Bump version to 0.3.32

are implemented, others on the way

android/lib/build-vosk.sh

@@ -105,7 +105,7 @@ make install
 
 # Kaldi itself
 cd $WORKDIR
-git clone -b android-mix --single-branch https://github.com/alphacep/kaldi
+git clone -b vosk-android --single-branch https://github.com/alphacep/kaldi
 cd $WORKDIR/kaldi/src
 CXX=$CXX CXXFLAGS="$ARCHFLAGS -O3 -DFST_NO_DYNAMIC_LINKING" ./configure --use-cuda=no \
     --mathlib=OPENBLAS_CLAPACK --shared \

csharp/demo/VoskDemo.csproj

@@ -11,7 +11,7 @@
   </PropertyGroup>
 
   <ItemGroup>
-    <PackageReference Include="Vosk" Version="0.3.30" />
+    <PackageReference Include="Vosk" Version="0.3.31" />
   </ItemGroup>
 
 </Project>

csharp/nuget/Vosk.nuspec

@@ -2,7 +2,7 @@
 <package>
   <metadata>
     <id>Vosk</id>
-    <version>0.3.30</version>
+    <version>0.3.32</version>
     <authors>Alpha Cephei Inc</authors>
     <owners>Alpha Cephei Inc</owners>
     <license type="expression">Apache-2.0</license>

go/example/test_simple.go

@@ -21,13 +21,8 @@ func main() {
 		log.Fatal(err)
 	}
 
-	spkModel, err := vosk.NewSpkModel("model-spk")
-	if err != nil {
-		log.Fatal(err)
-	}
-
 	sampleRate := 16000.0
-	rec, err := vosk.NewRecognizer(model, sampleRate, spkModel)
+	rec, err := vosk.NewRecognizer(model, sampleRate)
 	if err != nil {
 		log.Fatal(err)
 	}
@@ -39,12 +34,6 @@ func main() {
 	}
 	defer file.Close()
 
-	file, err := os.Open(filename)
-	if err != nil {
-		panic(err)
-	}
-	defer file.Close()
-
 	reader := bufio.NewReader(file)
 	buf := make([]byte, 4096)
 

go/vosk.go

@@ -18,6 +18,10 @@ func NewModel(modelPath string) (*VoskModel, error) {
 	return model, nil
 }
 
+func (m *VoskModel) Free() {
+	C.vosk_model_free(m.model)
+}
+
 func freeModel(model *VoskModel) {
 	C.vosk_model_free(model.model)
 }
@@ -48,6 +52,10 @@ func freeSpkModel(model *VoskSpkModel) {
 	C.vosk_spk_model_free(model.spkModel)
 }
 
+func(s *VoskSpkModel) Free() {
+	C.vosk_spk_model_free(s.spkModel)
+}
+
 // VoskRecognizer contains a reference to the C VoskRecognizer
 type VoskRecognizer struct {
 	rec *C.struct_VoskRecognizer
@@ -57,6 +65,10 @@ func freeRecognizer(recognizer *VoskRecognizer) {
 	C.vosk_recognizer_free(recognizer.rec)
 }
 
+func (r *VoskRecognizer) Free() {
+	C.vosk_recognizer_free(r.rec)
+}
+
 // NewRecognizer creates a new VoskRecognizer instance
 func NewRecognizer(model *VoskModel, sampleRate float64) (*VoskRecognizer, error) {
 	internal := C.vosk_recognizer_new(model.model, C.float(sampleRate))
@@ -110,7 +122,7 @@ func (r *VoskRecognizer) Result() []byte {
 
 // PartialResult returns a partial speech recognition result.
 func (r *VoskRecognizer) PartialResult() []byte {
-	return []byte(C.GoString(C.vosk_recognizer_result(r.rec)))
+	return []byte(C.GoString(C.vosk_recognizer_partial_result(r.rec)))
 }
 
 // FinalResult returns a speech recognition result. Same as result, but doesn't wait

java/demo/build.gradle

@@ -15,5 +15,5 @@ repositories {
 
 dependencies {
     implementation group: 'net.java.dev.jna', name: 'jna', version: '5.7.0'
-    implementation group: 'com.alphacephei', name: 'vosk', version: '0.3.30+'
+    implementation group: 'com.alphacephei', name: 'vosk', version: '0.3.32+'
 }

java/lib/build.gradle

@@ -5,7 +5,7 @@ plugins {
 
 archivesBaseName = 'vosk'
 group = 'com.alphacephei'
-version = '0.3.30'
+version = '0.3.32'
 
 repositories {
     mavenCentral()

java/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

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

@@ -1,10 +1,15 @@
 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) {
+    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) {

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

@@ -1,13 +1,18 @@
 package org.vosk;
 
 import com.sun.jna.PointerType;
+import java.io.IOException;
 
 public class SpeakerModel extends PointerType implements AutoCloseable {
     public SpeakerModel() {
     }
 
-    public SpeakerModel(String path) {
+    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

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

@@ -93,4 +93,10 @@ public class DecoderTest {
         }
         Assert.assertTrue(true);
     }
+
+
+    @Test(expected = IOException.class)
+    public void decoderTestException() throws IOException {
+        Model model = new Model("model_missing");
+    }
 }

nodejs/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vosk",
-  "version": "0.3.30",
+  "version": "0.3.32",
   "description": "Node binding for continuous offline voice recoginition with Vosk library.",
   "repository": {
     "type": "git",

python/setup.py

@@ -44,7 +44,7 @@ with open("README.md", "r") as fh:
 
 setuptools.setup(
     name="vosk",
-    version="0.3.31",
+    version="0.3.32",
     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

@@ -25,6 +25,9 @@ class Model(object):
     def __init__(self, model_path):
         self._handle = _c.vosk_model_new(model_path.encode('utf-8'))
 
+        if self._handle == _ffi.NULL:
+            raise Exception("Failed to create a model")
+
     def __del__(self):
         _c.vosk_model_free(self._handle)
 
@@ -36,6 +39,9 @@ class SpkModel(object):
     def __init__(self, model_path):
         self._handle = _c.vosk_spk_model_new(model_path.encode('utf-8'))
 
+        if self._handle == _ffi.NULL:
+            raise Exception("Failed to create a speaker model")
+
     def __del__(self):
         _c.vosk_spk_model_free(self._handle)
 
@@ -51,6 +57,9 @@ class KaldiRecognizer(object):
         else:
             raise TypeError("Unknown arguments")
 
+        if self._handle == _ffi.NULL:
+            raise Exception("Failed to create a recognizer")
+
     def __del__(self):
         _c.vosk_recognizer_free(self._handle)
 
@@ -64,7 +73,10 @@ class KaldiRecognizer(object):
         _c.vosk_recognizer_set_spk_model(self._handle, spk_model._handle)
 
     def AcceptWaveform(self, data):
-        return _c.vosk_recognizer_accept_waveform(self._handle, data, len(data))
+        res = _c.vosk_recognizer_accept_waveform(self._handle, data, len(data))
+        if res < 0:
+            raise Exception("Failed to process waveform")
+        return res
 
     def Result(self):
         return _ffi.string(_c.vosk_recognizer_result(self._handle)).decode('utf-8')

src/Makefile

@@ -23,6 +23,13 @@ VOSK_SOURCES= \
 	spk_model.cc \
 	vosk_api.cc
 
+VOSK_HEADERS= \
+	kaldi_recognizer.h \
+	language_model.h \
+	model.h \
+	spk_model.h \
+	vosk_api.h
+
 CFLAGS=-g -O3 -std=c++17 -fPIC -DFST_NO_DYNAMIC_LINKING \
 	-I. -I$(KALDI_ROOT)/src -I$(OPENFST_ROOT)/include $(EXTRA_CFLAGS) 
 
@@ -76,7 +83,7 @@ all: libvosk.$(EXT)
 libvosk.$(EXT): $(VOSK_SOURCES:.cc=.o)
 	$(CXX) --shared -s -o $@ $^ $(LIBS) -lm -latomic $(EXTRA_LDFLAGS)
 
-%.o: %.cc
+%.o: %.cc $(VOSK_HEADERS)
 	$(CXX) $(CFLAGS) -c -o $@ $<
 
 clean:

src/kaldi_recognizer.cc

@@ -143,11 +143,10 @@ KaldiRecognizer::~KaldiRecognizer() {
     delete decode_fst_;
     delete spk_feature_;
 
-    delete rnnlm_info_;
     delete lm_to_subtract_;
-    delete lm_to_subtract_scale_;
     delete carpa_to_add_;
     delete carpa_to_add_scale_;
+    delete rnnlm_info_;
     delete rnnlm_to_add_;
     delete rnnlm_to_add_scale_;
 
@@ -168,8 +167,12 @@ void KaldiRecognizer::InitState()
 void KaldiRecognizer::InitRescoring()
 {
     if (model_->graph_lm_fst_) {
-        lm_to_subtract_ = new fst::BackoffDeterministicOnDemandFst<StdArc>(*model_->graph_lm_fst_);
-        lm_to_subtract_scale_ = new fst::ScaleDeterministicOnDemandFst(-1.0, lm_to_subtract_);
+
+        fst::CacheOptions cache_opts(true, -1);
+        fst::ArcMapFstOptions mapfst_opts(cache_opts);
+        fst::StdToLatticeMapper<BaseFloat> mapper;
+
+        lm_to_subtract_ = new fst::ArcMapFst<fst::StdArc, LatticeArc, fst::StdToLatticeMapper<BaseFloat> >(*model_->graph_lm_fst_, mapper, mapfst_opts);
         carpa_to_add_ = new ConstArpaLmDeterministicFst(model_->const_arpa_);
 
         if (model_->rnnlm_enabled_) {
@@ -535,11 +538,12 @@ const char *KaldiRecognizer::NbestResult(CompactLattice &clat)
     for (int k = 0; k < nbest_lats.size(); k++) {
 
       Lattice nlat = nbest_lats[k];
-      RmEpsilon(&nlat);
-      CompactLattice nclat;
-      CompactLattice aligned_nclat;
-      ConvertLattice(nlat, &nclat);
 
+      CompactLattice nclat;
+      fst::Invert(&nlat);
+      DeterminizeLattice(nlat, &nclat);
+
+      CompactLattice aligned_nclat;
       if (model_->winfo_) {
           WordAlignLattice(nclat, *model_->trans_model_, *model_->winfo_, 0, &aligned_nclat);
       } else {
@@ -586,34 +590,40 @@ const char* KaldiRecognizer::GetResult()
         return StoreEmptyReturn();
     }
 
-    kaldi::CompactLattice clat;
-    kaldi::CompactLattice rlat;
+    // Original from decoder, subtracted graph weight, rescored with carpa, rescored with rnnlm
+    CompactLattice clat, slat, tlat, rlat;
+
     decoder_->GetLattice(true, &clat);
 
-    if (lm_to_subtract_scale_ && carpa_to_add_) {
-        TopSortCompactLatticeIfNeeded(&clat);
-        CompactLattice tlat;
-        fst::ComposeDeterministicOnDemandFst<StdArc> combined_lm(lm_to_subtract_scale_, carpa_to_add_);
-        ComposeCompactLatticeDeterministic(clat, &combined_lm, &tlat);
+    if (lm_to_subtract_ && carpa_to_add_) {
+        Lattice lat, composed_lat;
 
+        // Delete old score
+        ConvertLattice(clat, &lat);
+        fst::ScaleLattice(fst::GraphLatticeScale(-1.0), &lat);
+        fst::Compose(lat, *lm_to_subtract_, &composed_lat);
+        fst::Invert(&composed_lat);
+        DeterminizeLattice(composed_lat, &slat);
+        fst::ScaleLattice(fst::GraphLatticeScale(-1.0), &slat);
+
+        // Add CARPA score
+        TopSortCompactLatticeIfNeeded(&slat);
+        ComposeCompactLatticeDeterministic(slat, carpa_to_add_, &tlat);
+
+        // Rescore with RNNLM score on top if needed
         if (rnnlm_to_add_scale_) {
              ComposeLatticePrunedOptions compose_opts;
              compose_opts.lattice_compose_beam = 3.0;
              compose_opts.max_arcs = 3000;
-             TopSortCompactLatticeIfNeeded(&tlat);
              fst::ComposeDeterministicOnDemandFst<StdArc> combined_rnnlm(carpa_to_add_scale_, rnnlm_to_add_scale_);
+
+             TopSortCompactLatticeIfNeeded(&tlat);
              ComposeCompactLatticePruned(compose_opts, tlat,
                                          &combined_rnnlm, &rlat);
              rnnlm_to_add_->Clear();
         } else {
              rlat = tlat;
         }
-
-        kaldi::Lattice slat;
-        ConvertLattice(rlat, &slat);
-        fst::Invert(&slat);
-        DeterminizeLattice(slat, &rlat);
-
     } else {
         rlat = clat;
     }
@@ -642,7 +652,7 @@ const char* KaldiRecognizer::PartialResult()
         return StoreReturn(res.dump());
     }
 
-    kaldi::Lattice lat;
+    Lattice lat;
     decoder_->GetBestPath(false, &lat);
     vector<kaldi::int32> alignment, words;
     LatticeWeight weight;

src/kaldi_recognizer.h

@@ -82,11 +82,9 @@ class KaldiRecognizer {
         OnlineBaseFeature *spk_feature_ = nullptr;
 
         // Rescoring
-        fst::BackoffDeterministicOnDemandFst<fst::StdArc> *lm_to_subtract_ = nullptr;
-        fst::ScaleDeterministicOnDemandFst *lm_to_subtract_scale_ = nullptr;
+        fst::ArcMapFst<fst::StdArc, LatticeArc, fst::StdToLatticeMapper<BaseFloat> > *lm_to_subtract_ = nullptr;
         kaldi::ConstArpaLmDeterministicFst *carpa_to_add_ = nullptr;
         fst::ScaleDeterministicOnDemandFst *carpa_to_add_scale_ = nullptr;
-
         // RNNLM rescoring
         kaldi::rnnlm::KaldiRnnlmDeterministicFst* rnnlm_to_add_ = nullptr;
         fst::DeterministicOnDemandFst<fst::StdArc> *rnnlm_to_add_scale_ = nullptr;

src/model.cc

@@ -109,10 +109,10 @@ Model::Model(const char *model_path) : model_path_str_(model_path) {
 
     struct stat buffer;
     string am_v2_path = model_path_str_ + "/am/final.mdl";
-    string mfcc_v2_path = model_path_str_ + "/conf/mfcc.conf";
+    string model_conf_v2_path = model_path_str_ + "/conf/model.conf";
     string am_v1_path = model_path_str_ + "/final.mdl";
     string mfcc_v1_path = model_path_str_ + "/mfcc.conf";
-    if (stat(am_v2_path.c_str(), &buffer) == 0 && stat(mfcc_v2_path.c_str(), &buffer) == 0) {
+    if (stat(am_v2_path.c_str(), &buffer) == 0 && stat(model_conf_v2_path.c_str(), &buffer) == 0) {
         ConfigureV2();
         ReadDataFiles();
     } else if (stat(am_v1_path.c_str(), &buffer) == 0 && stat(mfcc_v1_path.c_str(), &buffer) == 0) {
@@ -168,6 +168,7 @@ void Model::ConfigureV1()
     std_fst_rxfilename_ = model_path_str_ + "/rescore/G.fst";
     final_ie_rxfilename_ = model_path_str_ + "/ivector/final.ie";
     mfcc_conf_rxfilename_ = model_path_str_ + "/mfcc.conf";
+    fbank_conf_rxfilename_ = model_path_str_ + "/fbank.conf";
     global_cmvn_stats_rxfilename_ = model_path_str_ + "/global_cmvn.stats";
     pitch_conf_rxfilename_ = model_path_str_ + "/pitch.conf";
     rnnlm_word_feats_rxfilename_ = model_path_str_ + "/rnnlm/word_feats.txt";
@@ -196,6 +197,7 @@ void Model::ConfigureV2()
     std_fst_rxfilename_ = model_path_str_ + "/rescore/G.fst";
     final_ie_rxfilename_ = model_path_str_ + "/ivector/final.ie";
     mfcc_conf_rxfilename_ = model_path_str_ + "/conf/mfcc.conf";
+    fbank_conf_rxfilename_ = model_path_str_ + "/conf/fbank.conf";
     global_cmvn_stats_rxfilename_ = model_path_str_ + "/am/global_cmvn.stats";
     pitch_conf_rxfilename_ = model_path_str_ + "/conf/pitch.conf";
     rnnlm_word_feats_rxfilename_ = model_path_str_ + "/rnnlm/word_feats.txt";
@@ -213,9 +215,17 @@ void Model::ReadDataFiles()
          " lattice-beam=" << nnet3_decoding_config_.lattice_beam;
     KALDI_LOG << "Silence phones " << endpoint_config_.silence_phones;
 
-    feature_info_.feature_type = "mfcc";
-    ReadConfigFromFile(mfcc_conf_rxfilename_, &feature_info_.mfcc_opts);
-    feature_info_.mfcc_opts.frame_opts.allow_downsample = true; // It is safe to downsample
+    if (stat(mfcc_conf_rxfilename_.c_str(), &buffer) == 0) {
+        feature_info_.feature_type = "mfcc";
+        ReadConfigFromFile(mfcc_conf_rxfilename_, &feature_info_.mfcc_opts);
+        feature_info_.mfcc_opts.frame_opts.allow_downsample = true; // It is safe to downsample
+    } else if (stat(fbank_conf_rxfilename_.c_str(), &buffer) == 0) {
+        feature_info_.feature_type = "fbank";
+        ReadConfigFromFile(fbank_conf_rxfilename_, &feature_info_.fbank_opts);
+        feature_info_.fbank_opts.frame_opts.allow_downsample = true; // It is safe to downsample
+    } else {
+        KALDI_ERR << "Failed to find feature config file";
+    }
 
     feature_info_.silence_weighting_config.silence_weight = 1e-3;
     feature_info_.silence_weighting_config.silence_phones_str = endpoint_config_.silence_phones;

src/model.h

@@ -66,6 +66,7 @@ protected:
     string std_fst_rxfilename_;
     string final_ie_rxfilename_;
     string mfcc_conf_rxfilename_;
+    string fbank_conf_rxfilename_;
     string global_cmvn_stats_rxfilename_;
     string pitch_conf_rxfilename_;
 

src/vosk_api.cc

@@ -27,11 +27,18 @@ using namespace kaldi;
 
 VoskModel *vosk_model_new(const char *model_path)
 {
-    return (VoskModel *)new Model(model_path);
+    try {
+        return (VoskModel *)new Model(model_path);
+    } catch (...) {
+        return nullptr;
+    }
 }
 
 void vosk_model_free(VoskModel *model)
 {
+    if (model == nullptr) {
+       return;
+    }
     ((Model *)model)->Unref();
 }
 
@@ -42,27 +49,46 @@ int vosk_model_find_word(VoskModel *model, const char *word)
 
 VoskSpkModel *vosk_spk_model_new(const char *model_path)
 {
-    return (VoskSpkModel *)new SpkModel(model_path);
+    try {
+        return (VoskSpkModel *)new SpkModel(model_path);
+    } catch (...) {
+        return nullptr;
+    }
 }
 
 void vosk_spk_model_free(VoskSpkModel *model)
 {
+    if (model == nullptr) {
+       return;
+    }
     ((SpkModel *)model)->Unref();
 }
 
 VoskRecognizer *vosk_recognizer_new(VoskModel *model, float sample_rate)
 {
-    return (VoskRecognizer *)new KaldiRecognizer((Model *)model, sample_rate);
+    try {
+        return (VoskRecognizer *)new KaldiRecognizer((Model *)model, sample_rate);
+    } catch (...) {
+        return nullptr;
+    }
 }
 
 VoskRecognizer *vosk_recognizer_new_spk(VoskModel *model, float sample_rate, VoskSpkModel *spk_model)
 {
-    return (VoskRecognizer *)new KaldiRecognizer((Model *)model, sample_rate, (SpkModel *)spk_model);
+    try {
+        return (VoskRecognizer *)new KaldiRecognizer((Model *)model, sample_rate, (SpkModel *)spk_model);
+    } catch (...) {
+        return nullptr;
+    }
 }
 
 VoskRecognizer *vosk_recognizer_new_grm(VoskModel *model, float sample_rate, const char *grammar)
 {
-    return (VoskRecognizer *)new KaldiRecognizer((Model *)model, sample_rate, grammar);
+    try {
+        return (VoskRecognizer *)new KaldiRecognizer((Model *)model, sample_rate, grammar);
+    } catch (...) {
+        return nullptr;
+    }
 }
 
 void vosk_recognizer_set_max_alternatives(VoskRecognizer *recognizer, int max_alternatives)
@@ -77,22 +103,37 @@ void vosk_recognizer_set_words(VoskRecognizer *recognizer, int words)
 
 void vosk_recognizer_set_spk_model(VoskRecognizer *recognizer, VoskSpkModel *spk_model)
 {
+    if (recognizer == nullptr || spk_model == nullptr) {
+       return;
+    }
     ((KaldiRecognizer *)recognizer)->SetSpkModel((SpkModel *)spk_model);
 }
 
 int vosk_recognizer_accept_waveform(VoskRecognizer *recognizer, const char *data, int length)
 {
-    return ((KaldiRecognizer *)(recognizer))->AcceptWaveform(data, length);
+    try {
+        return ((KaldiRecognizer *)(recognizer))->AcceptWaveform(data, length);
+    } catch (...) {
+        return -1;
+    }
 }
 
 int vosk_recognizer_accept_waveform_s(VoskRecognizer *recognizer, const short *data, int length)
 {
-    return ((KaldiRecognizer *)(recognizer))->AcceptWaveform(data, length);
+    try {
+        return ((KaldiRecognizer *)(recognizer))->AcceptWaveform(data, length);
+    } catch (...) {
+        return -1;
+    }
 }
 
 int vosk_recognizer_accept_waveform_f(VoskRecognizer *recognizer, const float *data, int length)
 {
-    return ((KaldiRecognizer *)(recognizer))->AcceptWaveform(data, length);
+    try {
+        return ((KaldiRecognizer *)(recognizer))->AcceptWaveform(data, length);
+    } catch (...) {
+        return -1;
+    }
 }
 
 const char *vosk_recognizer_result(VoskRecognizer *recognizer)

src/vosk_api.h

@@ -43,7 +43,7 @@ typedef struct VoskRecognizer VoskRecognizer;
 /** Loads model data from the file and returns the model object
  *
  * @param model_path: the path of the model on the filesystem
- @ @returns model object */
+ * @returns model object or NULL if problem occured */
 VoskModel *vosk_model_new(const char *model_path);
 
 
@@ -66,7 +66,7 @@ int vosk_model_find_word(VoskModel *model, const char *word);
 /** Loads speaker model data from the file and returns the model object
  *
  * @param model_path: the path of the model on the filesystem
- * @returns model object */
+ * @returns model object or NULL if problem occured */
 VoskSpkModel *vosk_spk_model_new(const char *model_path);
 
 
@@ -79,9 +79,13 @@ void vosk_spk_model_free(VoskSpkModel *model);
 
 /** Creates the recognizer object
  *
- *  The recognizers process the speech and return text using shared model data 
- *  @param sample_rate The sample rate of the audio you going to feed into the recognizer
- *  @returns 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 sample_rate 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 */
 VoskRecognizer *vosk_recognizer_new(VoskModel *model, float sample_rate);
 
 
@@ -90,9 +94,13 @@ VoskRecognizer *vosk_recognizer_new(VoskModel *model, float sample_rate);
  *  With the speaker recognition mode the recognizer not just recognize
  *  text but also return speaker vectors one can use for speaker identification
  *
- *  @param sample_rate The sample rate of the audio you going to feed into the recognizer
+ *  @param model       VoskModel containing static data for recognizer. Model can be
+ *                     shared across recognizers, even running in different threads.
+ *  @param sample_rate 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 spk_model speaker model for speaker identification
- *  @returns recognizer object */
+ *  @returns recognizer object or NULL if problem occured */
 VoskRecognizer *vosk_recognizer_new_spk(VoskModel *model, float sample_rate, VoskSpkModel *spk_model);
 
 
@@ -106,11 +114,15 @@ VoskRecognizer *vosk_recognizer_new_spk(VoskModel *model, float sample_rate, Vos
  *  Only recognizers with lookahead models support this type of quick configuration.
  *  Precompiled HCLG graph models are not supported.
  *
- *  @param sample_rate The sample rate of the audio you going to feed into the recognizer
+ *  @param model       VoskModel containing static data for recognizer. Model can be
+ *                     shared across recognizers, even running in different threads.
+ *  @param sample_rate 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 */
+ *  @returns recognizer object or NULL if problem occured */
 VoskRecognizer *vosk_recognizer_new_grm(VoskModel *model, float sample_rate, const char *grammar);
 
 
@@ -181,7 +193,9 @@ void vosk_recognizer_set_words(VoskRecognizer *recognizer, int words);
  *
  *  @param data - audio data in PCM 16-bit mono format
  *  @param length - length of the audio data
- *  @returns true if silence is occured and you can retrieve a new utterance with result method */
+ *  @returns 1 if silence is occured and you can retrieve a new utterance with result method 
+ *           0 if decoding continues
+ *           -1 if exception occured */
 int vosk_recognizer_accept_waveform(VoskRecognizer *recognizer, const char *data, int length);
 
 

travis/Dockerfile.dockcross

@@ -24,7 +24,7 @@ RUN apt-get update && \
 ARG OPENBLAS_ARCH=ARMV7
 ARG ARM_HARDWARE_OPTS="-mfloat-abi=hard -mfpu=neon"
 RUN cd /opt \
-    && git clone -b lookahead-1.8.0 --single-branch https://github.com/alphacep/kaldi \
+    && git clone -b vosk --single-branch https://github.com/alphacep/kaldi \
     && cd kaldi/tools \
     && git clone -b v0.3.13 --single-branch https://github.com/xianyi/OpenBLAS \
     && git clone -b v3.2.1  --single-branch https://github.com/alphacep/clapack \

travis/Dockerfile.manylinux

@@ -12,7 +12,7 @@ RUN yum -y update && yum -y install \
     && yum clean all
 
 RUN cd /opt \
-    && git clone -b lookahead-1.8.0 --single-branch https://github.com/alphacep/kaldi \
+    && git clone -b vosk --single-branch https://github.com/alphacep/kaldi \
     && cd /opt/kaldi/tools \
     && git clone -b v0.3.13 --single-branch https://github.com/xianyi/OpenBLAS \
     && git clone -b v3.2.1  --single-branch https://github.com/alphacep/clapack \