+            VoskPINVOKE.VoskBatchRecognizer_FinishStream(handle);
+        }
+
+        public void SetNLSML(bool nlsml)
+        {
+            VoskPINVOKE.VoskBatchRecognizer_SetNLSML(handle, Convert.ToInt32(nlsml));
+        }
+    }
+}

csharp/nuget/src/VoskPINVOKE.cs

@@ -65,6 +65,12 @@ class VoskPINVOKE {
   [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint="vosk_recognizer_reset")]
   public static extern void VoskRecognizer_Reset(global::System.Runtime.InteropServices.HandleRef jarg1);
 
+  [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint="vosk_recognizer_set_endpointer_mode")]
+  public static extern void VoskRecognizer_SetEndpointerMode(global::System.Runtime.InteropServices.HandleRef jarg1, int jarg2);
+
+  [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint="vosk_recognizer_set_endpointer_delays")]
+  public static extern void VoskRecognizer_SetEndpointerDelays(global::System.Runtime.InteropServices.HandleRef jarg1, float jarg2, float jarg3, float jarg4);
+
   [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint="vosk_set_log_level")]
   public static extern void SetLogLevel(int jarg1);
 
@@ -73,6 +79,40 @@ class VoskPINVOKE {
 
   [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint="vosk_gpu_thread_init")]
   public static extern void GpuThreadInit();
+
+  [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint = "vosk_batch_model_new")]
+  public static extern global::System.IntPtr new_BatchModel(string jarg1);
+
+  [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint = "vosk_batch_model_free")]
+  public static extern void delete_BatchModel(global::System.Runtime.InteropServices.HandleRef jarg1);
+
+  [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint = "vosk_batch_model_wait")]
+  public static extern void wait_BatchModel(global::System.Runtime.InteropServices.HandleRef jarg1);
+
+  [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint = "vosk_batch_recognizer_new")]
+  public static extern global::System.IntPtr new_VoskBatchRecognizer(global::System.Runtime.InteropServices.HandleRef jarg1, float jarg2);
+
+  [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint = "vosk_batch_recognizer_free")]
+  public static extern void delete_VoskBatchRecognizer(global::System.Runtime.InteropServices.HandleRef jarg1);
+
+  [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint = "vosk_batch_recognizer_accept_waveform")]
+  public static extern bool VoskBatchRecognizer_AcceptWaveform(global::System.Runtime.InteropServices.HandleRef jarg1, [global::System.Runtime.InteropServices.In, global::System.Runtime.InteropServices.MarshalAs(global::System.Runtime.InteropServices.UnmanagedType.LPArray)] byte[] jarg2, int jarg3);
+
+  [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint = "vosk_batch_recognizer_set_nlsml")]
+  public static extern void VoskBatchRecognizer_SetNLSML(global::System.Runtime.InteropServices.HandleRef jarg1, int jarg2);
+
+  [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint = "vosk_batch_recognizer_finish_stream")]
+  public static extern void VoskBatchRecognizer_FinishStream(global::System.Runtime.InteropServices.HandleRef jarg1);
+
+  [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint = "vosk_batch_recognizer_front_result")]
+  public static extern global::System.IntPtr VoskBatchRecognizer_FrontResult(global::System.Runtime.InteropServices.HandleRef jarg1);
+
+  [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint = "vosk_batch_recognizer_pop")]
+  public static extern void VoskBatchRecognizer_Pop(global::System.Runtime.InteropServices.HandleRef jarg1);
+
+  [global::System.Runtime.InteropServices.DllImport("libvosk", EntryPoint = "vosk_batch_recognizer_get_pending_chunks")]
+  public static extern int VoskBatchRecognizer_GetPendingChunks(global::System.Runtime.InteropServices.HandleRef jarg1);
+
 }
 
 }

csharp/nuget/src/VoskRecognizer.cs

@@ -1,5 +1,12 @@
 namespace Vosk {
 
+public enum EndpointerMode {
+    DEFAULT = 0,
+    SHORT = 1,
+    LONG = 2,
+    VERY_LONG = 3
+}
+
 public class VoskRecognizer : System.IDisposable {
   private System.Runtime.InteropServices.HandleRef handle;
 
@@ -91,6 +98,14 @@ public class VoskRecognizer : System.IDisposable {
     VoskPINVOKE.VoskRecognizer_Reset(handle);
   }
 
+  public void SetEndpointerMode(EndpointerMode mode) {
+    VoskPINVOKE.VoskRecognizer_SetEndpointerMode(handle, (int) mode);
+  }
+
+  public void SetEndpointerDelays(float t_start_max, float t_end, float t_max) {
+    VoskPINVOKE.VoskRecognizer_SetEndpointerDelays(handle, t_start_max, t_end, t_max);
+  }
+
 }
 
 }

go/batch.go

@@ -0,0 +1,99 @@
+package vosk
+
+// #cgo CPPFLAGS: -I ${SRCDIR}/../src
+// #cgo !windows LDFLAGS: -L ${SRCDIR}/../src -lvosk -ldl -lpthread
+// #cgo windows LDFLAGS: -L ${SRCDIR}/../src -lvosk -lpthread
+// #include <stdlib.h>
+// #include <vosk_api.h>
+import "C"
+import "unsafe"
+
+// VoskBatchModel contains a reference to the C VoskBatchModel
+type VoskBatchModel struct {
+	model *C.struct_VoskBatchModel
+}
+
+// NewBatchModel creates a new VoskBatchModel instance
+func NewBatchModel(modelPath string) (*VoskBatchModel, error) {
+	cmodelPath := C.CString(modelPath)
+	defer C.free(unsafe.Pointer(cmodelPath))
+	internal := C.vosk_batch_model_new(cmodelPath)
+	model := &VoskBatchModel{model: internal}
+	return model, nil
+}
+
+func (m *VoskBatchModel) Free() {
+	C.vosk_batch_model_free(m.model)
+}
+
+func (m *VoskBatchModel) Wait() {
+	C.vosk_batch_model_wait(m.model);
+}
+
+func freeBatchModel(model *VoskBatchModel) {
+	C.vosk_batch_model_free(model.model)
+}
+
+// VoskBatchRecognizer contains a reference to the C VoskBatchRecognizer
+type VoskBatchRecognizer struct {
+	rec *C.struct_VoskBatchRecognizer
+}
+
+func freeBatchRecognizer(recognizer *VoskBatchRecognizer) {
+	C.vosk_batch_recognizer_free(recognizer.rec)
+}
+
+func (r *VoskBatchRecognizer) Free() {
+	C.vosk_batch_recognizer_free(r.rec)
+}
+
+// NewBatchRecognizer creates a new VoskBatchRecognizer instance
+func NewBatchRecognizer(model *VoskBatchModel, sampleRate float64) (*VoskBatchRecognizer, error) {
+	internal := C.vosk_batch_recognizer_new(model.model, C.float(sampleRate))
+	rec := &VoskBatchRecognizer{rec: internal}
+	return rec, nil
+}
+
+// AcceptWaveform accepts and processes a new chunk of the voice data.
+func (r *VoskBatchRecognizer) AcceptWaveform(buffer []byte) {
+	cbuf := C.CBytes(buffer)
+	defer C.free(cbuf)
+	C.vosk_batch_recognizer_accept_waveform(r.rec, (*C.char)(cbuf), C.int(len(buffer)))
+}
+
+/** Set NLSML output
+ * @param nlsml - boolean value
+ */
+//void vosk_batch_recognizer_set_nlsml(VoskBatchRecognizer *recognizer, int nlsml);
+
+func (r *VoskBatchRecognizer) SetNlsml(nlsml int) {
+	C.vosk_batch_recognizer_set_nlsml(r.rec, C.int(nlsml))
+}
+
+/** Closes the stream */
+//void vosk_batch_recognizer_finish_stream(VoskBatchRecognizer *recognizer);
+
+func (r *VoskBatchRecognizer) FinishStream() {
+	C.vosk_batch_recognizer_finish_stream(r.rec)
+}
+
+/** Return results */
+//const char *vosk_batch_recognizer_front_result(VoskBatchRecognizer *recognizer);
+
+func (r *VoskBatchRecognizer) FrontResult() string {
+	return C.GoString(C.vosk_batch_recognizer_front_result(r.rec))
+}
+
+/** Release and free first retrieved result */
+//void vosk_batch_recognizer_pop(VoskBatchRecognizer *recognizer);
+
+func (r *VoskBatchRecognizer) Pop() {
+	C.vosk_batch_recognizer_pop(r.rec)
+}
+
+/** Get amount of pending chunks for more intelligent waiting */
+//int vosk_batch_recognizer_get_pending_chunks(VoskBatchRecognizer *recognizer);
+func (r *VoskBatchRecognizer) GetPendingChunks() int {
+	i := C.vosk_batch_recognizer_get_pending_chunks(r.rec)
+	return int(i)
+}

go/batch_example/README.md

@@ -0,0 +1,5 @@
+This example expects a `s16le` converted audio file and converts it to text in a
+manner that imitates the Python example of [test_gpu_batch.py](../python/example/test_gpu_batch.py).
+
+Note that the `libvosk.so` must be in the library path. This was successfully tested on
+Ubuntu 24.04 with Go 1.18, gcc-11, NVIDIA driver 570.172.08.

go/batch_example/test_batch.go

@@ -0,0 +1,54 @@
+package main
+
+import (
+	"flag"
+	"fmt"
+	"io"
+	"log"
+	"os"
+
+	vosk "github.com/alphacep/vosk-api/go"
+)
+
+func main() {
+	var filename string
+	flag.StringVar(&filename, "f", "", "file to transcribe")
+	flag.Parse()
+
+	vosk.GPUInit()
+
+	model, err := vosk.NewBatchModel("model")
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	rec, err := vosk.NewBatchRecognizer(model, 16000.0)
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	file, err := os.Open(filename)
+	if err != nil {
+		panic(err)
+	}
+	defer file.Close()
+
+	buf := make([]byte, 8000)
+
+	for {
+		if _, err := file.Read(buf); err != nil {
+			if err != io.EOF {
+				log.Fatal(err)
+			}
+
+			break
+		}
+		rec.AcceptWaveform(buf)
+		model.Wait()
+		if rec.FrontResult() != "" {
+			fmt.Println(rec.FrontResult())
+			rec.Pop()
+		}
+	}
+	// Is this needed? rec.FinishStream()
+}

go/example/test_simple.go

@@ -1,13 +1,12 @@
 package main
 
 import (
-	"bufio"
+	"encoding/json"
 	"flag"
 	"fmt"
 	"io"
 	"log"
 	"os"
-	"encoding/json"
 
 	vosk "github.com/alphacep/vosk-api/go"
 )
@@ -17,6 +16,8 @@ func main() {
 	flag.StringVar(&filename, "f", "", "file to transcribe")
 	flag.Parse()
 
+	vosk.GPUInit()
+
 	model, err := vosk.NewModel("model")
 	if err != nil {
 		log.Fatal(err)
@@ -38,11 +39,10 @@ func main() {
 	}
 	defer file.Close()
 
-	reader := bufio.NewReader(file)
 	buf := make([]byte, 4096)
 
 	for {
-		_, err := reader.Read(buf)
+		_, err := file.Read(buf)
 		if err != nil {
 			if err != io.EOF {
 				log.Fatal(err)

go/vosk.go

@@ -1,7 +1,8 @@
 package vosk
 
 // #cgo CPPFLAGS: -I ${SRCDIR}/../src
-// #cgo LDFLAGS: -L ${SRCDIR}/../src -lvosk -ldl -lpthread
+// #cgo !windows LDFLAGS: -L ${SRCDIR}/../src -lvosk -ldl -lpthread
+// #cgo windows LDFLAGS: -L ${SRCDIR}/../src -lvosk -lpthread
 // #include <stdlib.h>
 // #include <vosk_api.h>
 import "C"
@@ -102,6 +103,13 @@ func (r *VoskRecognizer) SetSpkModel(spkModel *VoskSpkModel) {
 	C.vosk_recognizer_set_spk_model(r.rec, spkModel.spkModel)
 }
 
+// SetGrm sets which phrases to recognize on an already initialized recognizer.
+func (r *VoskRecognizer) SetGrm(grammar string) {
+	cgrammar := C.CString(grammar)
+	defer C.free(unsafe.Pointer(cgrammar))
+	C.vosk_recognizer_set_grm(r.rec, cgrammar)
+}
+
 // SetMaxAlternatives configures the recognizer to output n-best results.
 func (r *VoskRecognizer) SetMaxAlternatives(maxAlternatives int) {
 	C.vosk_recognizer_set_max_alternatives(r.rec, C.int(maxAlternatives))
@@ -117,6 +125,16 @@ func (r *VoskRecognizer) SetPartialWords(words int) {
 	C.vosk_recognizer_set_partial_words(r.rec, C.int(words))
 }
 
+// SetEndpointerDelays sets the recognition timeouts, where startMax
+// is the timeout for stopping recognition in case of initial silence
+// (usually around 5), end is the timeout for stopping recognition
+// in milliseconds after we recognized something (usually around 0.5-1.0),
+// and max is the timeout for forcing utterance end in milliseconds
+// (usually around 20-30).
+func (r *VoskRecognizer) SetEndpointerDelays(startMax, end, max float64) {
+	C.vosk_recognizer_set_endpointer_delays(r.rec, C.float(startMax), C.float(end), C.float(max))
+}
+
 // AcceptWaveform accepts and processes a new chunk of the voice data.
 func (r *VoskRecognizer) AcceptWaveform(buffer []byte) int {
 	cbuf := C.CBytes(buffer)

java/demo/build.gradle

@@ -11,6 +11,5 @@ repositories {
 }
 
 dependencies {
-    implementation group: 'net.java.dev.jna', name: 'jna', version: '5.7.0'
-    implementation group: 'com.alphacephei', name: 'vosk', version: '0.3.40+'
+    implementation group: 'com.alphacephei', name: 'vosk', version: '0.3.75'
 }

java/lib/build.gradle

@@ -16,7 +16,7 @@ repositories {
 
 archivesBaseName = 'vosk'
 group = 'com.alphacephei'
-version = '0.3.45'
+version = '0.3.75'
 
 mavenPublish {
     group = 'com.alphacephei'
@@ -25,7 +25,7 @@ mavenPublish {
 }
 
 dependencies {
-    api group: 'net.java.dev.jna', name: 'jna', version: '5.7.0'
+    api group: 'net.java.dev.jna', name: 'jna', version: '5.18.1'
     testImplementation 'junit:junit:4.13'
 }
 

java/lib/src/main/java/org/vosk/LibVosk.java

@@ -1,7 +1,6 @@
 package org.vosk;
 
 import com.sun.jna.Native;
-import com.sun.jna.Library;
 import com.sun.jna.Platform;
 import com.sun.jna.Pointer;
 import java.io.File;
@@ -13,8 +12,9 @@ import java.nio.file.StandardCopyOption;
 public class LibVosk {
 
     private static void unpackDll(File targetDir, String lib) throws IOException {
-        InputStream source = LibVosk.class.getResourceAsStream("/win32-x86-64/" + lib + ".dll");
-        Files.copy(source, new File(targetDir, lib + ".dll").toPath(), StandardCopyOption.REPLACE_EXISTING);
+        try (InputStream source = LibVosk.class.getResourceAsStream("/win32-x86-64/" + lib + ".dll")) {
+            Files.copy(source, new File(targetDir, lib + ".dll").toPath(), StandardCopyOption.REPLACE_EXISTING);
+        }
     }
 
     static {
@@ -23,7 +23,7 @@ public class LibVosk {
             // We have to unpack dependencies
             try {
                 // To get a tmp folder we unpack small library and mark it for deletion
-                File tmpFile = Native.extractFromResourcePath("/win32-x86-64/empty");
+                File tmpFile = Native.extractFromResourcePath("/win32-x86-64/empty", LibVosk.class.getClassLoader());
                 File tmpDir = tmpFile.getParentFile();
                 new File(tmpDir, tmpFile.getName() + ".x").createNewFile();
 
@@ -78,10 +78,30 @@ public class LibVosk {
 
     public static native String vosk_recognizer_partial_result(Pointer recognizer);
 
+    public static native void vosk_recognizer_set_grm(Pointer recognizer, String grammar);
+
     public static native void vosk_recognizer_reset(Pointer recognizer);
 
+    public static native void vosk_recognizer_set_endpointer_mode(Pointer recognizer, int mode);
+
+    public static native void vosk_recognizer_set_endpointer_delays(Pointer recognizer, float t_start_max, float t_end, float t_max);
+
     public static native void vosk_recognizer_free(Pointer recognizer);
 
+    public static native Pointer vosk_text_processor_new(String verbalizer, String tagger);
+
+    public static native void vosk_text_processor_free(Pointer processor);
+
+    public static native String vosk_text_processor_itn(Pointer processor, String input);
+
+    /**
+     * Set log level for Kaldi messages.
+     *
+     *  @param loglevel the level
+     *     0 - default value to print info and error messages but no debug
+     *     less than 0 - don't print info messages
+     *     greater than 0 - more verbose mode
+     */
     public static void setLogLevel(LogLevel loglevel) {
         vosk_set_log_level(loglevel.getValue());
     }

java/lib/src/main/java/org/vosk/Recognizer.java

@@ -4,6 +4,17 @@ import com.sun.jna.PointerType;
 import java.io.IOException;
 
 public class Recognizer extends PointerType implements AutoCloseable {
+    /**
+     * Creates the recognizer object.
+     *
+     * The recognizers process the speech and return text using shared model data
+     * @param model       VoskModel containing static data for recognizer. Model can be
+     *                    shared across recognizers, even running in different threads.
+     * @param sampleRate The sample rate of the audio you are going to feed into the recognizer.
+     *                    Make sure this rate matches the audio content, it is a common
+     *                    issue causing accuracy problems.
+     * @throws IOException if the recognizer could not be created
+     */
     public Recognizer(Model model, float sampleRate) throws IOException {
         super(LibVosk.vosk_recognizer_new(model, sampleRate));
 
@@ -12,30 +23,141 @@ public class Recognizer extends PointerType implements AutoCloseable {
         }
     }
 
-    public Recognizer(Model model, float sampleRate, SpeakerModel spkModel) {
+    /**
+     * Creates the recognizer object with speaker recognition.
+     *
+     *  With the speaker recognition mode the recognizer not just recognize
+     *  text but also return speaker vectors one can use for speaker identification
+     *
+     *  @param model       VoskModel containing static data for recognizer. Model can be
+     *                     shared across recognizers, even running in different threads.
+     *  @param sampleRate The sample rate of the audio you are going to feed into the recognizer.
+     *                     Make sure this rate matches the audio content, it is a common
+     *                     issue causing accuracy problems.
+     *  @param spkModel speaker model for speaker identification
+     *  @throws IOException if the recognizer could not be created
+     */
+    public Recognizer(Model model, float sampleRate, SpeakerModel spkModel) throws IOException {
         super(LibVosk.vosk_recognizer_new_spk(model.getPointer(), sampleRate, spkModel.getPointer()));
+
+        if (getPointer() == null) {
+            throw new IOException("Failed to create a recognizer");
+        }
     }
 
-    public Recognizer(Model model, float sampleRate, String grammar) {
+    /**
+     * Creates the recognizer object with the phrase list.
+     *
+     *  Sometimes when you want to improve recognition accuracy and when you don't need
+     *  to recognize large vocabulary you can specify a list of phrases to recognize. This
+     *  will improve recognizer speed and accuracy but might return [unk] if user said
+     *  something different.
+     *
+     *  Only recognizers with lookahead models support this type of quick configuration.
+     *  Precompiled HCLG graph models are not supported.
+     *
+     *  @param model       VoskModel containing static data for recognizer. Model can be
+     *                     shared across recognizers, even running in different threads.
+     *  @param sampleRate The sample rate of the audio you are going to feed into the recognizer.
+     *                     Make sure this rate matches the audio content, it is a common
+     *                     issue causing accuracy problems.
+     *  @param grammar The string with the list of phrases to recognize as JSON array of strings,
+     *                 for example "["one two three four five", "[unk]"]".
+     *  @throws IOException if the recognizer could not be created
+     */
+    public Recognizer(Model model, float sampleRate, String grammar) throws IOException {
         super(LibVosk.vosk_recognizer_new_grm(model.getPointer(), sampleRate, grammar));
+
+        if (getPointer() == null) {
+            throw new IOException("Failed to create a recognizer");
+        }
     }
 
+    /**
+     * Configures recognizer to output n-best results.
+     *
+     * <pre>
+     *   {
+     *      "alternatives": [
+     *          { "text": "one two three four five", "confidence": 0.97 },
+     *          { "text": "one two three for five", "confidence": 0.03 },
+     *      ]
+     *   }
+     * </pre>
+     *
+     * @param maxAlternatives - maximum alternatives to return from recognition results
+     */
     public void setMaxAlternatives(int maxAlternatives) {
         LibVosk.vosk_recognizer_set_max_alternatives(this.getPointer(), maxAlternatives);
     }
 
+    /** Enables words with times in the output
+     *
+     * <pre>
+     *   "result" : [{
+     *       "conf" : 1.000000,
+     *       "end" : 1.110000,
+     *       "start" : 0.870000,
+     *       "word" : "what"
+     *     }, {
+     *       "conf" : 1.000000,
+     *       "end" : 1.530000,
+     *       "start" : 1.110000,
+     *       "word" : "zero"
+     *     }, {
+     *       "conf" : 1.000000,
+     *       "end" : 1.950000,
+     *       "start" : 1.530000,
+     *       "word" : "zero"
+     *     }, {
+     *       "conf" : 1.000000,
+     *       "end" : 2.340000,
+     *       "start" : 1.950000,
+     *       "word" : "zero"
+     *     }, {
+     *       "conf" : 1.000000,
+     *       "end" : 2.610000,
+     *       "start" : 2.340000,
+     *       "word" : "one"
+     *     }],
+     * </pre>
+     *
+     * @param words - boolean value
+     */
     public void setWords(boolean words) {
         LibVosk.vosk_recognizer_set_words(this.getPointer(), words);
     }
 
+    /**
+     * Like above return words and confidences in partial results.
+     *
+     * @param partial_words - boolean value
+     */
     public void setPartialWords(boolean partial_words) {
         LibVosk.vosk_recognizer_set_partial_words(this.getPointer(), partial_words);
     }
 
+    /**
+     * Adds speaker model to already initialized recognizer.
+     *
+     * Can add speaker recognition model to already created recognizer.
+     * Helps to initialize speaker recognition for grammar-based recognizer.
+     *
+     * @param spkModel Speaker recognition model
+     */
     public void setSpeakerModel(SpeakerModel spkModel) {
         LibVosk.vosk_recognizer_set_spk_model(this.getPointer(), spkModel.getPointer());
     }
 
+    /**
+     * Accept and process new chunk of voice data.
+     *
+     *  @param data - audio data in PCM 16-bit mono format
+     *  @param len - length of the audio data
+     *  @return 1 if silence is occurred and you can retrieve a new utterance with result method
+     *           0 if decoding continues
+     *           -1 if exception occurred
+     */
     public boolean acceptWaveForm(byte[] data, int len) {
         return LibVosk.vosk_recognizer_accept_waveform(this.getPointer(), data, len);
     }
@@ -48,22 +170,104 @@ public class Recognizer extends PointerType implements AutoCloseable {
         return LibVosk.vosk_recognizer_accept_waveform_f(this.getPointer(), data, len);
     }
 
+    /**
+     * Returns speech recognition result
+     *
+     * @return the result in JSON format which contains decoded line, decoded
+     *          words, times in seconds and confidences. You can parse this result
+     *          with any json parser
+     *
+     * <pre>
+     *  {
+     *    "text" : "what zero zero zero one"
+     *  }
+     * </pre>
+     *
+     * If alternatives enabled it returns result with alternatives, see also #setMaxAlternatives().
+     *
+     * If word times enabled returns word time, see also #setWordTimes().
+     */
     public String getResult() {
         return LibVosk.vosk_recognizer_result(this.getPointer());
     }
 
+    /**
+     * Returns partial speech recognition.
+     *
+     * @return partial speech recognition text which is not yet finalized.
+     *          result may change as recognizer process more data.
+     *
+     * <pre>
+     * {
+     *    "partial" : "cyril one eight zero"
+     * }
+     * </pre>
+     */
     public String getPartialResult() {
         return LibVosk.vosk_recognizer_partial_result(this.getPointer());
     }
 
+    /**
+     * Returns speech recognition result. Same as result, but doesn't wait for silence.
+     *  You usually call it in the end of the stream to get final bits of audio. It
+     *  flushes the feature pipeline, so all remaining audio chunks got processed.
+     *
+     *  @return speech result in JSON format.
+     */
     public String getFinalResult() {
         return LibVosk.vosk_recognizer_final_result(this.getPointer());
     }
 
+    /**
+     * Reconfigures recognizer to use grammar.
+     *
+     * @param grammar      Set of phrases in JSON array of strings or "[]" to use default model graph.
+     * @see #Recognizer(Model, float, String)
+     */
+    public void setGrammar(String grammar) {
+        LibVosk.vosk_recognizer_set_grm(this.getPointer(), grammar);
+    }
+
+    /**
+     * Resets the recognizer.
+     * Resets current results so the recognition can continue from scratch.
+     */
     public void reset() {
         LibVosk.vosk_recognizer_reset(this.getPointer());
     }
 
+    /**
+     * Endpointer delay mode
+     */
+    public class EndpointerMode {
+        public static final int DEFAULT = 0;
+        public static final int SHORT = 1;
+        public static final int LONG = 2;
+        public static final int VERY_LONG = 3;
+    }
+
+    /**
+     * Configures endpointer mode for recognizer
+     */
+    public void setEndpointerMode(int mode) {
+        LibVosk.vosk_recognizer_set_endpointer_mode(this.getPointer(), mode);
+    }
+
+    /**
+     * Set endpointer delays
+     *
+     * @param t_start_max     timeout for stopping recognition in case of initial silence (usually around 5.0)
+     * @param t_end           timeout for stopping recognition in milliseconds after we recognized something (usually around 0.5 - 1.0)
+     * @param t_max           timeout for forcing utterance end in milliseconds (usually around 20-30)
+     **/
+    public void setEndpointerDelays(float t_start_max, float t_end, float t_max) {
+        LibVosk.vosk_recognizer_set_endpointer_delays(this.getPointer(), t_start_max, t_end, t_max);
+    }
+
+    /**
+     * Releases recognizer object.
+     * Underlying model is also unreferenced and if needed, released.
+     */
     @Override
     public void close() {
         LibVosk.vosk_recognizer_free(this.getPointer());

java/lib/src/main/java/org/vosk/SpeakerModel.java

@@ -3,10 +3,28 @@ package org.vosk;
 import com.sun.jna.PointerType;
 import java.io.IOException;
 
+/**
+ * Helps to initialize speaker recognition for grammar-based recognizer.
+ */
 public class SpeakerModel extends PointerType implements AutoCloseable {
     public SpeakerModel() {
     }
 
+    /**
+     * Loads speaker model data from the file.
+     *
+     * The path must contain:
+     * - a config file: mfcc.conf
+     * - kaldi nnet: final.ext.raw
+     * - mean.vec
+     * - transform.mat
+     *
+     * @param path the path of the model on the filesystem
+     * @throws IOException if the model could not be created
+     *
+     * @see <a href="http://kaldi-asr.org/doc/structkaldi_1_1MfccOptions.html">Kaldi MfccOptions</a>
+     * @see <a href="http://kaldi-asr.org/doc/classkaldi_1_1nnet3_1_1Nnet.html">Kaldi Nnet</a>
+     */
     public SpeakerModel(String path) throws IOException {
         super(LibVosk.vosk_spk_model_new(path));
 

java/lib/src/test/java/org/vosk/test/DecoderTest.java

@@ -15,8 +15,10 @@ import javax.sound.sampled.UnsupportedAudioFileException;
 
 import org.vosk.LogLevel;
 import org.vosk.Recognizer;
+import org.vosk.Recognizer.EndpointerMode;
 import org.vosk.LibVosk;
 import org.vosk.Model;
+import org.vosk.TextProcessor;
 
 public class DecoderTest {
 
@@ -95,9 +97,24 @@ public class DecoderTest {
         Assert.assertTrue(true);
     }
 
+    @Test
+    public void decoderEndpointerDelays() throws IOException, UnsupportedAudioFileException {
+        try (Model model = new Model("model");
+            Recognizer recognizer = new Recognizer(model, 16000)) {
+            recognizer.setEndpointerMode(EndpointerMode.VERY_LONG);
+            recognizer.setEndpointerDelays(5.0f, 3.0f, 50.0f);
+	}
+        Assert.assertTrue(true);
+    }
 
     @Test(expected = IOException.class)
     public void decoderTestException() throws IOException {
         Model model = new Model("model_missing");
     }
+
+    @Test
+    public void testItn() throws IOException {
+        TextProcessor p = new TextProcessor("model/itn/en_itn_tagger.fst", "model/itn/en_itn_verbalizer.fst");
+        System.out.println(p.itn("as easy as one two three"));
+    }
 }

kotlin/.gitignore

@@ -0,0 +1,3 @@
+.gradle/
+.idea/
+build/

kotlin/AUTHORS

@@ -0,0 +1 @@
+Doomsdayrs doomsdayrs@gmail.com

kotlin/MAINTENANCE.md

@@ -0,0 +1,8 @@
+# Maintenance
+
+To maintain this module, please ensure the following.
+1. Kotlin version is kept up to date.
+This will be found in the plugins block of the [build.gradle.kts](./build.gradle.kts).
+Ensure both multiplatform and serialization have the same value.
+2. Ensure dependencies are up to date
+3. Ensure that the android min & target sdks are up to date.

kotlin/README.md

@@ -0,0 +1,63 @@
+# vosk-api-kotlin
+
+The vosk-api wrapped using Kotlin Multiplatform.
+
+## Usage
+
+The following are ways to use this wrapper.
+
+### JVM
+
+For Java & Android targets.
+
+```kotlin
+dependencoes {
+    val voskVersion = "0.4.0-alpha0"
+
+    // Generic
+    implementation("com.alphacephei:vosk-api-kotlin:$voskVersion")
+
+    // Android
+    implementation("com.alphacephei:vosk-api-kotlin-android:$voskVersion")
+}
+```
+
+## Building
+
+To build this project, follow the following steps.
+
+1. Install `libvosk`.
+This can be done from [source][source install] (parent monorepo) 
+or [downloaded][download].
+2. Download a [Vosk model](https://alphacephei.com/vosk/models) to use.
+  - It is suggested to use a small model to speed up tests.
+3. Once both are downloaded and placed into a proper location (hopefully following UNIX specification).
+Set the following environment variables:
+  - `VOSK_MODEL` to the path of the model.
+  - `VOSK_PATH` to the path of `libvosk`.
+These are used by the various tests to operate.
+4. Now that the required steps are complete, one can run `./gradlew build`.
+
+## Kotlin/Native
+
+Currently, the native target is disabled due to a lack of vosk-api packaging on platforms.
+Further worsened by the fact that installing the vosk-api on Linux systems is a chore.
+
+First, either install from [source][source install]
+or [download][download] and install into the proper unix directories as expected in
+[libvosk.def](./src/nativeInterop/cinterop/libvosk.def).
+
+To enable development for Kotlin/Native, do either for the following.
+- Add `NATIVE_EXPERIMENT=true` to your environment.
+- Go into [build.gradle.kts](./build.gradle.kts) & find `enableNative` & set the right side to true.
+
+Afterwards, when syncing the project, the native source sets will become available to work on.
+It is suggested to run `cinteropLibvoskNative` to generate the Kotlin C bindings to work with.
+
+## Future
+
+- Possibly target Kotlin/JS
+- Possibly target Kotlin/Objective-C (?)
+
+[source install]: https://alphacephei.com/vosk/install
+[download]: https://github.com/alphacep/vosk-api/releases/latest

kotlin/build.gradle.kts

@@ -0,0 +1,204 @@
+import org.jetbrains.dokka.gradle.DokkaTask
+import org.jetbrains.kotlin.gradle.ExperimentalKotlinGradlePluginApi
+import org.jetbrains.kotlin.gradle.dsl.JvmTarget
+
+/*
+ * Copyright 2020 Alpha Cephei Inc. & Doomsdayrs
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+plugins {
+	kotlin("multiplatform") version "2.0.0"
+	id("com.android.library")
+	`maven-publish`
+	id("org.jetbrains.dokka") version "1.9.20"
+	kotlin("plugin.serialization") version "2.0.0"
+}
+
+group = "com.alphacephei"
+version = "0.3.75"
+
+repositories {
+	google()
+	mavenCentral()
+}
+
+val dokkaOutputDir = "$buildDir/dokka"
+
+tasks.getByName<DokkaTask>("dokkaHtml") {
+	outputDirectory.set(file(dokkaOutputDir))
+}
+
+val deleteDokkaOutputDir by tasks.register<Delete>("deleteDokkaOutputDirectory") {
+	delete(dokkaOutputDir)
+}
+
+val javadocJar = tasks.register<Jar>("javadocJar") {
+	dependsOn(deleteDokkaOutputDir, tasks.dokkaHtml)
+	archiveClassifier.set("javadoc")
+	from(dokkaOutputDir)
+}
+
+fun org.jetbrains.kotlin.gradle.dsl.KotlinMultiplatformExtension.native(
+	configure: org.jetbrains.kotlin.gradle.plugin.mpp.KotlinNativeTargetWithHostTests.() -> Unit = {}
+) {
+	when {
+		org.jetbrains.kotlin.konan.target.HostManager.hostIsMingw -> mingwX64("native")
+		org.jetbrains.kotlin.konan.target.HostManager.hostIsLinux -> linuxX64("native")
+		org.jetbrains.kotlin.konan.target.HostManager.hostIsMac -> if (org.jetbrains.kotlin.konan.target.HostManager.hostArch() == "arm64") {
+			macosArm64("native")
+		} else {
+			macosX64("native")
+		}
+
+		else -> error("Unsupported Host OS: ${org.jetbrains.kotlin.konan.target.HostManager.hostOs()}")
+	}.apply(configure)
+}
+
+kotlin {
+	jvm {
+		@OptIn(ExperimentalKotlinGradlePluginApi::class)
+		compilerOptions {
+			jvmTarget.set(JvmTarget.JVM_17)
+		}
+
+		testRuns["test"].executionTask.configure {
+			useJUnitPlatform()
+			environment("MODEL", "VOSK_MODEL")
+			//environment("MODEL", "/home/doomsdayrs/Downloads/vosk-model-small-en-us-0.15/")
+			environment("LIBRARY", "VOSK_PATH")
+			//environment("LIBRARY", "/usr/local/lib64/libvosk/libvosk.so")
+			environment("AUDIO", "$projectDir/../python/example/test.wav")
+		}
+	}
+
+	androidTarget {
+		publishAllLibraryVariants()
+	}
+
+	/**
+	 * If native target should be enabled or not.
+	 *
+	 * Currently disabled as there is no proper packaging distribution currently.
+	 */
+	@Suppress("SimplifyBooleanWithConstants") // Ignore, the false is for overrides
+	val enableNative = System.getenv("NATIVE_EXPERIMENT") == "true" || false
+
+	if (enableNative)
+		native {
+			val main by compilations.getting
+			val libvosk by main.cinterops.creating
+
+			binaries {
+				sharedLib()
+			}
+		}
+
+
+	@OptIn(ExperimentalKotlinGradlePluginApi::class)
+	applyDefaultHierarchyTemplate {
+		withJvm()
+		withAndroidTarget()
+
+		if (enableNative)
+			withNative()
+	}
+
+	publishing {
+		publications {
+			withType<MavenPublication> {
+				artifact(javadocJar)
+				pom {
+					url.set("http://www.alphacephei.com.com/vosk/")
+					licenses {
+						license {
+							name.set("The Apache License, Version 2.0")
+							url.set("http://www.apache.org/licenses/LICENSE-2.0.txt")
+						}
+					}
+					developers {
+						developer {
+							id.set("com.alphacephei")
+							name.set("Alpha Cephei Inc")
+							email.set("contact@alphacephei.com")
+						}
+					}
+					scm {
+						connection.set("scm:git:git://github.com/alphacep/vosk-api.git")
+						url.set("https://github.com/alphacep/vosk-api/")
+					}
+				}
+			}
+		}
+	}
+
+	val jna_version = "5.14.0"
+	val coroutines_version = "1.7.3"
+
+	sourceSets {
+		val commonMain by getting {
+			dependencies {
+				api("org.jetbrains.kotlinx:kotlinx-serialization-json:1.7.0")
+				api("org.jetbrains.kotlinx:kotlinx-coroutines-core:$coroutines_version")
+			}
+		}
+		val commonTest by getting {
+			dependencies {
+				implementation(kotlin("test"))
+				implementation("org.jetbrains.kotlinx:kotlinx-coroutines-test:$coroutines_version")
+			}
+		}
+		val jvmMain by getting {
+			dependencies {
+				api("net.java.dev.jna:jna:$jna_version")
+			}
+		}
+		val jvmTest by getting
+		if (enableNative) {
+			val nativeMain by getting
+		}
+		val androidMain by getting {
+			dependsOn(jvmMain)
+			dependencies {
+				api("net.java.dev.jna:jna:$jna_version@aar")
+			}
+		}
+		val androidUnitTest by getting {
+			dependencies {
+				implementation("junit:junit:4.13.2")
+			}
+		}
+	}
+}
+
+android {
+	namespace = "com.alphacephei.library"
+	compileSdk = 34
+	sourceSets["main"].manifest.srcFile("src/androidMain/AndroidManifest.xml")
+	defaultConfig {
+		minSdk = 24
+		targetSdk = 34
+	}
+	compileOptions {
+		sourceCompatibility = JavaVersion.VERSION_17
+		targetCompatibility = JavaVersion.VERSION_17
+	}
+	publishing {
+		multipleVariants {
+			withSourcesJar()
+			withJavadocJar()
+			allVariants()
+		}
+	}
+}

kotlin/settings.gradle.kts

@@ -0,0 +1,31 @@
+/*
+ * Copyright 2020 Alpha Cephei Inc. & Doomsdayrs
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+pluginManagement {
+	repositories {
+		google()
+		gradlePluginPortal()
+		mavenCentral()
+	}
+	resolutionStrategy {
+		eachPlugin {
+			if (requested.id.namespace == "com.android") {
+				useModule("com.android.tools.build:gradle:8.3.0")
+			}
+		}
+	}
+}
+rootProject.name = "vosk-api-kotlin"

kotlin/src/androidMain/AndroidManifest.xml

@@ -0,0 +1,6 @@
+<?xml version="1.0" encoding="utf-8"?>
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+    package="com.alphacephei.library">
+
+    <uses-permission android:name="android.permission.RECORD_AUDIO" />
+</manifest>

kotlin/src/androidMain/kotlin/org/vosk/android/RecognitionListener.kt

@@ -0,0 +1,46 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.vosk.android
+
+/**
+ * Interface to receive recognition results
+ */
+interface RecognitionListener {
+	/**
+	 * Called when partial recognition result is available.
+	 */
+	fun onPartialResult(hypothesis: String)
+
+	/**
+	 * Called after silence occured.
+	 */
+	fun onResult(hypothesis: String)
+
+	/**
+	 * Called after stream end.
+	 */
+	fun onFinalResult(hypothesis: String)
+
+	/**
+	 * Called when an error occurs.
+	 */
+	fun onError(exception: Exception)
+
+	/**
+	 * Called after timeout expired
+	 */
+	fun onTimeout()
+}

kotlin/src/androidMain/kotlin/org/vosk/android/SpeechService.kt

@@ -0,0 +1,241 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.vosk.android
+
+import android.annotation.SuppressLint
+import android.media.AudioFormat
+import android.media.AudioRecord
+import android.media.MediaRecorder.AudioSource
+import android.os.Handler
+import android.os.Looper
+import org.vosk.Recognizer
+import java.io.IOException
+import kotlin.math.roundToInt
+
+/**
+ * Service that records audio in a thread, passes it to a recognizer and emits
+ * recognition results. Recognition events are passed to a client using
+ * [RecognitionListener]
+ */
+class SpeechService @Throws(IOException::class) constructor(
+	private val recognizer: Recognizer,
+	sampleRate: Float
+) {
+	private val sampleRate: Int
+	private val bufferSize: Int
+	private val recorder: AudioRecord
+	private var recognizerThread: RecognizerThread? = null
+	private val mainHandler = Handler(Looper.getMainLooper())
+
+	/**
+	 * Creates speech service. Service holds the AudioRecord object, so you
+	 * need to call [.shutdown] in order to properly finalize it.
+	 *
+	 * @throws IOException thrown if audio recorder can not be created for some reason.
+	 */
+	init {
+		this.sampleRate = sampleRate.toInt()
+		bufferSize = (this.sampleRate * BUFFER_SIZE_SECONDS).roundToInt()
+		@SuppressLint("MissingPermission")
+		recorder = AudioRecord(
+			AudioSource.VOICE_RECOGNITION, this.sampleRate,
+			AudioFormat.CHANNEL_IN_MONO,
+			AudioFormat.ENCODING_PCM_16BIT, bufferSize * 2
+		)
+		if (recorder.state == AudioRecord.STATE_UNINITIALIZED) {
+			recorder.release()
+			throw IOException(
+				"Failed to initialize recorder. Microphone might be already in use."
+			)
+		}
+	}
+
+	/**
+	 * Starts recognition. Does nothing if recognition is active.
+	 *
+	 * @return true if recognition was actually started
+	 */
+	fun startListening(listener: RecognitionListener): Boolean {
+		if (null != recognizerThread) return false
+		recognizerThread = RecognizerThread(listener)
+		recognizerThread!!.start()
+		return true
+	}
+
+	/**
+	 * Starts recognition. After specified timeout listening stops and the
+	 * endOfSpeech signals about that. Does nothing if recognition is active.
+	 *
+	 *
+	 * timeout - timeout in milliseconds to listen.
+	 *
+	 * @return true if recognition was actually started
+	 */
+	fun startListening(listener: RecognitionListener, timeout: Int): Boolean {
+		if (null != recognizerThread) return false
+		recognizerThread = RecognizerThread(listener, timeout)
+		recognizerThread!!.start()
+		return true
+	}
+
+	private fun stopRecognizerThread(): Boolean {
+		if (null == recognizerThread) return false
+		try {
+			recognizerThread!!.interrupt()
+			recognizerThread!!.join()
+		} catch (e: InterruptedException) {
+			// Restore the interrupted status.
+			Thread.currentThread().interrupt()
+		}
+		recognizerThread = null
+		return true
+	}
+
+	/**
+	 * Stops recognition. Listener should receive final result if there is
+	 * any. Does nothing if recognition is not active.
+	 *
+	 * @return true if recognition was actually stopped
+	 */
+	fun stop(): Boolean {
+		return stopRecognizerThread()
+	}
+
+	/**
+	 * Cancel recognition. Do not post any new events, simply cancel processing.
+	 * Does nothing if recognition is not active.
+	 *
+	 * @return true if recognition was actually stopped
+	 */
+	fun cancel(): Boolean {
+		if (recognizerThread != null) {
+			recognizerThread!!.setPause(true)
+		}
+		return stopRecognizerThread()
+	}
+
+	/**
+	 * Shutdown the recognizer and release the recorder
+	 */
+	fun shutdown() {
+		recorder.release()
+	}
+
+	fun setPause(paused: Boolean) {
+		if (recognizerThread != null) {
+			recognizerThread!!.setPause(paused)
+		}
+	}
+
+	/**
+	 * Resets recognizer in a thread, starts recognition over again
+	 */
+	fun reset() {
+		if (recognizerThread != null) {
+			recognizerThread!!.reset()
+		}
+	}
+
+	private inner class RecognizerThread @JvmOverloads constructor(
+		var listener: RecognitionListener,
+		timeout: Int = Companion.NO_TIMEOUT
+	) : Thread() {
+		private var remainingSamples: Int
+		private val timeoutSamples: Int
+
+		@Volatile
+		private var paused = false
+
+		@Volatile
+		private var reset = false
+
+		init {
+			timeoutSamples = if (timeout != Companion.NO_TIMEOUT) {
+				timeout * sampleRate / 1000
+			} else {
+				Companion.NO_TIMEOUT
+			}
+			remainingSamples = timeoutSamples
+		}
+
+		/**
+		 * When we are paused, don't process audio by the recognizer and don't emit
+		 * any listener results
+		 *
+		 * @param paused the status of pause
+		 */
+		fun setPause(paused: Boolean) {
+			this.paused = paused
+		}
+
+		/**
+		 * Set reset state to signal reset of the recognizer and start over
+		 */
+		fun reset() {
+			reset = true
+		}
+
+		override fun run() {
+			recorder.startRecording()
+			if (recorder.recordingState == AudioRecord.RECORDSTATE_STOPPED) {
+				recorder.stop()
+				val ioe = IOException(
+					"Failed to start recording. Microphone might be already in use."
+				)
+				mainHandler.post { listener.onError(ioe) }
+			}
+			val buffer = ShortArray(bufferSize)
+			while (!interrupted()
+				&& (timeoutSamples == Companion.NO_TIMEOUT || remainingSamples > 0)
+			) {
+				val nread = recorder.read(buffer, 0, buffer.size)
+				if (paused) {
+					continue
+				}
+				if (reset) {
+					recognizer.reset()
+					reset = false
+				}
+				if (nread < 0) throw RuntimeException("error reading audio buffer")
+				if (recognizer.acceptWaveform(buffer)) {
+					val result = recognizer.result
+					mainHandler.post { listener.onResult(result) }
+				} else {
+					val partialResult = recognizer.partialResult
+					mainHandler.post { listener.onPartialResult(partialResult) }
+				}
+				if (timeoutSamples != NO_TIMEOUT) {
+					remainingSamples -= nread
+				}
+			}
+			recorder.stop()
+			if (!paused) {
+				// If we met timeout signal that speech ended
+				if (timeoutSamples != NO_TIMEOUT && remainingSamples <= 0) {
+					mainHandler.post { listener.onTimeout() }
+				} else {
+					val finalResult = recognizer.finalResult
+					mainHandler.post { listener.onFinalResult(finalResult) }
+				}
+			}
+		}
+	}
+
+	companion object {
+		private const val NO_TIMEOUT = -1
+		private const val BUFFER_SIZE_SECONDS = 0.2f
+	}
+}

kotlin/src/androidMain/kotlin/org/vosk/android/SpeechStreamService.kt

@@ -0,0 +1,151 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.vosk.android
+
+import android.os.Handler
+import android.os.Looper
+import org.vosk.Recognizer
+import java.io.IOException
+import java.io.InputStream
+import kotlin.math.roundToInt
+
+/**
+ * Service that recognizes stream audio in a  thread, passes it to a recognizer and emits
+ * recognition results. Recognition events are passed to a client using
+ * [RecognitionListener]
+ */
+class SpeechStreamService(
+	private val recognizer: Recognizer,
+	inputStream: InputStream,
+	sampleRate: Float
+) {
+	private val inputStream: InputStream
+	private val sampleRate: Int
+	private val bufferSize: Int
+	private var recognizerThread: Thread? = null
+	private val mainHandler = Handler(Looper.getMainLooper())
+
+	/**
+	 * Creates speech service.
+	 */
+	init {
+		this.sampleRate = sampleRate.toInt()
+		this.inputStream = inputStream
+		bufferSize = (this.sampleRate * BUFFER_SIZE_SECONDS * 2).roundToInt()
+	}
+
+	/**
+	 * Starts recognition. Does nothing if recognition is active.
+	 *
+	 * @return true if recognition was actually started
+	 */
+	fun start(listener: RecognitionListener): Boolean {
+		if (null != recognizerThread) return false
+		recognizerThread = RecognizerThread(listener)
+		recognizerThread!!.start()
+		return true
+	}
+
+	/**
+	 * Starts recognition. After specified timeout listening stops and the
+	 * endOfSpeech signals about that. Does nothing if recognition is active.
+	 *
+	 *
+	 * timeout - timeout in milliseconds to listen.
+	 *
+	 * @return true if recognition was actually started
+	 */
+	fun start(listener: RecognitionListener, timeout: Int): Boolean {
+		if (null != recognizerThread) return false
+		recognizerThread = RecognizerThread(listener, timeout)
+		recognizerThread!!.start()
+		return true
+	}
+
+	/**
+	 * Stops recognition. All listeners should receive final result if there is
+	 * any. Does nothing if recognition is not active.
+	 *
+	 * @return true if recognition was actually stopped
+	 */
+	fun stop(): Boolean {
+		if (null == recognizerThread) return false
+		try {
+			recognizerThread!!.interrupt()
+			recognizerThread!!.join()
+		} catch (e: InterruptedException) {
+			// Restore the interrupted status.
+			Thread.currentThread().interrupt()
+		}
+		recognizerThread = null
+		return true
+	}
+
+	private inner class RecognizerThread @JvmOverloads constructor(
+		var listener: RecognitionListener,
+		timeout: Int = Companion.NO_TIMEOUT
+	) : Thread() {
+		private var remainingSamples: Int
+		private val timeoutSamples: Int
+
+		init {
+			if (timeout != Companion.NO_TIMEOUT) timeoutSamples =
+				timeout * sampleRate / 1000 else timeoutSamples = Companion.NO_TIMEOUT
+			remainingSamples = timeoutSamples
+		}
+
+		override fun run() {
+			val buffer = ByteArray(bufferSize)
+			while (!interrupted()
+				&& (timeoutSamples == Companion.NO_TIMEOUT || remainingSamples > 0)
+			) {
+				try {
+					val nread = inputStream.read(buffer, 0, buffer.size)
+					if (nread < 0) {
+						break
+					} else {
+						val isSilence: Boolean = recognizer.acceptWaveform(buffer)
+						if (isSilence) {
+							val result = recognizer.result
+							mainHandler.post { listener.onResult(result) }
+						} else {
+							val partialResult = recognizer.partialResult
+							mainHandler.post { listener.onPartialResult(partialResult) }
+						}
+					}
+					if (timeoutSamples != NO_TIMEOUT) {
+						remainingSamples -= nread
+					}
+				} catch (e: IOException) {
+					mainHandler.post { listener.onError(e) }
+				}
+			}
+
+			// If we met timeout signal that speech ended
+			if (timeoutSamples != NO_TIMEOUT && remainingSamples <= 0) {
+				mainHandler.post { listener.onTimeout() }
+			} else {
+				val finalResult = recognizer.finalResult
+				mainHandler.post { listener.onFinalResult(finalResult) }
+			}
+		}
+	}
+
+	companion object {
+		private const val NO_TIMEOUT = -1
+		private const val BUFFER_SIZE_SECONDS = 0.2f
+	}
+}

kotlin/src/androidMain/kotlin/org/vosk/android/StorageService.kt

@@ -0,0 +1,138 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.vosk.android
+
+import android.content.Context
+import android.content.res.AssetManager
+import android.os.Environment
+import android.os.Handler
+import android.os.Looper
+import android.util.Log
+import org.vosk.Model
+import java.io.*
+import java.util.concurrent.Executor
+import java.util.concurrent.Executors
+import java.util.function.Consumer
+
+/**
+ * Provides utility methods to sync model files to external storage to allow
+ * C++ code access them. Relies on file named "uuid" to track updates.
+ */
+object StorageService {
+	private val TAG = StorageService::class.simpleName
+
+	@JvmStatic
+	fun unpack(
+		context: Context,
+		sourcePath: String,
+		targetPath: String,
+		completeCallback: Consumer<Model>,
+		errorCallback: Consumer<IOException>
+	) {
+		val executor: Executor =
+			Executors.newSingleThreadExecutor() // change according to your requirements
+		val handler = Handler(Looper.getMainLooper())
+		executor.execute {
+			try {
+				val outputPath = sync(context, sourcePath, targetPath)
+				val model = Model(outputPath)
+				handler.post { completeCallback.accept(model) }
+			} catch (e: IOException) {
+				handler.post { errorCallback.accept(e) }
+			}
+		}
+	}
+
+	@JvmStatic
+	@Throws(IOException::class)
+	fun sync(context: Context, sourcePath: String, targetPath: String): String {
+		val assetManager = context.assets
+		val externalFilesDir = context.getExternalFilesDir(null)
+			?: throw IOException(
+				"cannot get external files dir, "
+						+ "external storage state is " + Environment.getExternalStorageState()
+			)
+		val targetDir = File(externalFilesDir, targetPath)
+		val resultPath = File(targetDir, sourcePath).absolutePath
+		val sourceUUID = readLine(assetManager.open("$sourcePath/uuid"))
+		try {
+			val targetUUID = readLine(FileInputStream(File(targetDir, "$sourcePath/uuid")))
+			if (targetUUID == sourceUUID) return resultPath
+		} catch (e: FileNotFoundException) {
+			// ignore
+		}
+		deleteContents(targetDir)
+		copyAssets(assetManager, sourcePath, targetDir)
+
+		// Copy uuid
+		copyFile(assetManager, "$sourcePath/uuid", targetDir)
+		return resultPath
+	}
+
+	@Throws(IOException::class)
+	private fun readLine(inputStream: InputStream): String {
+		return BufferedReader(InputStreamReader(inputStream)).use { it.readLine() }
+	}
+
+	private fun deleteContents(dir: File): Boolean {
+		val files = dir.listFiles()
+		var success = true
+		if (files != null) {
+			for (file in files) {
+				if (file.isDirectory) {
+					success = success and deleteContents(file)
+				}
+				if (!file.delete()) {
+					success = false
+				}
+			}
+		}
+		return success
+	}
+
+	@Throws(IOException::class)
+	private fun copyAssets(assetManager: AssetManager, path: String, outPath: File) {
+		val assets = assetManager.list(path) ?: return
+		if (assets.isEmpty()) {
+			if (!path.endsWith("uuid")) copyFile(assetManager, path, outPath)
+		} else {
+			val dir = File(outPath, path)
+			if (!dir.exists()) {
+				Log.v(TAG, "Making directory " + dir.absolutePath)
+				if (!dir.mkdirs()) {
+					Log.v(TAG, "Failed to create directory " + dir.absolutePath)
+				}
+			}
+			for (asset in assets) {
+				copyAssets(assetManager, "$path/$asset", outPath)
+			}
+		}
+	}
+
+	@Throws(IOException::class)
+	private fun copyFile(assetManager: AssetManager, fileName: String, outPath: File) {
+		Log.v(TAG, "Copy $fileName to $outPath")
+		assetManager.open(fileName).use { inputStream ->
+			FileOutputStream("$outPath/$fileName").use { out ->
+				val buffer = ByteArray(4000)
+				var read: Int
+				while (inputStream.read(buffer).also { read = it } != -1) {
+					out.write(buffer, 0, read)
+				}
+			}
+		}
+	}
+}

kotlin/src/commonMain/kotlin/org/vosk/AcceptWaveformException.kt

@@ -0,0 +1,22 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+/**
+ * Thrown when a [Recognizer] cannot accept a given waveform
+ */
+class AcceptWaveformException(data: Any) : Exception("Could not accept waveform: $data")

kotlin/src/commonMain/kotlin/org/vosk/BatchModel.kt

@@ -0,0 +1,38 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import org.vosk.exception.ModelException
+
+/**
+ * Batch model object
+ *
+ * @since 26 / 12 / 2022
+ * @constructor Creates the batch recognizer object
+ */
+expect class BatchModel @Throws(ModelException::class) constructor(path: String) : Freeable {
+
+	/**
+	 *  Releases batch model object
+	 */
+	override fun free()
+
+	/**
+	 * Wait for the processing
+	 */
+	fun await()
+}

kotlin/src/commonMain/kotlin/org/vosk/BatchRecognizer.kt

@@ -0,0 +1,62 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+/**
+ * Batch recognizer object
+ *
+ * @since 26 / 12 / 2022
+ * @constructor Creates batch recognizer object
+ */
+expect class BatchRecognizer constructor(model: BatchModel, sampleRate: Float) : Freeable {
+
+	/**
+	 * Releases batch recognizer object
+	 */
+	override fun free()
+
+	/**
+	 * Accept batch voice data
+	 */
+	fun acceptWaveform(data: ByteArray)
+
+	/**
+	 * Set NLSML output
+	 * @param nlsml - boolean value
+	 */
+	fun setNLSML(nlsml: Boolean)
+
+	/**
+	 *  Closes the stream
+	 */
+	fun finishStream()
+
+	/**
+	 * Return results
+	 */
+	val frontResult: String
+
+	/**
+	 * Release and free first retrieved result
+	 */
+	fun pop()
+
+	/**
+	 * Get amount of pending chunks for more intelligent waiting
+	 */
+	val pendingChunks: Int
+}

kotlin/src/commonMain/kotlin/org/vosk/EndPointerMode.kt

@@ -0,0 +1,27 @@
+/*
+ * Copyright 2024 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+/**
+ * VoskEpMode
+ */
+enum class EndPointerMode {
+	ANSWER_DEFAULT,
+	ANSWER_SHORT,
+	ANSWER_LONG,
+	ANSWER_VERY_LONG
+}

kotlin/src/commonMain/kotlin/org/vosk/Freeable.kt

@@ -0,0 +1,31 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+/**
+ * Denotes an object that must be freed afterwards.
+ *
+ * On JVM, This is done via AutoClosable.
+ */
+@Suppress("SpellCheckingInspection")
+interface Freeable {
+
+	/**
+	 * Dereference the related object
+	 */
+	fun free()
+}

kotlin/src/commonMain/kotlin/org/vosk/LogLevel.kt

@@ -0,0 +1,40 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+/**
+ * Log level for Kaldi messages.
+ *
+ * @since 26 / 12 / 2022
+ */
+enum class LogLevel(val value: Int) {
+
+	/**
+	 * Don't print info messages
+	 */
+	WARNINGS(-1),
+
+	/**
+	 * Default value to print info and error messages but no debug
+	 */
+	INFO(0),
+
+	/**
+	 * More verbose mode
+	 */
+	DEBUG(1);
+}

kotlin/src/commonMain/kotlin/org/vosk/Model.kt

@@ -0,0 +1,52 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import org.vosk.exception.IOException
+
+
+/**
+ * Model stores all the data required for recognition
+ *
+ * It contains static data and can be shared across processing
+ * threads.
+ *
+ * @since 26 / 12 / 2022
+ * @constructor Loads model data from the file and returns the model object
+ * @param path the path of the model on the filesystem
+ * @throws IOException if the path provided is invalid
+ */
+expect class Model @Throws(IOException::class) constructor(path: String) : Freeable {
+
+	/**
+	 * Check if a word can be recognized by the model
+	 * @param word: the word
+	 * @returns the word symbol if @param word exists inside the model
+	 * or -1 otherwise.
+	 * Reminding that word symbol 0 is for <epsilon>
+	 */
+	fun findWord(word: String): Int
+
+	/**
+	 * Releases the model memory
+	 *
+	 *  The model object is reference-counted so if some recognizer
+	 *  depends on this model, model might still stay alive. When
+	 *  last recognizer is released, model will be released too.
+	 */
+	override fun free()
+}

kotlin/src/commonMain/kotlin/org/vosk/Recognizer.kt

@@ -0,0 +1,273 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import org.vosk.exception.RecognizerException
+
+/**
+ * Recognizer object is the main object which processes data.
+ *
+ * Each recognizer usually runs in own thread and takes audio as input.
+ * Once audio is processed recognizer returns JSON object as a string
+ * which represent decoded information - words, confidences, times, n-best lists,
+ * speaker information and so on
+ *
+ * @since 26 / 12 / 2022
+ */
+expect class Recognizer : Freeable {
+	/**
+	 * Creates the recognizer object
+	 *
+	 *  The recognizers process the speech and return text using shared model data
+	 *  @param model       VoskModel containing static data for recognizer. Model can be
+	 *                     shared across recognizers, even running in different threads.
+	 *  @param sampleRate The sample rate of the audio you going to feed into the recognizer.
+	 *                     Make sure this rate matches the audio content, it is a common
+	 *                     issue causing accuracy problems.
+	 * @throws RecognizerException if a problem occurred
+	 */
+	@Throws(RecognizerException::class)
+	constructor(model: Model, sampleRate: Float)
+
+	/**
+	 * Creates the recognizer object with speaker recognition
+	 *
+	 *  With the speaker recognition mode the recognizer not just recognize
+	 *  text but also return speaker vectors one can use for speaker identification
+	 *
+	 *  @param model       VoskModel containing static data for recognizer. Model can be
+	 *                     shared across recognizers, even running in different threads.
+	 *  @param sampleRate The sample rate of the audio you going to feed into the recognizer.
+	 *                     Make sure this rate matches the audio content, it is a common
+	 *                     issue causing accuracy problems.
+	 *  @param speakerModel speaker model for speaker identification
+	 * @throws RecognizerException if a problem occurred
+	 */
+	@Throws(RecognizerException::class)
+	constructor(model: Model, sampleRate: Float, speakerModel: SpeakerModel)
+
+	/**
+	 * Creates the recognizer object with the phrase list
+	 *
+	 *  Sometimes when you want to improve recognition accuracy and when you don't need
+	 *  to recognize large vocabulary you can specify a list of phrases to recognize. This
+	 *  will improve recognizer speed and accuracy but might return [unk] if user said
+	 *  something different.
+	 *
+	 *  Only recognizers with lookahead models support this type of quick configuration.
+	 *  Precompiled HCLG graph models are not supported.
+	 *
+	 *  @param model       VoskModel containing static data for recognizer. Model can be
+	 *                     shared across recognizers, even running in different threads.
+	 *  @param sampleRate The valuesample rate of the audio you going to feed into the recognizer.
+	 *                     Make sure this rate matches the audio content, it is a common
+	 *                     issue causing accuracy problems.
+	 *  @param grammar The string with the list of phrases to recognize as JSON array of strings,
+	 *                 for example "["one two three four five", "[unk]"]".
+	 *
+	 * @throws RecognizerException if a problem occurred
+	 */
+	@Throws(RecognizerException::class)
+	constructor(model: Model, sampleRate: Float, grammar: String)
+
+	/**
+	 * Adds speaker model to already initialized recognizer
+	 *
+	 * Can add speaker recognition model to already created recognizer. Helps to initialize
+	 * speaker recognition for grammar-based recognizer.
+	 *
+	 * @param speakerModel Speaker recognition model
+	 */
+	fun setSpeakerModel(speakerModel: SpeakerModel)
+
+
+	/**
+	 * Reconfigures recognizer to use grammar
+	 *
+	 * @param grammar      Set of phrases in JSON array of strings or "[]" to use default model graph.
+	 *                     See also vosk_recognizer_new_grm
+	 */
+	fun setGrammar(grammar: String)
+
+	/**
+	 * Configures recognizer to output n-best results
+	 *
+	 * <pre>
+	 *   {
+	 *      "alternatives": [
+	 *          { "text": "one two three four five", "confidence": 0.97 },
+	 *          { "text": "one two three for five", "confidence": 0.03 },
+	 *      ]
+	 *   }
+	 * </pre>
+	 *
+	 * @param maxAlternatives - maximum alternatives to return from recognition results
+	 */
+	fun setMaxAlternatives(maxAlternatives: Int)
+
+	/**
+	 * Enables words with times in the output
+	 *
+	 * <pre>
+	 *   "result" : [{
+	 *       "conf" : 1.000000,
+	 *       "end" : 1.110000,
+	 *       "start" : 0.870000,
+	 *       "word" : "what"
+	 *     }, {
+	 *       "conf" : 1.000000,
+	 *       "end" : 1.530000,
+	 *       "start" : 1.110000,
+	 *       "word" : "zero"
+	 *     }, {
+	 *       "conf" : 1.000000,
+	 *       "end" : 1.950000,
+	 *       "start" : 1.530000,
+	 *       "word" : "zero"
+	 *     }, {
+	 *       "conf" : 1.000000,
+	 *       "end" : 2.340000,
+	 *       "start" : 1.950000,
+	 *       "word" : "zero"
+	 *     }, {
+	 *       "conf" : 1.000000,
+	 *       "end" : 2.610000,
+	 *       "start" : 2.340000,
+	 *       "word" : "one"
+	 *     }],
+	 * </pre>
+	 *
+	 * C equivalent = vosk_recognizer_set_words
+	 * @param words - boolean value
+	 */
+	fun setOutputWordTimes(words: Boolean)
+
+	/**
+	 * Like [setOutputWordTimes] return words and confidences in partial results
+	 *
+	 * @param partialWords - boolean value
+	 */
+	fun setPartialWords(partialWords: Boolean)
+
+	/**
+	 * Set NLSML output
+	 * @param nlsml - boolean value
+	 */
+	fun setNLSML(nlsml: Boolean)
+
+
+	/**
+	 * Set endpointer scaling factor
+	 *
+	 * @param mode Endpointer mode
+	 **/
+	fun setEndPointerMode(mode: EndPointerMode)
+
+	/**
+	 * Set endpointer delays
+	 *
+	 * @param tStartMax timeout for stopping recognition in case of initial silence (usually around 5.0)
+	 * @param tEnd      timeout for stopping recognition in milliseconds after we recognized something (usually around 0.5 - 1.0)
+	 * @param tMax      timeout for forcing utterance end in milliseconds (usually around 20-30)
+	 **/
+	fun setEndPointerDelays(tStartMax: Float, tEnd: Float, tMax: Float)
+
+	/**
+	 * Accept voice data
+	 *
+	 *  accept and process new chunk of voice data
+	 *
+	 *  @param data Audio data in PCM 16-bit mono format.
+	 *  @param length Length of the audio data.
+	 *  @returns
+	 *       1 - If silence is occurred and you can retrieve a new utterance with result method
+	 *       0 - If decoding continues
+	 *      -1 - If exception occurred
+	 */
+	@Throws(AcceptWaveformException::class)
+	fun acceptWaveform(data: ByteArray): Boolean
+
+	/**
+	 * Same as [acceptWaveform] but the version with the short data for language bindings where you have
+	 *  audio as array of shorts
+	 */
+	@Throws(AcceptWaveformException::class)
+	fun acceptWaveform(data: ShortArray): Boolean
+
+	/**
+	 * Same as [acceptWaveform] but the version with the float data for language bindings where you have
+	 *  audio as array of floats
+	 */
+	@Throws(AcceptWaveformException::class)
+	fun acceptWaveform(data: FloatArray): Boolean
+
+	/**
+	 * Returns speech recognition result
+	 *
+	 * @returns the result in JSON format which contains decoded line, decoded
+	 *          words, times in seconds and confidences. You can parse this result
+	 *          with any json parser
+	 *
+	 * <pre>
+	 *  {
+	 *    "text" : "what zero zero zero one"
+	 *  }
+	 * </pre>
+	 *
+	 * If alternatives enabled it returns result with alternatives, see also [setMaxAlternatives].
+	 *
+	 * If word times enabled returns word time, see also [setOutputWordTimes].
+	 */
+	val result: String
+
+	/**
+	 * Returns partial speech recognition
+	 *
+	 * @returns partial speech recognition text which is not yet finalized.
+	 *          result may change as recognizer process more data.
+	 *
+	 * <pre>
+	 * {
+	 *    "partial" : "cyril one eight zero"
+	 * }
+	 * </pre>
+	 */
+	val finalResult: String
+
+	/**
+	 * Returns speech recognition result. Same as result, but doesn't wait for silence
+	 *  You usually call it in the end of the stream to get final bits of audio. It
+	 *  flushes the feature pipeline, so all remaining audio chunks got processed.
+	 *
+	 *  @returns speech result in JSON format.
+	 */
+	val partialResult: String
+
+	/**
+	 * Resets the recognizer
+	 *
+	 *  Resets current results so the recognition can continue from scratch
+	 */
+	fun reset()
+
+	/**
+	 * Releases recognizer object
+	 *
+	 *  Underlying model is also unreferenced and if needed released
+	 */
+	override fun free()
+}

kotlin/src/commonMain/kotlin/org/vosk/SpeakerModel.kt

@@ -0,0 +1,40 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import org.vosk.exception.ModelException
+
+/**
+ * Speaker model is the same as model but contains the data
+ * for speaker identification.
+ *
+ * @since 26 / 12 / 2022
+ * @constructor Loads speaker model data from the file and returns the model object
+ * @param path the path of the model on the filesystem
+ * @throws ModelException if the path provided is invalid
+ */
+expect class SpeakerModel @Throws(ModelException::class) constructor(path: String) : Freeable {
+
+	/**
+	 * Releases the model memory
+	 *
+	 * The model object is reference-counted so if some recognizer
+	 * depends on this model, model might still stay alive. When
+	 * last recognizer is released, model will be released too.
+	 */
+	override fun free()
+}

kotlin/src/commonMain/kotlin/org/vosk/TextProcessor.kt

@@ -0,0 +1,32 @@
+/*
+ * Copyright 2024 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+/**
+ * Inverse text normalization
+ *
+ * @since 2024/06/19
+ * @constructor Create text processor
+ */
+expect class TextProcessor constructor(tagger: Char, verbalizer: Char) : Freeable {
+
+	/** Release text processor */
+	override fun free()
+
+	/** Convert string */
+	fun itn(input: Char): Char
+}

kotlin/src/commonMain/kotlin/org/vosk/Vosk.kt

@@ -0,0 +1,46 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+/**
+ * 26 / 12 / 2022
+ *
+ * Control overarching features of libvosk.
+ */
+expect object Vosk {
+	/**
+	 * Set log level for Kaldi messages
+	 *
+	 *  @param logLevel the level
+	 */
+	fun setLogLevel(logLevel: LogLevel)
+
+	/**
+	 *  Init, automatically select a CUDA device and allow multithreading.
+	 *  Must be called once from the main thread.
+	 *  Has no effect if HAVE_CUDA flag is not set.
+	 */
+	fun gpuInit()
+
+
+	/**
+	 *  Init CUDA device in a multi-threaded environment.
+	 *  Must be called for each thread.
+	 *  Has no effect if HAVE_CUDA flag is not set.
+	 */
+	fun gpuThreadInit()
+}

kotlin/src/commonMain/kotlin/org/vosk/exception/IOException.kt

@@ -0,0 +1,22 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk.exception
+
+/**
+ * Internal common IO exception. On JVM this is just a type alias.
+ */
+expect open class IOException(message: String?) : Exception

kotlin/src/commonMain/kotlin/org/vosk/exception/ModelException.kt

@@ -0,0 +1,22 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk.exception
+
+/**
+ * Thrown when there is an exception creating a model.
+ */
+class ModelException(path: String): IOException("Failed to find model: $path")

kotlin/src/commonMain/kotlin/org/vosk/exception/RecognizerException.kt

@@ -0,0 +1,22 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk.exception
+
+/**
+ * Thrown when the recognizer fails to be created
+ */
+class RecognizerException: IOException("Failed to create recognizer.")

kotlin/src/commonMain/kotlin/org/vosk/json/Alternative.kt

@@ -0,0 +1,31 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk.json
+
+import kotlinx.serialization.Serializable
+
+/**
+ * Represents an alternative result.
+ *
+ * @since 26 / 12 / 2022
+ */
+@Serializable
+data class Alternative(
+	val confidence: Double,
+	val result: List<Result> = emptyList(),
+	val text: String
+)

kotlin/src/commonMain/kotlin/org/vosk/json/Extension.kt

@@ -0,0 +1,61 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk.json
+
+import kotlinx.serialization.decodeFromString
+import kotlinx.serialization.json.Json
+import kotlinx.serialization.json.encodeToJsonElement
+import org.vosk.Recognizer
+import org.vosk.Model
+import org.vosk.exception.RecognizerException
+
+/*
+ * 26 / 12 / 2022
+ */
+
+/**
+ * Vosk JSON encoder
+ */
+val voskJson = Json { encodeDefaults = true }
+
+/**
+ * Get the result as a JSON object
+ */
+fun Recognizer.resultAsJson(): ResultOutput =
+	voskJson.decodeFromString(result)
+
+/**
+ * Get the final result as a JSON object
+ */
+fun Recognizer.finalResultAsJson(): ResultOutput =
+	voskJson.decodeFromString(finalResult)
+
+/**
+ * Get the partial result as a JSON object
+ */
+fun Recognizer.partialResultAsJson(): PartialResultOutput =
+	voskJson.decodeFromString(partialResult)
+
+
+/**
+ * Create a [Recognizer], but using a list for grammar instead.
+ *
+ * The grammar list is converted into a JSON array
+ */
+@Throws(RecognizerException::class)
+fun Recognizer(model: Model, sampleRate: Float, grammar: List<String>) =
+	Recognizer(model, sampleRate, voskJson.encodeToJsonElement(grammar).toString())

kotlin/src/commonMain/kotlin/org/vosk/json/PartialResultOutput.kt

@@ -0,0 +1,32 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk.json
+
+import kotlinx.serialization.SerialName
+import kotlinx.serialization.Serializable
+
+/**
+ * Represents a partial result
+ *
+ * @since 26 / 12 / 2022
+ */
+@Serializable
+data class PartialResultOutput(
+	val partial: String,
+	@SerialName("partial_result")
+	val partialResult: List<Result> = emptyList(),
+)

kotlin/src/commonMain/kotlin/org/vosk/json/Result.kt

@@ -0,0 +1,32 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk.json
+
+import kotlinx.serialization.Serializable
+
+/**
+ * Represents a result for any given word.
+ *
+ * @since 26 / 12 / 2022
+ */
+@Serializable
+data class Result(
+	val conf: Double? = null,
+	val end: Double,
+	val start: Double,
+	val word: String,
+)

kotlin/src/commonMain/kotlin/org/vosk/json/ResultOutput.kt

@@ -0,0 +1,31 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk.json
+
+import kotlinx.serialization.Serializable
+
+/**
+ * Result output
+ *
+ * @since 26 / 12 / 2022
+ */
+@Serializable
+data class ResultOutput(
+	val alternatives: List<Alternative> = emptyList(),
+	val result: List<Result> = emptyList(),
+	val text: String? = null
+)

kotlin/src/commonMain/kotlin/org/vosk/json/WaveformResult.kt

@@ -0,0 +1,38 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk.json
+
+/**
+ * For extension functions transforming (input streams->recognizers) into flows.
+ */
+sealed interface WaveformResult {
+
+	/**
+	 * A result made after a period of silence.
+	 */
+	data class Result(val result: ResultOutput) : WaveformResult
+
+	/**
+	 * A partial result that is being made in progress.
+	 */
+	data class PartialResult(val result: PartialResultOutput) : WaveformResult
+
+	/**
+	 * A final result made after there is no more content to feed.
+	 */
+	data class FinalResult(val result: ResultOutput) : WaveformResult
+}

kotlin/src/jvmMain/kotlin/org/vosk/BatchModel.kt

@@ -0,0 +1,83 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import com.sun.jna.PointerType
+import org.vosk.exception.ModelException
+import java.io.File
+import java.nio.file.Path
+import kotlin.io.path.absolutePathString
+
+
+/**
+ * Batch model object
+ *
+ * @since 26 / 12 / 2022
+ */
+actual class BatchModel : Freeable, PointerType, AutoCloseable {
+
+	/**
+	 * Empty constructor for JNA
+	 */
+	constructor()
+
+	/**
+	 * Creates the batch recognizer object
+	 */
+	@Throws(ModelException::class)
+	actual constructor(path: String) : super(
+		LibVosk.vosk_batch_model_new(path) ?: throw ModelException(path)
+	)
+
+	/**
+	 * Constructor using a Path, will retrieve absolutePath
+	 *
+	 * @param path to batch model
+	 */
+	@Throws(ModelException::class)
+	constructor(path: Path) : this(path.absolutePathString())
+
+	/**
+	 * Constructor using a File, will retrieve absolutePath
+	 *
+	 * @param file to batch model
+	 */
+	@Throws(ModelException::class)
+	constructor(file: File) : this(file.absolutePath)
+
+	/**
+	 * Releases batch model object
+	 */
+	actual override fun free() {
+		LibVosk.vosk_batch_model_free(this)
+	}
+
+	/**
+	 * Wait for the processing
+	 */
+	actual fun await() {
+		LibVosk.vosk_batch_model_wait(this)
+	}
+
+	/**
+	 * @see free
+	 */
+	override fun close() {
+		free()
+	}
+
+}

kotlin/src/jvmMain/kotlin/org/vosk/BatchRecognizer.kt

@@ -0,0 +1,95 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import com.sun.jna.PointerType
+
+
+/**
+ * Batch recognizer object
+ *
+ * @since 26 / 12 / 2022
+ * @constructor Creates batch recognizer object
+ */
+actual class BatchRecognizer : Freeable, PointerType, AutoCloseable {
+
+	/**
+	 * Empty constructor for JNA
+	 */
+	constructor()
+
+	/**
+	 * Creates batch recognizer object
+	 */
+	actual constructor(model: BatchModel, sampleRate: Float) :
+			super(LibVosk.vosk_batch_recognizer_new(model, sampleRate))
+
+	/**
+	 * Releases batch recognizer object
+	 */
+	actual override fun free() {
+		LibVosk.vosk_batch_recognizer_free(this);
+	}
+
+	/**
+	 * Accept batch voice data
+	 */
+	actual fun acceptWaveform(data: ByteArray) {
+		LibVosk.vosk_batch_recognizer_accept_waveform(this, data, data.size)
+	}
+
+	/**
+	 * Set NLSML output
+	 * @param nlsml - boolean value
+	 */
+	actual fun setNLSML(nlsml: Boolean) {
+		LibVosk.vosk_batch_recognizer_set_nlsml(this, nlsml)
+	}
+
+	/**
+	 *  Closes the stream
+	 */
+	actual fun finishStream() {
+		LibVosk.vosk_batch_recognizer_finish_stream(this)
+	}
+
+	/**
+	 * Return results
+	 */
+	actual val frontResult: String
+		get() = LibVosk.vosk_batch_recognizer_front_result(this)
+
+	/**
+	 * Release and free first retrieved result
+	 */
+	actual fun pop() {
+		LibVosk.vosk_batch_recognizer_pop(this)
+	}
+
+	/**
+	 * Get amount of pending chunks for more intelligent waiting
+	 */
+	actual val pendingChunks: Int
+		get() = LibVosk.vosk_batch_recognizer_get_pending_chunks(this)
+
+	/**
+	 * @see free
+	 */
+	override fun close() {
+		free()
+	}
+}

kotlin/src/jvmMain/kotlin/org/vosk/LibVosk.kt

@@ -0,0 +1,213 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import com.sun.jna.Native
+import com.sun.jna.Platform
+import com.sun.jna.Pointer
+import java.io.File
+import java.io.IOException
+import java.io.InputStream
+import java.nio.file.Files
+import java.nio.file.StandardCopyOption
+
+/**
+ * 26 / 12 / 2022
+ */
+@Suppress("FunctionName")
+internal object LibVosk {
+
+	@JvmStatic
+	@Deprecated(
+		"LibVosk is now for internal JNA, use Vosk instead",
+		ReplaceWith("Vosk.setLogLevel(logLevel)", "org.vosk.Vosk")
+	)
+	fun setLogLevel(logLevel: LogLevel) {
+		Vosk.setLogLevel(logLevel)
+	}
+
+	@Throws(IOException::class)
+	private fun unpackDll(targetDir: File, lib: String) {
+		Vosk::class.java.getResourceAsStream("/win32-x86-64/$lib.dll")!!.use {
+			Files.copy(
+				it,
+				File(targetDir, "$lib.dll").toPath(),
+				StandardCopyOption.REPLACE_EXISTING
+			)
+		}
+	}
+
+	init {
+		when {
+			Platform.isAndroid() -> {
+				Native.register(LibVosk::class.java, "vosk")
+			}
+
+			Platform.isWindows() -> {
+				// We have to unpack dependencies
+				try {
+					// To get a tmp folder we unpack small library and mark it for deletion
+					val tmpFile: File = Native.extractFromResourcePath(
+						"/win32-x86-64/empty",
+						LibVosk::class.java.classLoader
+					)
+					val tmpDir = tmpFile.parentFile!!
+					File(tmpDir, tmpFile.name + ".x").createNewFile()
+
+					// Now unpack dependencies
+					unpackDll(tmpDir, "libwinpthread-1");
+					unpackDll(tmpDir, "libgcc_s_seh-1");
+					unpackDll(tmpDir, "libstdc++-6");
+
+				} catch (e: IOException) {
+					// Nothing for now, it will fail on next step
+				} finally {
+					Native.register(LibVosk::class.java, "libvosk");
+				}
+			}
+
+			else -> {
+				Native.register(LibVosk::class.java, "vosk");
+			}
+		}
+	}
+
+
+	external fun vosk_model_new(path: String): Pointer?
+
+	external fun vosk_model_free(model: Model)
+
+	external fun vosk_model_find_word(model: Model, word: String): Int
+
+
+	external fun vosk_spk_model_new(path: String): Pointer?
+
+	external fun vosk_spk_model_free(model: SpeakerModel)
+
+
+	external fun vosk_recognizer_new(model: Model, sampleRate: Float): Pointer?
+
+	external fun vosk_recognizer_new_spk(
+		model: Model,
+		sampleRate: Float,
+		spkModel: SpeakerModel
+	): Pointer?
+
+	external fun vosk_recognizer_new_grm(
+		model: Model,
+		sampleRate: Float,
+		grammar: String?
+	): Pointer?
+
+	external fun vosk_recognizer_set_spk_model(recognizer: Recognizer, spk_model: SpeakerModel)
+
+	external fun vosk_recognizer_set_grm(recognizer: Recognizer, grammar: String)
+
+	external fun vosk_recognizer_set_max_alternatives(recognizer: Recognizer, maxAlternatives: Int)
+
+	external fun vosk_recognizer_set_words(recognizer: Recognizer, words: Boolean)
+
+	external fun vosk_recognizer_set_partial_words(recognizer: Recognizer, partial_words: Boolean)
+
+	external fun vosk_recognizer_set_nlsml(recognizer: Recognizer, nlsml: Boolean)
+
+
+	external fun vosk_recognizer_accept_waveform(
+		recognizer: Recognizer,
+		data: ByteArray?,
+		len: Int
+	): Int
+
+	external fun vosk_recognizer_accept_waveform_s(
+		recognizer: Recognizer,
+		data: ShortArray?,
+		len: Int
+	): Int
+
+	external fun vosk_recognizer_accept_waveform_f(
+		recognizer: Recognizer,
+		data: FloatArray?,
+		len: Int
+	): Int
+
+
+	external fun vosk_recognizer_result(recognizer: Recognizer): String
+
+	external fun vosk_recognizer_final_result(recognizer: Recognizer): String
+
+	external fun vosk_recognizer_partial_result(recognizer: Recognizer): String
+
+
+	external fun vosk_recognizer_reset(recognizer: Recognizer)
+
+	external fun vosk_recognizer_free(recognizer: Recognizer)
+
+	external fun vosk_set_log_level(level: Int)
+
+	external fun vosk_gpu_init()
+
+	external fun vosk_gpu_thread_init()
+
+
+	external fun vosk_batch_model_new(path: String): Pointer?
+
+	external fun vosk_batch_model_free(model: BatchModel)
+
+	external fun vosk_batch_model_wait(model: BatchModel)
+
+	external fun vosk_batch_recognizer_new(batchModel: BatchModel, sampleRate: Float): Pointer
+
+	external fun vosk_batch_recognizer_free(recognizer: BatchRecognizer)
+
+	external fun vosk_batch_recognizer_accept_waveform(
+		recognizer: BatchRecognizer,
+		data: ByteArray?,
+		length: Int
+	)
+
+	external fun vosk_batch_recognizer_set_nlsml(
+		recognizer: BatchRecognizer,
+		nlsml: Boolean
+	)
+
+	external fun vosk_batch_recognizer_finish_stream(
+		recognizer: BatchRecognizer
+	)
+
+	external fun vosk_batch_recognizer_front_result(
+		recognizer: BatchRecognizer
+	): String
+
+	external fun vosk_batch_recognizer_pop(recognizer: BatchRecognizer)
+
+	external fun vosk_batch_recognizer_get_pending_chunks(recognizer: BatchRecognizer): Int
+
+	external fun vosk_text_processor_new(tagger: Char, verbalizer: Char): Pointer
+
+	external fun vosk_text_processor_free(processor: TextProcessor)
+
+	external fun vosk_text_processor_itn(processor: TextProcessor, input: Char): Char
+
+	external fun vosk_recognizer_set_endpointer_mode(recognizer: Recognizer, ordinal: Int)
+
+	external fun vosk_recognizer_set_endpointer_delays(
+		recognizer: Recognizer,
+		tStartMax: Float,
+		tEnd: Float,
+		tMax: Float
+	)
+}

kotlin/src/jvmMain/kotlin/org/vosk/Model.kt

@@ -0,0 +1,95 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import com.sun.jna.PointerType
+import org.vosk.exception.IOException
+import org.vosk.exception.ModelException
+import java.io.File
+import java.nio.file.Path
+import kotlin.io.path.absolutePathString
+
+
+/**
+ * Model stores all the data required for recognition
+ *
+ * It contains static data and can be shared across processing
+ * threads.
+ *
+ * @since 26 / 12 / 2022
+ * @constructor Loads model data from the file and returns the model object
+ * @throws IOException if the path provided is invalid
+ */
+actual class Model : Freeable, PointerType, AutoCloseable {
+
+	/**
+	 * Empty constructor for JNA
+	 */
+	constructor()
+
+	/**
+	 * Loads model data from the file and returns the model object
+	 */
+	@Throws(IOException::class)
+	actual constructor(path: String) : super(
+		LibVosk.vosk_model_new(path) ?: throw ModelException(path)
+	)
+
+	/**
+	 * Constructor using a Path, will retrieve absolutePath
+	 *
+	 * @param path to batch model
+	 */
+	@Throws(IOException::class)
+	constructor(path: Path) : this(path.absolutePathString())
+
+	/**
+	 * Constructor using a File, will retrieve absolutePath
+	 *
+	 * @param file to batch model
+	 */
+	@Throws(IOException::class)
+	constructor(file: File) : this(file.absolutePath)
+
+	/**
+	 * Check if a word can be recognized by the model
+	 * @param word: the word
+	 * @returns the word symbol if @param word exists inside the model
+	 * or -1 otherwise.
+	 * Reminding that word symbol 0 is for <epsilon>
+	 */
+	actual fun findWord(word: String): Int =
+		LibVosk.vosk_model_find_word(this, word)
+
+	/**
+	 * Releases the model memory
+	 *
+	 *  The model object is reference-counted so if some recognizer
+	 *  depends on this model, model might still stay alive. When
+	 *  last recognizer is released, model will be released too.
+	 */
+	actual override fun free() {
+		LibVosk.vosk_model_free(this)
+	}
+
+	/**
+	 * @see free
+	 */
+	override fun close() {
+		free()
+	}
+}

kotlin/src/jvmMain/kotlin/org/vosk/Recognizer.kt

@@ -0,0 +1,353 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import com.sun.jna.PointerType
+import org.vosk.exception.RecognizerException
+
+/**
+ * Recognizer object is the main object which processes data.
+ *
+ * Each recognizer usually runs in own thread and takes audio as input.
+ * Once audio is processed recognizer returns JSON object as a string
+ * which represent decoded information - words, confidences, times, n-best lists,
+ * speaker information and so on
+ *
+ * @since 26 / 12 / 2022
+ */
+actual class Recognizer : Freeable, PointerType, AutoCloseable {
+
+	/**
+	 * Empty constructor for JNA
+	 */
+	constructor()
+
+	/**
+	 * Creates the recognizer object
+	 *
+	 *  The recognizers process the speech and return text using shared model data
+	 *  @param model       VoskModel containing static data for recognizer. Model can be
+	 *                     shared across recognizers, even running in different threads.
+	 *  @param sampleRate The sample rate of the audio you going to feed into the recognizer.
+	 *                     Make sure this rate matches the audio content, it is a common
+	 *                     issue causing accuracy problems.
+	 * @throws RecognizerException if a problem occurred
+	 */
+	@Throws(RecognizerException::class)
+	actual constructor(model: Model, sampleRate: Float) :
+			super(LibVosk.vosk_recognizer_new(model, sampleRate) ?: throw RecognizerException())
+
+	/**
+	 * Creates the recognizer object with speaker recognition
+	 *
+	 *  With the speaker recognition mode the recognizer not just recognize
+	 *  text but also return speaker vectors one can use for speaker identification
+	 *
+	 *  @param model       VoskModel containing static data for recognizer. Model can be
+	 *                     shared across recognizers, even running in different threads.
+	 *  @param sampleRate The sample rate of the audio you going to feed into the recognizer.
+	 *                     Make sure this rate matches the audio content, it is a common
+	 *                     issue causing accuracy problems.
+	 *  @param speakerModel speaker model for speaker identification
+	 * @throws RecognizerException if a problem occurred
+	 */
+	@Throws(RecognizerException::class)
+	actual constructor(
+		model: Model,
+		sampleRate: Float,
+		speakerModel: SpeakerModel
+	) : super(
+		LibVosk.vosk_recognizer_new_spk(model, sampleRate, speakerModel)
+			?: throw RecognizerException()
+	)
+
+	/**
+	 * Creates the recognizer object with the phrase list
+	 *
+	 *  Sometimes when you want to improve recognition accuracy and when you don't need
+	 *  to recognize large vocabulary you can specify a list of phrases to recognize. This
+	 *  will improve recognizer speed and accuracy but might return [unk] if user said
+	 *  something different.
+	 *
+	 *  Only recognizers with lookahead models support this type of quick configuration.
+	 *  Precompiled HCLG graph models are not supported.
+	 *
+	 *  @param model       VoskModel containing static data for recognizer. Model can be
+	 *                     shared across recognizers, even running in different threads.
+	 *  @param sampleRate The valuesample rate of the audio you going to feed into the recognizer.
+	 *                     Make sure this rate matches the audio content, it is a common
+	 *                     issue causing accuracy problems.
+	 *  @param grammar The string with the list of phrases to recognize as JSON array of strings,
+	 *                 for example "["one two three four five", "[unk]"]".
+	 *
+	 * @throws RecognizerException if a problem occurred
+	 */
+	@Throws(RecognizerException::class)
+	actual constructor(
+		model: Model,
+		sampleRate: Float,
+		grammar: String
+	) : super(
+		LibVosk.vosk_recognizer_new_grm(model, sampleRate, grammar) ?: throw RecognizerException()
+	)
+
+	/**
+	 * JVM analog of Kotlin extension
+	 * @see [org.vosk.json.Recognizer]
+	 */
+	@Throws(RecognizerException::class)
+	constructor(
+		model: Model,
+		sampleRate: Float,
+		grammar: List<String>
+	) : super(
+		// We need the full qualifer to avoid any build issues
+		@Suppress("RemoveRedundantQualifierName")
+		org.vosk.json.Recognizer(model, sampleRate, grammar).pointer
+	)
+
+	/**
+	 * Adds speaker model to already initialized recognizer
+	 *
+	 * Can add speaker recognition model to already created recognizer. Helps to initialize
+	 * speaker recognition for grammar-based recognizer.
+	 *
+	 * @param speakerModel Speaker recognition model
+	 */
+	actual fun setSpeakerModel(speakerModel: SpeakerModel) {
+		LibVosk.vosk_recognizer_set_spk_model(this, speakerModel)
+	}
+
+	/**
+	 * Reconfigures recognizer to use grammar
+	 *
+	 * @param grammar      Set of phrases in JSON array of strings or "[]" to use default model graph.
+	 *                     See also vosk_recognizer_new_grm
+	 */
+	actual fun setGrammar(grammar: String) {
+		LibVosk.vosk_recognizer_set_grm(this, grammar)
+	}
+
+	/**
+	 * Configures recognizer to output n-best results
+	 *
+	 * <pre>
+	 *   {
+	 *      "alternatives": [
+	 *          { "text": "one two three four five", "confidence": 0.97 },
+	 *          { "text": "one two three for five", "confidence": 0.03 },
+	 *      ]
+	 *   }
+	 * </pre>
+	 *
+	 * @param maxAlternatives - maximum alternatives to return from recognition results
+	 */
+	actual fun setMaxAlternatives(maxAlternatives: Int) {
+		LibVosk.vosk_recognizer_set_max_alternatives(this, maxAlternatives)
+	}
+
+	/**
+	 * Enables words with times in the output
+	 *
+	 * <pre>
+	 *   "result" : [{
+	 *       "conf" : 1.000000,
+	 *       "end" : 1.110000,
+	 *       "start" : 0.870000,
+	 *       "word" : "what"
+	 *     }, {
+	 *       "conf" : 1.000000,
+	 *       "end" : 1.530000,
+	 *       "start" : 1.110000,
+	 *       "word" : "zero"
+	 *     }, {
+	 *       "conf" : 1.000000,
+	 *       "end" : 1.950000,
+	 *       "start" : 1.530000,
+	 *       "word" : "zero"
+	 *     }, {
+	 *       "conf" : 1.000000,
+	 *       "end" : 2.340000,
+	 *       "start" : 1.950000,
+	 *       "word" : "zero"
+	 *     }, {
+	 *       "conf" : 1.000000,
+	 *       "end" : 2.610000,
+	 *       "start" : 2.340000,
+	 *       "word" : "one"
+	 *     }],
+	 * </pre>
+	 *
+	 * @param words - boolean value
+	 */
+	actual fun setOutputWordTimes(words: Boolean) {
+		LibVosk.vosk_recognizer_set_words(this, words)
+	}
+
+	/**
+	 * Like [setOutputWordTimes] return words and confidences in partial results
+	 *
+	 * @param partialWords - boolean value
+	 */
+	actual fun setPartialWords(partialWords: Boolean) {
+		LibVosk.vosk_recognizer_set_partial_words(this, partialWords)
+	}
+
+	/**
+	 * Set NLSML output
+	 * @param nlsml - boolean value
+	 */
+	actual fun setNLSML(nlsml: Boolean) {
+		LibVosk.vosk_recognizer_set_nlsml(this, nlsml)
+	}
+
+	/**
+	 * Accept voice data
+	 *
+	 *  accept and process new chunk of voice data
+	 *
+	 *  @param data Audio data in PCM 16-bit mono format.
+	 *  @param length Length of the audio data.
+	 *  @returns
+	 *       1 - If silence is occurred and you can retrieve a new utterance with result method
+	 *       0 - If decoding continues
+	 *      -1 - If exception occurred
+	 */
+	@Throws(AcceptWaveformException::class)
+	actual fun acceptWaveform(data: ByteArray): Boolean {
+		val result = LibVosk.vosk_recognizer_accept_waveform(this, data, data.size)
+		if (result == -1) throw AcceptWaveformException(data)
+		return result == 1
+	}
+
+	/**
+	 * Same as [acceptWaveform] but the version with the short data for language bindings where you have
+	 *  audio as array of shorts
+	 */
+	@Throws(AcceptWaveformException::class)
+	actual fun acceptWaveform(data: ShortArray): Boolean {
+		val result = LibVosk.vosk_recognizer_accept_waveform_s(this, data, data.size)
+		if (result == -1) throw AcceptWaveformException(data)
+		return result == 1
+	}
+
+	/**
+	 * Same as [acceptWaveform] but the version with the float data for language bindings where you have
+	 *  audio as array of floats
+	 */
+	@Throws(AcceptWaveformException::class)
+	actual fun acceptWaveform(data: FloatArray): Boolean {
+		val result = LibVosk.vosk_recognizer_accept_waveform_f(this, data, data.size)
+		if (result == -1) throw AcceptWaveformException(data)
+		return result == 1
+	}
+
+
+	/**
+	 * Returns speech recognition result
+	 *
+	 * @returns the result in JSON format which contains decoded line, decoded
+	 *          words, times in seconds and confidences. You can parse this result
+	 *          with any json parser
+	 *
+	 * <pre>
+	 *  {
+	 *    "text" : "what zero zero zero one"
+	 *  }
+	 * </pre>
+	 *
+	 * If alternatives enabled it returns result with alternatives, see also [setMaxAlternatives].
+	 *
+	 * If word times enabled returns word time, see also [setOutputWordTimes].
+	 */
+	actual val result: String
+		get() = LibVosk.vosk_recognizer_result(this)
+
+	/**
+	 * Returns partial speech recognition
+	 *
+	 * @returns partial speech recognition text which is not yet finalized.
+	 *          result may change as recognizer process more data.
+	 *
+	 * <pre>
+	 * {
+	 *    "partial" : "cyril one eight zero"
+	 * }
+	 * </pre>
+	 */
+	actual val finalResult: String
+		get() = LibVosk.vosk_recognizer_final_result(this)
+
+	/**
+	 * Returns speech recognition result. Same as result, but doesn't wait for silence
+	 *  You usually call it in the end of the stream to get final bits of audio. It
+	 *  flushes the feature pipeline, so all remaining audio chunks got processed.
+	 *
+	 *  @returns speech result in JSON format.
+	 */
+	actual val partialResult: String
+		get() = LibVosk.vosk_recognizer_partial_result(this)
+
+	/**
+	 * Resets the recognizer
+	 *
+	 *  Resets current results so the recognition can continue from scratch
+	 */
+	actual fun reset() {
+		LibVosk.vosk_recognizer_reset(this)
+	}
+
+	/**
+	 * Releases recognizer object
+	 *
+	 *  Underlying model is also unreferenced and if needed released
+	 */
+	actual override fun free() {
+		LibVosk.vosk_recognizer_free(this)
+	}
+
+	/**
+	 * @see free
+	 */
+	override fun close() {
+		free()
+	}
+
+	/**
+	 * Set endpointer scaling factor
+	 *
+	 * @param mode Endpointer mode
+	 **/
+	actual fun setEndPointerMode(mode: EndPointerMode) {
+		LibVosk.vosk_recognizer_set_endpointer_mode(this, mode.ordinal)
+	}
+
+	/**
+	 * Set endpointer delays
+	 *
+	 * @param tStartMax timeout for stopping recognition in case of initial silence (usually around 5.0)
+	 * @param tEnd      timeout for stopping recognition in milliseconds after we recognized something (usually around 0.5 - 1.0)
+	 * @param tMax      timeout for forcing utterance end in milliseconds (usually around 20-30)
+	 **/
+	actual fun setEndPointerDelays(
+		tStartMax: Float,
+		tEnd: Float,
+		tMax: Float
+	) {
+		LibVosk.vosk_recognizer_set_endpointer_delays(this, tStartMax, tEnd, tMax)
+	}
+}

kotlin/src/jvmMain/kotlin/org/vosk/SpeakerModel.kt

@@ -0,0 +1,84 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import com.sun.jna.PointerType
+import org.vosk.exception.ModelException
+import java.io.File
+import java.nio.file.Path
+import kotlin.io.path.absolutePathString
+
+/**
+ * Speaker model is the same as model but contains the data
+ * for speaker identification.
+ *
+ * @since 26 / 12 / 2022
+ * @constructor Loads speaker model data from the file and returns the model object
+ * @throws ModelException if the path provided is invalid
+ */
+actual class SpeakerModel : Freeable, PointerType, AutoCloseable {
+
+	/**
+	 * Empty constructor for JNA
+	 */
+	constructor()
+
+	/**
+	 * Loads speaker model data from the file and returns the model object
+	 *
+	 * @param path the path of the model on the filesystem
+	 */
+	@Throws(ModelException::class)
+	actual constructor(path: String) : super(
+		LibVosk.vosk_spk_model_new(path) ?: throw ModelException(path)
+	)
+
+	/**
+	 * Constructor using a Path, will retrieve absolutePath
+	 *
+	 * @param path to batch model
+	 */
+	@Throws(ModelException::class)
+	constructor(path: Path) : this(path.absolutePathString())
+
+	/**
+	 * Constructor using a File, will retrieve absolutePath
+	 *
+	 * @param file to batch model
+	 */
+	@Throws(ModelException::class)
+	constructor(file: File) : this(file.absolutePath)
+
+	/**
+	 * Releases the model memory
+	 *
+	 * The model object is reference-counted so if some recognizer
+	 * depends on this model, model might still stay alive. When
+	 * last recognizer is released, model will be released too.
+	 */
+	actual override fun free() {
+		LibVosk.vosk_spk_model_free(this)
+	}
+
+	/**
+	 * @see free
+	 */
+	override fun close() {
+		free()
+	}
+
+}

kotlin/src/jvmMain/kotlin/org/vosk/TextProcessor.kt

@@ -0,0 +1,51 @@
+/*
+ * Copyright 2024 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import com.sun.jna.PointerType
+
+/**
+ * Inverse text normalization
+ *
+ * @since 2024/06/19
+ */
+actual class TextProcessor :
+	Freeable, PointerType, AutoCloseable {
+
+	/**
+	 * Create text processor
+	 */
+	actual constructor(tagger: Char, verbalizer: Char) :
+			super(LibVosk.vosk_text_processor_new(tagger, verbalizer))
+
+
+	/** Release text processor */
+	actual override fun free() {
+		LibVosk.vosk_text_processor_free(this)
+	}
+
+	/** Convert string */
+	actual fun itn(input: Char): Char =
+		LibVosk.vosk_text_processor_itn(this, input)
+
+	/**
+	 * @see free
+	 */
+	override fun close() {
+		free()
+	}
+}

kotlin/src/jvmMain/kotlin/org/vosk/Vosk.kt

@@ -0,0 +1,55 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+
+/**
+ * 26 / 12 / 2022
+ *
+ * Control overarching features of libvosk.
+ */
+actual object Vosk {
+	/**
+	 * Set log level for Kaldi messages
+	 *
+	 *  @param logLevel the level
+	 */
+	@JvmStatic
+	actual fun setLogLevel(logLevel: LogLevel) {
+		LibVosk.vosk_set_log_level(logLevel.value)
+	}
+
+	/**
+	 *  Init, automatically select a CUDA device and allow multithreading.
+	 *  Must be called once from the main thread.
+	 *  Has no effect if HAVE_CUDA flag is not set.
+	 */
+	@JvmStatic
+	actual fun gpuInit() {
+		LibVosk.vosk_gpu_init()
+	}
+
+	/**
+	 *  Init CUDA device in a multi-threaded environment.
+	 *  Must be called for each thread.
+	 *  Has no effect if HAVE_CUDA flag is not set.
+	 */
+	@JvmStatic
+	actual fun gpuThreadInit() {
+		LibVosk.vosk_gpu_thread_init()
+	}
+}

kotlin/src/jvmMain/kotlin/org/vosk/WaveformExt.kt

@@ -0,0 +1,55 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import kotlinx.coroutines.Dispatchers
+import kotlinx.coroutines.flow.Flow
+import kotlinx.coroutines.flow.flowOn
+import kotlinx.coroutines.flow.map
+import org.vosk.json.WaveformResult
+import org.vosk.json.finalResultAsJson
+import org.vosk.json.partialResultAsJson
+import org.vosk.json.resultAsJson
+
+/**
+ * Feed an [Flow] of [ByteArray] into a [Recognizer].
+ *
+ * The returned flow will emit an [WaveformResult] for each result parsed.
+ *
+ * This expects a null terminator to signify the end of a stream.
+ *
+ * This will not close the [recognizer] at the end of reading.
+ *
+ * Any exceptions will be fed into the flow,
+ *  and should be collected via [kotlinx.coroutines.flow.catch].
+ *
+ * Flows on [Dispatchers.IO] to prevent blocking the main thread.
+ *
+ * @see [Recognizer.acceptWaveform]
+ */
+fun Flow<ByteArray?>.feed(recognizer: Recognizer): Flow<WaveformResult> =
+	map {
+		if (it != null) {
+			if (recognizer.acceptWaveform(it)) {
+				WaveformResult.Result(recognizer.resultAsJson())
+			} else {
+				WaveformResult.PartialResult(recognizer.partialResultAsJson())
+			}
+		} else {
+			WaveformResult.FinalResult(recognizer.finalResultAsJson())
+		}
+	}.flowOn(Dispatchers.IO)

kotlin/src/jvmMain/kotlin/org/vosk/exception/IOException.kt

@@ -0,0 +1,22 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk.exception
+
+/**
+ * Analog of [java.io.IOException]
+ */
+actual typealias IOException = java.io.IOException

kotlin/src/jvmTest/kotlin/DecoderTest.kt

@@ -0,0 +1,212 @@
+/*
+ * Copyright 2023 Alpha Cephei Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+import kotlinx.coroutines.*
+import kotlinx.coroutines.flow.*
+import kotlinx.coroutines.test.runTest
+import org.vosk.*
+import java.io.BufferedInputStream
+import java.io.FileInputStream
+import java.io.IOException
+import java.nio.ByteBuffer
+import java.nio.ByteOrder
+import javax.sound.sampled.AudioSystem
+import javax.sound.sampled.UnsupportedAudioFileException
+import kotlin.test.Test
+
+class DecoderTest {
+	val modelPath = System.getenv("MODEL")
+	val testFile = System.getenv("AUDIO")
+
+	init {
+		System.load(System.getenv("LIBRARY"))
+		Vosk.setLogLevel(LogLevel.DEBUG)
+	}
+
+	@Test
+	fun grammarList() {
+		Model(modelPath).use { model ->
+			Recognizer(model, 16000f, listOf("one")).apply {
+				setMaxAlternatives(10)
+				setOutputWordTimes(true)
+				setPartialWords(true)
+			}
+		}
+	}
+
+	@Test
+	@Throws(IOException::class, UnsupportedAudioFileException::class)
+	fun decoderTest() {
+		Model(modelPath).use { model ->
+			AudioSystem.getAudioInputStream(BufferedInputStream(FileInputStream(testFile)))
+				.use { ais ->
+					Recognizer(model, 16000f).apply {
+						setMaxAlternatives(10)
+						setOutputWordTimes(true)
+						setPartialWords(true)
+					}.use { recognizer ->
+						val b = ByteArray(4096)
+						while (ais.read(b) >= 0) {
+							if (recognizer.acceptWaveform(b)) {
+								println(recognizer.result)
+							} else {
+								println(recognizer.partialResult)
+							}
+						}
+						println(recognizer.finalResult)
+					}
+				}
+		}
+	}
+
+	@OptIn(ExperimentalCoroutinesApi::class)
+	@Test
+	@Throws(IOException::class, UnsupportedAudioFileException::class)
+	fun decoderTestFlow() = runTest {
+		Model(modelPath).use { model ->
+			Recognizer(model, 16000f).apply {
+				setMaxAlternatives(10)
+				setOutputWordTimes(true)
+				setPartialWords(true)
+			}.use { recognizer ->
+				AudioSystem.getAudioInputStream(BufferedInputStream(FileInputStream(testFile)))
+					.use { ais ->
+						flow {
+							val b = ByteArray(4096)
+							while (ais.read(b) >= 0) {
+								emit(b)
+							}
+							emit(null)
+						}.flowOn(Dispatchers.IO)
+							.feed(recognizer)
+							.collect {
+								println(it)
+							}
+					}
+			}
+		}
+	}
+
+	/**
+	 * This test aims to simulate the situation for Dicio,
+	 *  In which we can receive the audio input stream before the recognizer is setup.
+	 *
+	 *  It is recommended to use a large model on desktop to properly see how long it takes to load up.
+	 */
+	@Test
+	@Throws(IOException::class, UnsupportedAudioFileException::class)
+	fun decoderTestFlowBuffered() {
+		// Scope to buffer into
+		val scope = CoroutineScope(Dispatchers.IO)
+
+		AudioSystem.getAudioInputStream(BufferedInputStream(FileInputStream(testFile))).use { ais ->
+			val byteFlow = flow {
+				val b = ByteArray(4096)
+				while (ais.read(b) >= 0) {
+					emit(b)
+				}
+				emit(null)
+			}.flowOn(Dispatchers.IO)
+				.shareIn(scope, SharingStarted.Eagerly, 100)
+
+			// Tell us the current buffer size
+			println("Buffered size:" + byteFlow.replayCache.size)
+
+			var startTime = System.currentTimeMillis()
+
+			Model(modelPath).use { model ->
+				var resultTime = System.currentTimeMillis() - startTime
+				println("Model initialized in: $resultTime")
+				startTime = System.currentTimeMillis()
+
+				Recognizer(model, 16000f).apply {
+					setMaxAlternatives(10)
+					setOutputWordTimes(true)
+					setPartialWords(true)
+				}.use { recognizer ->
+					resultTime = System.currentTimeMillis() - startTime
+					println("Recognizer initialized in: $resultTime")
+					println("Buffered size:" + byteFlow.replayCache.size)
+
+					runBlocking {
+						byteFlow
+							.feed(recognizer)
+							.take(byteFlow.replayCache.size)
+							.collect {
+								println(it)
+							}
+					}
+				}
+			}
+		}
+	}
+
+	@Test
+	@Throws(IOException::class, UnsupportedAudioFileException::class)
+	fun decoderTestShort() {
+		Model(modelPath).use { model ->
+			AudioSystem.getAudioInputStream(BufferedInputStream(FileInputStream(testFile)))
+				.use { ais ->
+					Recognizer(model, 16000f).use { recognizer ->
+						val b = ByteArray(4096)
+						val s = ShortArray(2048)
+						while (ais.read(b) >= 0) {
+							ByteBuffer.wrap(b).order(ByteOrder.LITTLE_ENDIAN).asShortBuffer().get(s)
+							if (recognizer.acceptWaveform(s)) {
+								println(recognizer.result)
+							} else {
+								println(recognizer.partialResult)
+							}
+						}
+						println(recognizer.finalResult)
+					}
+				}
+		}
+	}
+
+	@Test
+	@Throws(IOException::class, UnsupportedAudioFileException::class)
+	fun decoderTestGrammar() {
+		Model(modelPath).use { model ->
+			AudioSystem.getAudioInputStream(BufferedInputStream(FileInputStream(testFile)))
+				.use { ais ->
+					Recognizer(
+						model, 16000f, "[\"one two three four five six seven eight nine zero oh\"]"
+					).use { recognizer ->
+						val b = ByteArray(4096)
+						while (ais.read(b) >= 0) {
+							if (recognizer.acceptWaveform(b)) {
+								println(recognizer.result)
+							} else {
+								println(recognizer.partialResult)
+							}
+						}
+						println(recognizer.finalResult)
+					}
+				}
+		}
+	}
+
+	@Test
+	fun decoderTestException() {
+		try {
+			val model = Model("model_missing")
+			assert(false)
+		} catch (e: IOException) {
+			assert(true)
+		}
+	}
+}

kotlin/src/nativeInterop/cinterop/libvosk.def

@@ -0,0 +1,25 @@
+headers = libvosk/vosk_api.h
+package = libvosk
+
+noStringConversion = \
+	vosk_batch_recognizer_accept_waveform \
+	vosk_recognizer_accept_waveform
+
+compilerOpts.linux = \
+	-I/usr/include/libvosk/ \
+	-I/usr/local/include/libvosk/ \
+	-I/usr/include/ \
+	-I/usr/local/include/
+
+compilerOpts.linux_x64 = \
+	-I/usr/lib64/libvosk/ \
+    -I/usr/local/lib64/libvosk/
+
+linkerOpts.linux = \
+	-L/usr/lib/ \
+	-L/usr/local/lib/ \
+	-llibvosk
+
+linkerOpts.linux_x64 = \
+	-L/usr/lib64/ \
+	-L/usr/local/lib64/

kotlin/src/nativeMain/kotlin/org/vosk/BatchModel.kt

@@ -0,0 +1,50 @@
+/*
+ * Copyright 2020 Alpha Cephei Inc. & Doomsdayrs
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import cnames.structs.VoskBatchModel
+import kotlinx.cinterop.CPointer
+import libvosk.vosk_batch_model_free
+import libvosk.vosk_batch_model_new
+import libvosk.vosk_batch_model_wait
+
+/**
+ * Batch model object
+ *
+ * 26 / 12 / 2022
+ */
+actual class BatchModel(val pointer: CPointer<VoskBatchModel>) : Freeable {
+	/**
+	 * Creates the batch recognizer object
+	 */
+	@Throws(IOException::class)
+	actual constructor(path: String) : this(vosk_batch_model_new(path) ?: throw ioException(path))
+
+	/**
+	 *  Releases batch model object
+	 */
+	actual override fun free() {
+		vosk_batch_model_free(pointer)
+	}
+
+	/**
+	 * Wait for the processing
+	 */
+	actual fun await() {
+		vosk_batch_model_wait(pointer)
+	}
+}

kotlin/src/nativeMain/kotlin/org/vosk/BatchRecognizer.kt

@@ -0,0 +1,86 @@
+/*
+ * Copyright 2020 Alpha Cephei Inc. & Doomsdayrs
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import cnames.structs.VoskBatchRecognizer
+import kotlinx.cinterop.CPointer
+import kotlinx.cinterop.toCValues
+import kotlinx.cinterop.toKString
+import libvosk.*
+
+/**
+ * Batch recognizer object
+ *
+ * 26 / 12 / 2022
+ */
+actual class BatchRecognizer(val pointer: CPointer<VoskBatchRecognizer>) : Freeable {
+	/**
+	 * Creates batch recognizer object
+	 */
+	actual constructor(
+		model: BatchModel,
+		sampleRate: Float
+	) : this(vosk_batch_recognizer_new(model.pointer, sampleRate)!!)
+
+	/**
+	 * Releases batch recognizer object
+	 */
+	actual override fun free() {
+		vosk_batch_recognizer_free(pointer)
+	}
+
+	/**
+	 * Accept batch voice data
+	 */
+	actual fun acceptWaveform(data: ByteArray) {
+		vosk_batch_recognizer_accept_waveform(pointer, data.toCValues(), data.size)
+	}
+
+	/**
+	 * Set NLSML output
+	 * @param nlsml - boolean value
+	 */
+	actual fun setNLSML(nlsml: Boolean) {
+		vosk_batch_recognizer_set_nlsml(pointer, nlsml.toInt())
+	}
+
+	/**
+	 *  Closes the stream
+	 */
+	actual fun finishStream() {
+		vosk_batch_recognizer_finish_stream(pointer)
+	}
+
+	/**
+	 * Return results
+	 */
+	actual val frontResult: String
+		get() = vosk_batch_recognizer_front_result(pointer)!!.toKString()
+
+	/**
+	 * Release and free first retrieved result
+	 */
+	actual fun pop() {
+		vosk_batch_recognizer_pop(pointer)
+	}
+
+	/**
+	 * Get amount of pending chunks for more intelligent waiting
+	 */
+	actual val pendingChunks: Int
+		get() = vosk_batch_recognizer_get_pending_chunks(pointer)
+}

kotlin/src/nativeMain/kotlin/org/vosk/FreeableExt.kt

@@ -0,0 +1,23 @@
+/*
+ * Copyright 2020 Alpha Cephei Inc. & Doomsdayrs
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+/**
+ * Use a [Freeable], then immediately free it and return any values.
+ */
+fun <T : Freeable, R> T.use(block: (T) -> R): R =
+	block(this).also { free() }

kotlin/src/nativeMain/kotlin/org/vosk/IOException.kt

@@ -0,0 +1,3 @@
+package org.vosk
+
+actual class IOException actual constructor(message: String?) : Exception()

kotlin/src/nativeMain/kotlin/org/vosk/Model.kt

@@ -0,0 +1,63 @@
+/*
+ * Copyright 2020 Alpha Cephei Inc. & Doomsdayrs
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import cnames.structs.VoskModel
+import kotlinx.cinterop.CPointer
+import libvosk.vosk_model_find_word
+import libvosk.vosk_model_free
+import libvosk.vosk_model_new
+
+/**
+ * Model stores all the data required for recognition
+ *
+ * It contains static data and can be shared across processing
+ * threads.
+ *
+ * 26 / 12 / 2022
+ */
+actual class Model(val pointer: CPointer<VoskModel>) : Freeable {
+	/**
+	 * Loads model data from the file and returns the model object
+	 *
+	 * @param path: the path of the model on the filesystem
+	 * @returns model object or NULL if problem occured
+	 */
+	@Throws(IOException::class)
+	actual constructor(path: String) : this(vosk_model_new(path) ?: throw ioException(path))
+
+	/**
+	 * Check if a word can be recognized by the model
+	 * @param word: the word
+	 * @returns the word symbol if @param word exists inside the model
+	 * or -1 otherwise.
+	 * Reminding that word symbol 0 is for <epsilon>
+	 */
+	actual fun findWord(word: String): Int =
+		vosk_model_find_word(pointer, word)
+
+	/**
+	 * Releases the model memory
+	 *
+	 *  The model object is reference-counted so if some recognizer
+	 *  depends on this model, model might still stay alive. When
+	 *  last recognizer is released, model will be released too.
+	 */
+	actual override fun free() {
+		vosk_model_free(pointer)
+	}
+}

kotlin/src/nativeMain/kotlin/org/vosk/Recognizer.kt

@@ -0,0 +1,310 @@
+/*
+ * Copyright 2020 Alpha Cephei Inc. & Doomsdayrs
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import kotlinx.cinterop.CPointer
+import kotlinx.cinterop.toCValues
+import kotlinx.cinterop.toKString
+import libvosk.*
+
+/**
+ * Recognizer object is the main object which processes data.
+ *
+ * Each recognizer usually runs in own thread and takes audio as input.
+ * Once audio is processed recognizer returns JSON object as a string
+ * which represent decoded information - words, confidences, times, n-best lists,
+ * speaker information and so on
+ *
+ * 26 / 12 / 2022
+ */
+actual class Recognizer(val pointer: CPointer<VoskRecognizer>) : Freeable {
+	/**
+	 * Creates the recognizer object
+	 *
+	 *  The recognizers process the speech and return text using shared model data
+	 *  @param model       VoskModel containing static data for recognizer. Model can be
+	 *                     shared across recognizers, even running in different threads.
+	 *  @param sampleRate The sample rate of the audio you going to feed into the recognizer.
+	 *                     Make sure this rate matches the audio content, it is a common
+	 *                     issue causing accuracy problems.
+	 *  @returns recognizer object or NULL if problem occured
+	 */
+	actual constructor(
+		model: Model,
+		sampleRate: Float
+	) : this(
+		vosk_recognizer_new(
+			model.pointer,
+			sampleRate
+		)!!
+	)
+
+	/**
+	 * Creates the recognizer object with speaker recognition
+	 *
+	 *  With the speaker recognition mode the recognizer not just recognize
+	 *  text but also return speaker vectors one can use for speaker identification
+	 *
+	 *  @param model       VoskModel containing static data for recognizer. Model can be
+	 *                     shared across recognizers, even running in different threads.
+	 *  @param sampleRate The sample rate of the audio you going to feed into the recognizer.
+	 *                     Make sure this rate matches the audio content, it is a common
+	 *                     issue causing accuracy problems.
+	 *  @param speakerModel speaker model for speaker identification
+	 *  @returns recognizer object or NULL if problem occured
+	 */
+	actual constructor(
+		model: Model,
+		sampleRate: Float,
+		speakerModel: SpeakerModel
+	) : this(
+		vosk_recognizer_new_spk(
+			model.pointer,
+			sampleRate,
+			speakerModel.pointer
+		)!!
+	)
+
+	/**
+	 * Creates the recognizer object with the phrase list
+	 *
+	 *  Sometimes when you want to improve recognition accuracy and when you don't need
+	 *  to recognize large vocabulary you can specify a list of phrases to recognize. This
+	 *  will improve recognizer speed and accuracy but might return [unk] if user said
+	 *  something different.
+	 *
+	 *  Only recognizers with lookahead models support this type of quick configuration.
+	 *  Precompiled HCLG graph models are not supported.
+	 *
+	 *  @param model       VoskModel containing static data for recognizer. Model can be
+	 *                     shared across recognizers, even running in different threads.
+	 *  @param sampleRate The sample rate of the audio you going to feed into the recognizer.
+	 *                     Make sure this rate matches the audio content, it is a common
+	 *                     issue causing accuracy problems.
+	 *  @param grammar The string with the list of phrases to recognize as JSON array of strings,
+	 *                 for example "["one two three four five", "[unk]"]".
+	 *
+	 *  @returns recognizer object or NULL if problem occured
+	 */
+	actual constructor(
+		model: Model,
+		sampleRate: Float,
+		grammar: String
+	) : this(
+		vosk_recognizer_new_grm(
+			model.pointer,
+			sampleRate,
+			grammar
+		)!!
+	)
+
+	/**
+	 * Adds speaker model to already initialized recognizer
+	 *
+	 * Can add speaker recognition model to already created recognizer. Helps to initialize
+	 * speaker recognition for grammar-based recognizer.
+	 *
+	 * @param speakerModel Speaker recognition model
+	 */
+	actual fun setSpeakerModel(speakerModel: SpeakerModel) {
+		vosk_recognizer_set_spk_model(pointer, speakerModel.pointer)
+	}
+
+	/**
+	 * Reconfigures recognizer to use grammar
+	 *
+	 * @param grammar      Set of phrases in JSON array of strings or "[]" to use default model graph.
+	 *                     See also vosk_recognizer_new_grm
+	 */
+	actual fun setGrammar(grammar: String) {
+		vosk_recognizer_set_grm(pointer, grammar)
+	}
+
+	/**
+	 * Configures recognizer to output n-best results
+	 *
+	 * <pre>
+	 *   {
+	 *      "alternatives": [
+	 *          { "text": "one two three four five", "confidence": 0.97 },
+	 *          { "text": "one two three for five", "confidence": 0.03 },
+	 *      ]
+	 *   }
+	 * </pre>
+	 *
+	 * @param maxAlternatives - maximum alternatives to return from recognition results
+	 */
+	actual fun setMaxAlternatives(maxAlternatives: Int) {
+		vosk_recognizer_set_max_alternatives(pointer, maxAlternatives)
+	}
+
+	/**
+	 * Enables words with times in the output
+	 *
+	 * <pre>
+	 *   "result" : [{
+	 *       "conf" : 1.000000,
+	 *       "end" : 1.110000,
+	 *       "start" : 0.870000,
+	 *       "word" : "what"
+	 *     }, {
+	 *       "conf" : 1.000000,
+	 *       "end" : 1.530000,
+	 *       "start" : 1.110000,
+	 *       "word" : "zero"
+	 *     }, {
+	 *       "conf" : 1.000000,
+	 *       "end" : 1.950000,
+	 *       "start" : 1.530000,
+	 *       "word" : "zero"
+	 *     }, {
+	 *       "conf" : 1.000000,
+	 *       "end" : 2.340000,
+	 *       "start" : 1.950000,
+	 *       "word" : "zero"
+	 *     }, {
+	 *       "conf" : 1.000000,
+	 *       "end" : 2.610000,
+	 *       "start" : 2.340000,
+	 *       "word" : "one"
+	 *     }],
+	 * </pre>
+	 *
+	 * @param words - boolean value
+	 */
+	actual fun setWords(words: Boolean) {
+		vosk_recognizer_set_words(pointer, words.toInt())
+	}
+
+	/**
+	 * Like above return words and confidences in partial results
+	 *
+	 * @param partialWords - boolean value
+	 */
+	actual fun setPartialWords(partialWords: Boolean) {
+		vosk_recognizer_set_partial_words(pointer, partialWords.toInt())
+	}
+
+	/**
+	 * Set NLSML output
+	 * @param nlsml - boolean value
+	 */
+	actual fun setNLSML(nlsml: Boolean) {
+		vosk_recognizer_set_nlsml(pointer, nlsml.toInt())
+	}
+
+	/**
+	 * Accept voice data
+	 *
+	 *  accept and process new chunk of voice data
+	 *
+	 *  @param data - audio data in PCM 16-bit mono format
+	 *  @param length - length of the audio data
+	 *  @returns 1 if silence is occured and you can retrieve a new utterance with result method
+	 *           0 if decoding continues
+	 *           -1 if exception occured
+	 */
+	@Throws(AcceptWaveformException::class)
+	actual fun acceptWaveform(data: ByteArray): Boolean {
+		val result = vosk_recognizer_accept_waveform(pointer, data.toCValues(), data.size)
+		if (result == -1) throw AcceptWaveformException(data)
+		return result == 1
+	}
+
+	/**
+	 * Same as above but the version with the short data for language bindings where you have
+	 *  audio as array of shorts
+	 */
+	@Throws(AcceptWaveformException::class)
+	actual fun acceptWaveform(data: ShortArray): Boolean {
+		val result = vosk_recognizer_accept_waveform_s(pointer, data.toCValues(), data.size)
+		if (result == -1) throw AcceptWaveformException(data)
+		return result == 1
+	}
+
+	/**
+	 * Same as above but the version with the float data for language bindings where you have
+	 *  audio as array of floats
+	 */
+	@Throws(AcceptWaveformException::class)
+	actual fun acceptWaveform(data: FloatArray): Boolean {
+		val result = vosk_recognizer_accept_waveform_f(pointer, data.toCValues(), data.size)
+		if (result == -1) throw AcceptWaveformException(data)
+		return result == 1
+	}
+
+	/**
+	 * Returns speech recognition result
+	 *
+	 * @returns the result in JSON format which contains decoded line, decoded
+	 *          words, times in seconds and confidences. You can parse this result
+	 *          with any json parser
+	 *
+	 * <pre>
+	 *  {
+	 *    "text" : "what zero zero zero one"
+	 *  }
+	 * </pre>
+	 *
+	 * If alternatives enabled it returns result with alternatives, see also vosk_recognizer_set_max_alternatives().
+	 *
+	 * If word times enabled returns word time, see also vosk_recognizer_set_word_times().
+	 */
+	actual val result: String
+		get() = vosk_recognizer_result(pointer)!!.toKString()
+
+	/**
+	 * Returns partial speech recognition
+	 *
+	 * @returns partial speech recognition text which is not yet finalized.
+	 *          result may change as recognizer process more data.
+	 *
+	 * <pre>
+	 * {
+	 *    "partial" : "cyril one eight zero"
+	 * }
+	 * </pre>
+	 */
+	actual val finalResult: String
+		get() = vosk_recognizer_result(pointer)!!.toKString()
+
+	/**
+	 * Returns speech recognition result. Same as result, but doesn't wait for silence
+	 *  You usually call it in the end of the stream to get final bits of audio. It
+	 *  flushes the feature pipeline, so all remaining audio chunks got processed.
+	 *
+	 *  @returns speech result in JSON format.
+	 */
+	actual val partialResult: String
+		get() = vosk_recognizer_partial_result(pointer)!!.toKString()
+
+	/**
+	 * Resets the recognizer
+	 *
+	 *  Resets current results so the recognition can continue from scratch */
+	actual fun reset() {
+		vosk_recognizer_reset(pointer)
+	}
+
+	/**
+	 * Releases recognizer object
+	 *
+	 *  Underlying model is also unreferenced and if needed released */
+	actual override fun free() {
+		vosk_recognizer_free(pointer)
+	}
+}

kotlin/src/nativeMain/kotlin/org/vosk/SpeakerModel.kt

@@ -0,0 +1,50 @@
+/*
+ * Copyright 2020 Alpha Cephei Inc. & Doomsdayrs
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import cnames.structs.VoskSpkModel
+import kotlinx.cinterop.CPointer
+import libvosk.vosk_spk_model_free
+import libvosk.vosk_spk_model_new
+
+/**
+ * Speaker model is the same as model but contains the data
+ * for speaker identification.
+ *
+ * 26 / 12 / 2022
+ */
+actual class SpeakerModel(val pointer: CPointer<VoskSpkModel>) : Freeable {
+	/**
+	 * Loads speaker model data from the file and returns the model object
+	 *
+	 * @param path: the path of the model on the filesystem
+	 * @returns model object or NULL if problem occurred
+	 */
+	@Throws(IOException::class)
+	actual constructor(path: String) : this(vosk_spk_model_new(path) ?: throw ioException(path))
+
+	/**
+	 * Releases the model memory
+	 *
+	 * The model object is reference-counted so if some recognizer
+	 * depends on this model, model might still stay alive. When
+	 * last recognizer is released, model will be released too.
+	 */
+	actual override fun free() {
+		vosk_spk_model_free(pointer)
+	}
+}

kotlin/src/nativeMain/kotlin/org/vosk/Utils.kt

@@ -0,0 +1,26 @@
+/*
+ * Copyright 2020 Alpha Cephei Inc. & Doomsdayrs
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+/*
+ * 26 / 12 / 2022
+ */
+
+/**
+ * Converts a boolean to an int
+ */
+internal inline fun Boolean.toInt() = if (this) 1 else 0

kotlin/src/nativeMain/kotlin/org/vosk/Vosk.kt

@@ -0,0 +1,52 @@
+/*
+ * Copyright 2020 Alpha Cephei Inc. & Doomsdayrs
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.vosk
+
+import libvosk.vosk_set_log_level
+import libvosk.vosk_gpu_init
+import libvosk.vosk_gpu_thread_init
+
+/**
+ * 26 / 12 / 2022
+ */
+actual object Vosk {
+	/** Set log level for Kaldi messages
+	 *
+	 *  @param logLevel the level
+	 */
+	actual fun setLogLevel(logLevel: LogLevel) {
+		vosk_set_log_level(logLevel.value)
+	}
+
+	/**
+	 *  Init, automatically select a CUDA device and allow multithreading.
+	 *  Must be called once from the main thread.
+	 *  Has no effect if HAVE_CUDA flag is not set.
+	 */
+	actual fun gpuInit() {
+		vosk_gpu_init()
+	}
+
+	/**
+	 *  Init CUDA device in a multi-threaded environment.
+	 *  Must be called for each thread.
+	 *  Has no effect if HAVE_CUDA flag is not set.
+	 */
+	actual fun gpuThreadInit() {
+		vosk_gpu_thread_init()
+	}
+}

nodejs/index.js

@@ -69,16 +69,18 @@ const vosk_recognizer_ptr = ref.refType(vosk_recognizer);
 
 let soname;
 if (os.platform() == 'win32') {
-    // Update path to load dependent dlls
-    let currentPath = process.env.Path;
-    let dllDirectory = path.resolve(path.join(__dirname, "lib", "win-x86_64"));
-    process.env.Path = dllDirectory + path.delimiter + currentPath;
+	// Update path to load dependent dlls
+	let currentPath = process.env.Path;
+	let dllDirectory = path.resolve(path.join(__dirname, 'lib', 'win-x86_64'));
+	process.env.Path = dllDirectory + path.delimiter + currentPath;
 
-    soname = path.join(__dirname, "lib", "win-x86_64", "libvosk.dll")
+	soname = path.join(__dirname, 'lib', 'win-x86_64', 'libvosk.dll');
 } else if (os.platform() == 'darwin') {
-    soname = path.join(__dirname, "lib", "osx-universal", "libvosk.dylib")
+	soname = path.join(__dirname, 'lib', 'osx-universal', 'libvosk.dylib');
+} else if (os.platform() == 'linux' && os.arch() == 'arm64') {
+	soname = path.join(__dirname, 'lib', 'linux-arm64', 'libvosk.so');
 } else {
-    soname = path.join(__dirname, "lib", "linux-x86_64", "libvosk.so")
+	soname = path.join(__dirname, 'lib', 'linux-x86_64', 'libvosk.so');
 }
 
 const libvosk = ffi.Library(soname, {
@@ -128,6 +130,9 @@ class Model {
          * @type {unknown}
          */
         this.handle = libvosk.vosk_model_new(modelPath);
+        if (!this.handle) {
+            throw new Error('Failed to create a model.');
+        }
     }
 
     /**
@@ -161,6 +166,9 @@ class SpeakerModel {
          * @type {unknown}
          */
         this.handle = libvosk.vosk_spk_model_new(modelPath);
+        if (!this.handle) {
+            throw new Error('Failed to create a speaker model.');
+        }
     }
 
     /**
@@ -235,6 +243,10 @@ class Recognizer {
             : hasOwnProperty(param, 'grammar')
                 ? libvosk.vosk_recognizer_new_grm(model.handle, sampleRate, JSON.stringify(param.grammar))
                 : libvosk.vosk_recognizer_new(model.handle, sampleRate);
+
+        if (!this.handle) {
+            throw new Error('Failed to create a recognizer.');
+        }
     }
 
     /**

nodejs/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vosk",
-  "version": "0.3.45",
+  "version": "0.3.75",
   "description": "Node binding for continuous offline voice recoginition with Vosk library.",
   "repository": {
     "type": "git",

python/example/test_ep.py

@@ -0,0 +1,55 @@
+#!/usr/bin/env python3
+
+import wave
+import sys
+
+from vosk import Model, KaldiRecognizer, SetLogLevel, EndpointerMode
+
+# You can set log level to -1 to disable debug messages
+SetLogLevel(0)
+
+wf = wave.open(sys.argv[1], "rb")
+if wf.getnchannels() != 1 or wf.getsampwidth() != 2 or wf.getcomptype() != "NONE":
+    print("Audio file must be WAV format mono PCM.")
+    sys.exit(1)
+
+model = Model(lang="en-us")
+
+# You can also init model by name or with a folder path
+# model = Model(model_name="vosk-model-en-us-0.21")
+# model = Model("models/en")
+
+rec = KaldiRecognizer(model, wf.getframerate())
+rec.SetWords(True)
+rec.SetPartialWords(True)
+rec.SetEndpointerMode(EndpointerMode.VERY_LONG)
+
+while True:
+    data = wf.readframes(4000)
+    if len(data) == 0:
+        break
+    if rec.AcceptWaveform(data):
+        print(rec.Result())
+    else:
+        print(rec.PartialResult())
+
+print(rec.FinalResult())
+
+
+wf = wave.open(sys.argv[1], "rb")
+if wf.getnchannels() != 1 or wf.getsampwidth() != 2 or wf.getcomptype() != "NONE":
+    print("Audio file must be WAV format mono PCM.")
+    sys.exit(1)
+
+rec.SetEndpointerDelays(0.5, 0.3, 10.0)
+
+while True:
+    data = wf.readframes(4000)
+    if len(data) == 0:
+        break
+    if rec.AcceptWaveform(data):
+        print(rec.Result())
+    else:
+        print(rec.PartialResult())
+
+print(rec.FinalResult())

python/example/test_gradio.py

@@ -1,40 +1,39 @@
-#!/usr/bin/env python3
-
-import json
-import gradio as gr
-
-from vosk import KaldiRecognizer, Model
-
-model = Model(lang="en-us")
-
-def transcribe(data, state):
-    sample_rate, audio_data = data
-    audio_data = (audio_data >> 16).astype("int16").tobytes()
-
-    if state is None:
-        rec = KaldiRecognizer(model, sample_rate)
-        result = []
-    else:
-        rec, result = state
-
-    if rec.AcceptWaveform(audio_data):
-        text_result = json.loads(rec.Result())["text"]
-        if text_result != "":
-            result.append(text_result)
-        partial_result = ""
-    else:
-        partial_result = json.loads(rec.PartialResult())["partial"] + " "
-
-    return "\n".join(result) + "\n" + partial_result, (rec, result)
-
-gr.Interface(
-    fn=transcribe,
-    inputs=[
-        gr.Audio(source="microphone", type="numpy", streaming=True),
-        "state"
-    ],
-    outputs=[
-        "textbox",
-        "state"
-    ],
-    live=True).launch(share=True)
+#!/usr/bin/env python3
+
+import json
+import gradio as gr
+
+from vosk import KaldiRecognizer, Model
+
+model = Model(lang="en-us")
+
+def transcribe(stream, new_chunk):
+
+    sample_rate, audio_data = new_chunk
+    audio_data = audio_data.tobytes()
+
+    if stream is None:
+        rec = KaldiRecognizer(model, sample_rate)
+        result = []
+    else:
+        rec, result = stream
+
+    if rec.AcceptWaveform(audio_data):
+        text_result = json.loads(rec.Result())["text"]
+        if text_result != "":
+            result.append(text_result)
+        partial_result = ""
+    else:
+        partial_result = json.loads(rec.PartialResult())["partial"] + " "
+
+    return (rec, result), "\n".join(result) + "\n" + partial_result
+
+gr.Interface(
+    fn=transcribe,
+    inputs=[
+        "state", gr.Audio(sources=["microphone"], type="numpy", streaming=True),
+    ],
+    outputs=[
+        "state", "text",
+    ],
+    live=True).launch(share=True)

python/example/test_itn.py

@@ -0,0 +1,11 @@
+#!/usr/bin/env python3
+
+import wave
+import sys
+
+from vosk import Processor
+
+proc = Processor("ru_itn_tagger.fst", "ru_itn_verbalizer.fst")
+print (proc.process("у нас десять яблок"))
+print (proc.process("у нас десять яблок и десять миллилитров воды точка"))
+print (proc.process("мы пришли в восемь часов пять минут"))

python/setup.py

@@ -45,7 +45,7 @@ with open("README.md", "rb") as fh:
 
 setuptools.setup(
     name="vosk",
-    version="0.3.45",
+    version="0.3.75",
     author="Alpha Cephei Inc",
     author_email="contact@alphacephei.com",
     description="Offline open source speech recognition API based on Kaldi and Vosk",

python/test/transcribe_scp.py

@@ -28,7 +28,7 @@ def recognize(line):
 
 def main():
     p = Pool(8)
-    texts = p.map(recognize, open(sys.argv[1], encoding="uft-8").readlines())
+    texts = p.map(recognize, open(sys.argv[1], encoding="utf-8").readlines())
     print ("\n".join(texts))
 
 main()

python/vosk/__init__.py

@@ -3,6 +3,7 @@ import sys
 import srt
 import datetime
 import json
+import enum
 
 import requests
 from urllib.request import urlretrieve
@@ -57,7 +58,8 @@ class Model:
             raise Exception("Failed to create a model")
 
     def __del__(self):
-        _c.vosk_model_free(self._handle)
+        if _c is not None:
+            _c.vosk_model_free(self._handle)
 
     def vosk_model_find_word(self, word):
         return _c.vosk_model_find_word(self._handle, word.encode("utf-8"))
@@ -140,6 +142,12 @@ class SpkModel:
     def __del__(self):
         _c.vosk_spk_model_free(self._handle)
 
+class EndpointerMode(enum.Enum):
+    DEFAULT = 0
+    SHORT = 1
+    LONG = 2
+    VERY_LONG = 3
+
 class KaldiRecognizer:
 
     def __init__(self, *args):
@@ -172,6 +180,12 @@ class KaldiRecognizer:
     def SetNLSML(self, enable_nlsml):
         _c.vosk_recognizer_set_nlsml(self._handle, 1 if enable_nlsml else 0)
 
+    def SetEndpointerMode(self, mode):
+        _c.vosk_recognizer_set_endpointer_mode(self._handle, mode.value)
+
+    def SetEndpointerDelays(self, t_start_max, t_end, t_max):
+        _c.vosk_recognizer_set_endpointer_delays(self._handle, t_start_max, t_end, t_max)
+
     def SetSpkModel(self, spk_model):
         _c.vosk_recognizer_set_spk_model(self._handle, spk_model._handle)
 
@@ -273,3 +287,17 @@ class BatchRecognizer:
 
     def GetPendingChunks(self):
         return _c.vosk_batch_recognizer_get_pending_chunks(self._handle)
+
+class Processor:
+
+    def __init__(self, *args):
+        self._handle = _c.vosk_text_processor_new(args[0].encode('utf-8'), args[1].encode('utf-8'))
+
+        if self._handle == _ffi.NULL:
+            raise Exception("Failed to create processor")
+
+    def __del__(self):
+        _c.vosk_text_processor_free(self._handle)
+
+    def process(self, text):
+        return _ffi.string(_c.vosk_text_processor_itn(self._handle, text.encode('utf-8'))).decode('utf-8')

python/vosk/transcriber/cli.py

@@ -15,8 +15,8 @@ parser.add_argument(
         "--model", "-m", type=str,
         help="model path")
 parser.add_argument(
-        "--server", "-s", const="ws://localhost:2700", action="store_const",
-        help="use server for recognition")
+        "--server", "-s", type=str,
+        help="use server for recognition. For example ws://localhost:2700")
 parser.add_argument(
         "--list-models", default=False, action="store_true",
         help="list available models")

python/vosk/transcriber/transcriber.py

@@ -99,7 +99,7 @@ class Transcriber:
             monologues = {"schemaVersion":"2.0", "monologues":[], "text":[]}
             for part in result:
                 if part["text"] != "":
-                    monologue["text"] += part["text"]
+                    monologues["text"] += [part["text"]]
             for _, res in enumerate(result):
                 if not "result" in res:
                     continue
@@ -133,6 +133,12 @@ class Transcriber:
             start_time = timer()
             proc = await self.resample_ffmpeg_async(input_file)
             result, tot_samples = await self.recognize_stream_server(proc)
+            await proc.wait()
+
+            # Bad input, continue
+            if tot_samples == 0:
+                 self.queue.task_done()
+                 continue
 
             processed_result = self.format_result(result)
             if output_file != "":
@@ -142,8 +148,6 @@ class Transcriber:
             else:
                 print(processed_result)
 
-            await proc.wait()
-
             elapsed = timer() - start_time
             logging.info("Execution time: {:.3f} sec; "\
                     "xRT {:.3f}".format(elapsed, float(elapsed) * (2 * SAMPLE_RATE) / tot_samples))
@@ -165,8 +169,10 @@ class Transcriber:
         rec = KaldiRecognizer(self.model, SAMPLE_RATE)
         rec.SetWords(True)
         result, tot_samples = self.recognize_stream(rec, stream)
-        processed_result = self.format_result(result)
+        if tot_samples == 0:
+            return
 
+        processed_result = self.format_result(result)
         if inputdata[1] != "":
             logging.info("File {} processing complete".format(inputdata[1]))
             with open(inputdata[1], "w", encoding="utf-8") as fh: