Move to 3-value endpointer config

android/build.gradle

@@ -10,7 +10,7 @@ buildscript {
 }
 
 allprojects {
-    version = '0.3.50'
+    version = '0.3.47'
 }
 
 subprojects {

android/lib/build-vosk.sh

@@ -118,7 +118,7 @@ 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 rnnlm
+make -j 8 online2 lm rnnlm
 
 # Vosk-api
 cd $WORKDIR

csharp/demo/VoskDemo.csproj

@@ -11,7 +11,7 @@
   </PropertyGroup>
 
   <ItemGroup>
-    <PackageReference Include="Vosk" Version="0.3.50" />
+    <PackageReference Include="Vosk" Version="0.3.45" />
   </ItemGroup>
 
 </Project>

csharp/nuget/Vosk.nuspec

@@ -2,7 +2,7 @@
 <package>
   <metadata>
     <id>Vosk</id>
-    <version>0.3.50</version>
+    <version>0.3.45</version>
     <authors>Alpha Cephei Inc</authors>
     <owners>Alpha Cephei Inc</owners>
     <license type="expression">Apache-2.0</license>

go/vosk.go

@@ -1,8 +1,7 @@
 package vosk
 
 // #cgo CPPFLAGS: -I ${SRCDIR}/../src
-// #cgo !windows LDFLAGS: -L ${SRCDIR}/../src -lvosk -ldl -lpthread
-// #cgo windows LDFLAGS: -L ${SRCDIR}/../src -lvosk -lpthread
+// #cgo LDFLAGS: -L ${SRCDIR}/../src -lvosk -ldl -lpthread
 // #include <stdlib.h>
 // #include <vosk_api.h>
 import "C"

java/demo/build.gradle

@@ -11,5 +11,5 @@ repositories {
 }
 
 dependencies {
-    implementation group: 'com.alphacephei', name: 'vosk', version: '0.3.50'
+    implementation group: 'com.alphacephei', name: 'vosk', version: '0.3.45'
 }

java/lib/build.gradle

@@ -16,7 +16,7 @@ repositories {
 
 archivesBaseName = 'vosk'
 group = 'com.alphacephei'
-version = '0.3.50'
+version = '0.3.45'
 
 mavenPublish {
     group = 'com.alphacephei'

kotlin/build.gradle.kts

@@ -26,7 +26,7 @@ plugins {
 }
 
 group = "com.alphacephei"
-version = "0.3.50"
+version = "0.4.0-alpha0"
 
 repositories {
 	google()

nodejs/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vosk",
-  "version": "0.3.50",
+  "version": "0.3.45",
   "description": "Node binding for continuous offline voice recoginition with Vosk library.",
   "repository": {
     "type": "git",

python/example/test_ep.py

@@ -41,7 +41,7 @@ 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)
+rec.SetEndpointerDelays(300, 500, 2000)
 
 while True:
     data = wf.readframes(4000)

python/example/test_gradio.py

@@ -1,39 +1,40 @@
-#!/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)
+#!/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)

python/example/test_itn.py

@@ -1,11 +0,0 @@
-#!/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="0.3.50",
+    version="0.3.46",
     author="Alpha Cephei Inc",
     author_email="contact@alphacephei.com",
     description="Offline open source speech recognition API based on Kaldi and Vosk",

python/vosk/__init__.py

@@ -144,9 +144,15 @@ class SpkModel:
 
 class EndpointerMode(enum.Enum):
     DEFAULT = 0
-    SHORT = 1
-    LONG = 2
-    VERY_LONG = 3
+    SINGLE_WORD = 1
+    VERY_SHORT = 2
+    SHORT = 3
+    STANDARD = 4
+    STANDARD_5 = 5
+    STANDARD_180 = 6
+    LONG = 7
+    VERY_LONG = 8
+    VERY_LONG_180 = 9
 
 class KaldiRecognizer:
 
@@ -183,8 +189,8 @@ class KaldiRecognizer:
     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 SetEndpointerDelays(self, sct, nit, t):
+        _c.vosk_recognizer_set_endpointer_delays(self._handle, sct, nit, t)
 
     def SetSpkModel(self, spk_model):
         _c.vosk_recognizer_set_spk_model(self._handle, spk_model._handle)
@@ -287,17 +293,3 @@ 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", type=str,
-        help="use server for recognition. For example ws://localhost:2700")
+        "--server", "-s", const="ws://localhost:2700", action="store_const",
+        help="use server for recognition")
 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"] != "":
-                    monologues["text"] += [part["text"]]
+                    monologues["text"] += part["text"]
             for _, res in enumerate(result):
                 if not "result" in res:
                     continue

ruby/vosk.gemspec

@@ -1,6 +1,6 @@
 Gem::Specification.new do |s|
   s.name        = "vosk"
-  s.version     = "0.3.50"
+  s.version     = "0.3.45"
   s.summary     = "Offline speech recognition API"
   s.description = "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."
   s.authors     = ["Alpha Cephei Inc"]

src/Makefile

@@ -23,18 +23,17 @@ VOSK_SOURCES= \
 	language_model.cc \
 	model.cc \
 	spk_model.cc \
-	vosk_api.cc \
-	postprocessor.cc
+	vosk_api.cc
 
 VOSK_HEADERS= \
 	recognizer.h \
 	language_model.h \
 	model.h \
 	spk_model.h \
-	vosk_api.h \
-        postprocessor.h
+	vosk_api.h
 
-CFLAGS=-g -O3 -std=c++17 -Wno-deprecated-declarations -fPIC -DFST_NO_DYNAMIC_LINKING -I. -I$(KALDI_ROOT)/src -I$(OPENFST_ROOT)/include $(EXTRA_CFLAGS)
+CFLAGS=-g -O3 -std=c++17 -Wno-deprecated-declarations -fPIC -DFST_NO_DYNAMIC_LINKING \
+       -I. -I$(KALDI_ROOT)/src -I$(OPENFST_ROOT)/include $(EXTRA_CFLAGS)
 
 LDFLAGS=
 

src/postprocessor.cc

@@ -1,66 +0,0 @@
-// Copyright (c) 2022 Zhendong Peng (pzd17@tsinghua.org.cn)
-//
-// 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.
-
-#include "postprocessor.h"
-
-using fst::TokenType;
-
-Processor::Processor(const std::string& tagger_path,
-                     const std::string& verbalizer_path) {
-  tagger_.reset(StdVectorFst::Read(tagger_path));
-  verbalizer_.reset(StdVectorFst::Read(verbalizer_path));
-  compiler_ = std::make_shared<StringCompiler<StdArc>>(TokenType::BYTE);
-  printer_ = std::make_shared<StringPrinter<StdArc>>(TokenType::BYTE);
-}
-
-std::string Processor::ShortestPath(const StdVectorFst& lattice) {
-  StdVectorFst shortest_path;
-  fst::ShortestPath(lattice, &shortest_path, 1, true);
-
-  std::string output;
-  printer_->operator()(shortest_path, &output);
-  return output;
-}
-
-std::string Processor::Compose(const std::string& input,
-                               const StdVectorFst* fst) {
-  StdVectorFst input_fst;
-  compiler_->operator()(input, &input_fst);
-
-  StdVectorFst lattice;
-  fst::Compose(input_fst, *fst, &lattice);
-  return ShortestPath(lattice);
-}
-
-std::string Processor::Tag(const std::string& input) {
-  if (input.empty()) {
-    return "";
-  }
-  return Compose(input, tagger_.get());
-}
-
-std::string Processor::Verbalize(const std::string& input) {
-  if (input.empty()) {
-    return "";
-  }
-  std::string output = Compose(input, verbalizer_.get());
-  output.erase(std::remove(output.begin(), output.end(), '\0'), output.end());
-  return output;
-}
-
-std::string Processor::Normalize(const std::string& input) {
-  return Verbalize(Tag(input));
-}
-
-

src/postprocessor.h

@@ -1,45 +0,0 @@
-// Copyright (c) 2022 Zhendong Peng (pzd17@tsinghua.org.cn)
-//
-// 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.
-
-#ifndef PROCESSOR_WETEXT_PROCESSOR_H_
-#define PROCESSOR_WETEXT_PROCESSOR_H_
-
-#include <memory>
-#include <string>
-
-#include "fst/fstlib.h"
-
-using fst::StdArc;
-using fst::StdVectorFst;
-using fst::StringCompiler;
-using fst::StringPrinter;
-
-class Processor {
- public:
-  Processor(const std::string& tagger_path, const std::string& verbalizer_path);
-  std::string Tag(const std::string& input);
-  std::string Verbalize(const std::string& input);
-  std::string Normalize(const std::string& input);
-
- private:
-  std::string ShortestPath(const StdVectorFst& lattice);
-  std::string Compose(const std::string& input, const StdVectorFst* fst);
-
-  std::shared_ptr<StdVectorFst> tagger_ = nullptr;
-  std::shared_ptr<StdVectorFst> verbalizer_ = nullptr;
-  std::shared_ptr<StringCompiler<StdArc>> compiler_ = nullptr;
-  std::shared_ptr<StringPrinter<StdArc>> printer_ = nullptr;
-};
-
-#endif  // PROCESSOR_WETEXT_PROCESSOR_H_

src/recognizer.cc

@@ -222,36 +222,110 @@ void Recognizer::SetNLSML(bool nlsml)
 
 void Recognizer::SetEndpointerMode(int mode)
 {
+    float rule1, rule2, rule3, rule4, rule5;
+    if (mode == 0) {
+        endpoint_config_ = model_->endpoint_config_;
+        return;
+    }
+
     float scale = 1.0;
     switch(mode) {
-        case 1:
-           scale = 0.75;
+        case 1: // SINGLE_WORD
+           rule2 = 0.1;
+           rule3 = 0.15;
+           rule4 = 0.2;
+           rule1 = 4.0;
+           rule5 = 7.0;
            break;
-        case 2:
-           scale = 1.50;
+        case 2: // VERY_SHORT
+           rule2 = 0.3;
+           rule3 = 0.3;
+           rule4 = 0.6;
+           rule1 = 3.0;
+           rule5 = 5.0;
            break;
-        case 3:
-           scale = 4.0;
+        case 3: // SHORT
+           rule2 = 0.3;
+           rule3 = 0.4;
+           rule4 = 0.6;
+           rule1 = 3.0;
+           rule5 = 5.0;
+           break;
+        case 4: // STANDARD
+           rule2 = 0.8;
+           rule3 = 0.9;
+           rule4 = 1.0;
+           rule1 = 3.0;
+           rule5 = 7.0;
+           break;
+        case 5: // STANDARD_5
+           rule2 = 0.8;
+           rule3 = 0.9;
+           rule4 = 1.5;
+           rule1 = 5.0;
+           rule5 = 7.0;
+           break;
+        case 6: // STANDARD_180
+           rule2 = 0.8;
+           rule3 = 0.9;
+           rule4 = 1.5;
+           rule1 = 3.0;
+           rule5 = 180.0;
+           break;
+        case 7: // LONG
+           rule2 = 1.0;
+           rule3 = 1.2;
+           rule4 = 2.0;
+           rule1 = 4.0;
+           rule5 = 10.0;
+           break;
+        case 8: // VERY_LONG
+           rule2 = 2.0;
+           rule3 = 2.5;
+           rule4 = 3.0;
+           rule1 = 4.0;
+           rule5 = 15.0;
+           break;
+        case 9: // VERY_LONG_180
+           rule2 = 2.0;
+           rule3 = 2.5;
+           rule4 = 3.0;
+           rule1 = 4.0;
+           rule5 = 180.0;
+           break;
+        default: // STANDARD
+           rule2 = 0.8;
+           rule3 = 0.9;
+           rule4 = 1.0;
+           rule1 = 5.0;
+           rule5 = 7.0;
            break;
     }
-    KALDI_LOG << "Updating endpointer scale " << scale;
+    KALDI_LOG << "Updating endpointer timeouts to " << rule1 << "," << rule2 << "," << rule3 << "," << rule4 << "," << rule5;
     endpoint_config_ = model_->endpoint_config_;
-    endpoint_config_.rule2.min_trailing_silence *= scale;
-    endpoint_config_.rule3.min_trailing_silence *= scale;
-    endpoint_config_.rule4.min_trailing_silence *= scale;
+    endpoint_config_.rule1.min_trailing_silence = rule1;
+    endpoint_config_.rule2.min_trailing_silence = rule2;
+    endpoint_config_.rule3.min_trailing_silence = rule3;
+    endpoint_config_.rule4.min_trailing_silence = rule4;
+    endpoint_config_.rule5.min_utterance_length = rule5;
 }
 
-void Recognizer::SetEndpointerDelays(float t_start_max, float t_end, float t_max)
+void Recognizer::SetEndpointerDelays(int sct, int nit, int t)
 {
     float rule1, rule2, rule3, rule4, rule5;
 
-    rule1 = t_start_max;
-    rule2 = t_end;
-    rule3 = t_end + 0.5;
-    rule4 = t_end + 1.0;
-    rule5 = t_max;
+    rule2 = sct / 1000.0;
+    if (sct < 500) {
+        rule3 = sct / 1000.0 * 1.5;
+        rule4 = sct / 1000.0 * 2;
+    } else  {
+        rule3 = sct / 1000.0 + 0.5;
+        rule4 = sct / 1000.0 + 1.0;
+    }
+    rule1 = nit / 1000.0;
+    rule5 = t / 1000.0;
 
-    KALDI_LOG << "Updating endpointer delays " << rule1 << "," << rule2 << "," << rule3 << "," << rule4 << "," << rule5;
+    KALDI_LOG << "Updating endpointer timeouts to " << rule1 << "," << rule2 << "," << rule3 << "," << rule4 << "," << rule5;
     endpoint_config_ = model_->endpoint_config_;
     endpoint_config_.rule1.min_trailing_silence = rule1;
     endpoint_config_.rule2.min_trailing_silence = rule2;
@@ -275,7 +349,7 @@ void Recognizer::SetSpkModel(SpkModel *spk_model)
 void Recognizer::SetGrm(char const *grammar)
 {
     if (state_ == RECOGNIZER_RUNNING) {
-        KALDI_ERR << "Can't add grammar to already running recognizer";
+        KALDI_ERR << "Can't add speaker model to already running recognizer";
         return;
     }
 

src/recognizer.h

@@ -53,7 +53,7 @@ class Recognizer {
         void SetPartialWords(bool partial_words);
         void SetNLSML(bool nlsml);
         void SetEndpointerMode(int mode);
-        void SetEndpointerDelays(float t_start_max, float t_end, float t_max);
+        void SetEndpointerDelays(int sct, int nit, int t);
         bool AcceptWaveform(const char *data, int len);
         bool AcceptWaveform(const short *sdata, int len);
         bool AcceptWaveform(const float *fdata, int len);

src/vosk_api.cc

@@ -17,7 +17,6 @@
 #include "recognizer.h"
 #include "model.h"
 #include "spk_model.h"
-#include "postprocessor.h"
 
 #if HAVE_CUDA
 #include "cudamatrix/cu-device.h"
@@ -138,12 +137,12 @@ void vosk_recognizer_set_endpointer_mode(VoskRecognizer *recognizer, VoskEndpoin
     ((Recognizer *)recognizer)->SetEndpointerMode(mode);
 }
 
-void vosk_recognizer_set_endpointer_delays(VoskRecognizer *recognizer, float t_start_max, float t_end, float t_max)
+void vosk_recognizer_set_endpointer_delays(VoskRecognizer *recognizer, int sct, int nit, int t)
 {
     if (recognizer == nullptr) {
        return;
     }
-    ((Recognizer *)recognizer)->SetEndpointerDelays(t_start_max, t_end, t_max);
+    ((Recognizer *)recognizer)->SetEndpointerDelays(sct, nit, t);
 }
 
 int vosk_recognizer_accept_waveform(VoskRecognizer *recognizer, const char *data, int length)
@@ -305,28 +304,3 @@ int vosk_batch_recognizer_get_pending_chunks(VoskBatchRecognizer *recognizer)
     return 0;
 #endif
 }
-
-VoskTextProcessor *vosk_text_processor_new(const char *tagger, const char *verbalizer)
-{
-    try {
-        return (VoskTextProcessor *)new Processor(tagger, verbalizer);
-    } catch (...) {
-        return nullptr;
-    }
-}
-
-void vosk_text_processor_free(VoskTextProcessor *processor)
-{
-    delete ((Processor *)processor);
-}
-
-char *vosk_text_processor_itn(VoskTextProcessor *processor, const char *input)
-{
-    Processor *wprocessor = (Processor *)processor;
-    std::string sinput(input);
-
-    std::string tagged_text = wprocessor->Tag(sinput);
-    std::string normalized_text = wprocessor->Verbalize(tagged_text);
-
-    return strdup(normalized_text.c_str());
-}

src/vosk_api.h

@@ -39,8 +39,6 @@ typedef struct VoskSpkModel VoskSpkModel;
  *  speaker information and so on */
 typedef struct VoskRecognizer VoskRecognizer;
 
-/** Inverse text normalization */
-typedef struct VoskTextProcessor VoskTextProcessor;
 
 /**
  * Batch model object
@@ -219,11 +217,17 @@ void vosk_recognizer_set_partial_words(VoskRecognizer *recognizer, int partial_w
  */
 void vosk_recognizer_set_nlsml(VoskRecognizer *recognizer, int nlsml);
 
-typedef enum VoskEpMode {
-    VOSK_EP_ANSWER_DEFAULT = 0,
-    VOSK_EP_ANSWER_SHORT = 1,
-    VOSK_EP_ANSWER_LONG = 2,
-    VOSK_EP_ANSWER_VERY_LONG = 3,
+typedef enum VoskEndpointerMode {
+    VOSK_EP_DEFAULT = 0,
+    VOSK_EP_SINGLE_WORD = 1,
+    VOSK_EP_VERY_SHORT = 2,
+    VOSK_EP_SHORT = 3,
+    VOSK_EP_STANDARD = 4,
+    VOSK_EP_STANDARD_5 = 5,
+    VOSK_EP_STANDARD_180 = 6,
+    VOSK_EP_LONG = 7,
+    VOSK_EP_VERY_LONG = 8,
+    VOSK_EP_VERY_LONG_180 = 9
 } VoskEndpointerMode;
 
 /**
@@ -236,11 +240,12 @@ void vosk_recognizer_set_endpointer_mode(VoskRecognizer *recognizer,  VoskEndpoi
 /**
  * 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)
+ * @param sct   speech complete timeout
+ * @param sint  speech incomplete timeout
+ * @param nit   no input timeout
+ * @param t     recognition timeout
  **/
-void vosk_recognizer_set_endpointer_delays(VoskRecognizer *recognizer, float t_start_max, float t_end, float t_max);
+void vosk_recognizer_set_endpointer_delays(VoskRecognizer *recognizer, int sct, int nit, int t);
 
 /** Accept voice data
  *
@@ -378,15 +383,6 @@ void vosk_batch_recognizer_pop(VoskBatchRecognizer *recognizer);
 /** Get amount of pending chunks for more intelligent waiting */
 int vosk_batch_recognizer_get_pending_chunks(VoskBatchRecognizer *recognizer);
 
-/** Create text processor */
-VoskTextProcessor *vosk_text_processor_new(const char *tagger, const char *verbalizer);
-
-/** Release text processor */
-void vosk_text_processor_free(VoskTextProcessor *processor);
-
-/** Convert string */
-char *vosk_text_processor_itn(VoskTextProcessor *processor, const char *input);
-
 #ifdef __cplusplus
 }
 #endif

travis/Dockerfile.dockcross

@@ -45,5 +45,5 @@ RUN cd /opt \
     && sed -i "s:TARGET_ARCH=\"\`uname -m\`\":TARGET_ARCH=$(echo $CROSS_TRIPLE|cut -d - -f 1):g" configure \
     && sed -i "s: -O1 : -O3 :g" makefiles/linux_openblas_arm.mk \
     && ./configure --mathlib=OPENBLAS_CLAPACK --shared --use-cuda=no \
-    && make -j 10 online2 rnnlm \
+    && make -j 10 online2 lm rnnlm \
     && find /opt/kaldi -name "*.o" -exec rm {} \;

travis/Dockerfile.dockcross-manylinux

@@ -35,5 +35,5 @@ RUN cd /opt \
     && sed -i "s:TARGET_ARCH=\"\`uname -m\`\":TARGET_ARCH=$(echo $CROSS_TRIPLE|cut -d - -f 1):g" configure \
     && sed -i "s: -O1 : -O3 :g" makefiles/linux_openblas_arm.mk \
     && ./configure --mathlib=OPENBLAS_CLAPACK --shared --use-cuda=no \
-    && make -j 10 online2 rnnlm \
+    && make -j 10 online2 lm rnnlm \
     && find /opt/kaldi -name "*.o" -exec rm {} \;

travis/Dockerfile.manylinux

@@ -30,5 +30,5 @@ RUN cd /opt \
     && ./configure --mathlib=OPENBLAS_CLAPACK --shared --use-cuda=no \
     && sed -i 's:-msse -msse2:-msse -msse2:g' kaldi.mk \
     && sed -i 's: -O1 : -O3 :g' kaldi.mk \
-    && make -j $(nproc) online2 rnnlm \
+    && make -j $(nproc) online2 lm rnnlm \
     && find /opt/kaldi -name "*.o" -exec rm {} \;

travis/Dockerfile.manylinux-mkl

@@ -27,5 +27,5 @@ RUN cd /opt \
     && ./configure --mathlib=MKL --shared --use-cuda=no \
     && sed -i 's:-msse -msse2:-msse -msse2 -mavx -mavx2:g' kaldi.mk \
     && sed -i 's: -O1 : -O3 :g' kaldi.mk \
-    && make -j $(nproc) online2 rnnlm \
+    && make -j $(nproc) online2 lm rnnlm \
     && find /opt/kaldi -name "*.o" -exec rm {} \;

travis/Dockerfile.win

@@ -62,4 +62,4 @@ RUN cd /opt/kaldi \
         --host=x86_64-w64-mingw32 --openblas-clapack-root=/opt/kaldi/local \
         --fst-root=/opt/kaldi/local --fst-version=1.8.0 \
     && make depend -j \
-    && make -j $(nproc) online2 rnnlm
+    && make -j $(nproc) online2 lm rnnlm

travis/Dockerfile.win32

@@ -61,4 +61,4 @@ RUN cd /opt/kaldi \
         --host=i686-w64-mingw32 --openblas-clapack-root=/opt/kaldi/local \
         --fst-root=/opt/kaldi/local --fst-version=1.8.0 \
     && make depend -j \
-    && make -j $(nproc) online2 rnnlm
+    && make -j $(nproc) online2 lm rnnlm

webjs/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vosk-js",
-  "version": "0.3.50",
+  "version": "0.3.45",
   "description": "Node binding for continuous voice recoginition through vosk-api.",
   "repository": {
     "type": "git",