Fix modified beam search for iOS and android (#76)

* Use Int type for sampling rate

* Fix swift

* Fix iOS

.github/scripts/.gitignore

@@ -1,2 +1,3 @@
 Makefile
 *.jar
+hs_err_pid*.log

.github/scripts/Main.kt

@@ -4,7 +4,7 @@ import android.content.res.AssetManager
 
 fun main() {
     var featConfig = FeatureConfig(
-        sampleRate = 16000.0f,
+        sampleRate = 16000,
         featureDim = 80,
     )
 
@@ -13,7 +13,7 @@ fun main() {
         decoder = "./sherpa-onnx-streaming-zipformer-en-2023-02-21/decoder-epoch-99-avg-1.onnx",
         joiner = "./sherpa-onnx-streaming-zipformer-en-2023-02-21/joiner-epoch-99-avg-1.onnx",
         tokens = "./sherpa-onnx-streaming-zipformer-en-2023-02-21/tokens.txt",
-        numThreads = 4,
+        numThreads = 1,
         debug = false,
     )
 
@@ -24,22 +24,31 @@ fun main() {
         featConfig = featConfig,
         endpointConfig = endpointConfig,
         enableEndpoint = true,
+        decodingMethod = "greedy_search",
+        maxActivePaths = 4,
     )
 
     var model = SherpaOnnx(
         assetManager = AssetManager(),
         config = config,
     )
+
     var samples = WaveReader.readWave(
         assetManager = AssetManager(),
         filename = "./sherpa-onnx-streaming-zipformer-en-2023-02-21/test_wavs/1089-134686-0001.wav",
     )
 
-    model.decodeSamples(samples!!)
+    model.acceptWaveform(samples!!, sampleRate=16000)
+    while (model.isReady()) {
+      model.decode()
+    }
 
     var tail_paddings = FloatArray(8000) // 0.5 seconds
-    model.decodeSamples(tail_paddings)
+    model.acceptWaveform(tail_paddings, sampleRate=16000)
-
     model.inputFinished()
+    while (model.isReady()) {
+      model.decode()
+    }
+
     println("results: ${model.text}")
 }

.gitignore

@@ -38,3 +38,4 @@ log.txt
 tags
 run-decode-file-python.sh
 android/SherpaOnnx/app/src/main/assets/
+*.ncnn.*

android/SherpaOnnx/app/src/main/java/com/k2fsa/sherpa/onnx/MainActivity.kt

@@ -121,7 +121,10 @@ class MainActivity : AppCompatActivity() {
             val ret = audioRecord?.read(buffer, 0, buffer.size)
             if (ret != null && ret > 0) {
                 val samples = FloatArray(ret) { buffer[it] / 32768.0f }
-                model.decodeSamples(samples)
+                model.acceptWaveform(samples, sampleRate=16000)
+                while (model.isReady()) {
+                    model.decode()
+                }
                 runOnUiThread {
                     val isEndpoint = model.isEndpoint()
                     val text = model.text
@@ -177,33 +180,17 @@ class MainActivity : AppCompatActivity() {
         val type = 0
         println("Select model type ${type}")
         val config = OnlineRecognizerConfig(
-            featConfig = getFeatureConfig(sampleRate = 16000.0f, featureDim = 80),
+            featConfig = getFeatureConfig(sampleRate = 16000, featureDim = 80),
             modelConfig = getModelConfig(type = type)!!,
             endpointConfig = getEndpointConfig(),
-            enableEndpoint = true
+            enableEndpoint = true,
+            decodingMethod = "greedy_search",
+            maxActivePaths = 4,
         )
 
         model = SherpaOnnx(
             assetManager = application.assets,
             config = config,
         )
-       /*
-        println("reading samples")
-        val samples = WaveReader.readWave(
-            assetManager = application.assets,
-            // filename = "sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20/test_wavs/0.wav",
-            filename = "sherpa-onnx-lstm-zh-2023-02-20/test_wavs/0.wav",
-            // filename="sherpa-onnx-lstm-en-2023-02-17/test_wavs/1089-134686-0001.wav"
-        )
-        println("samples read done!")
-
-        model.decodeSamples(samples!!)
-
-        val tailPaddings = FloatArray(8000) // 0.5 seconds
-        model.decodeSamples(tailPaddings)
-
-        println("result is: ${model.text}")
-        model.reset()
-        */
     }
 }

android/SherpaOnnx/app/src/main/java/com/k2fsa/sherpa/onnx/SherpaOnnx.kt

@@ -24,7 +24,7 @@ data class OnlineTransducerModelConfig(
 )
 
 data class FeatureConfig(
-    var sampleRate: Float = 16000.0f,
+    var sampleRate: Int = 16000,
     var featureDim: Int = 80,
 )
 
@@ -32,7 +32,9 @@ data class OnlineRecognizerConfig(
     var featConfig: FeatureConfig = FeatureConfig(),
     var modelConfig: OnlineTransducerModelConfig,
     var endpointConfig: EndpointConfig = EndpointConfig(),
-    var enableEndpoint: Boolean,
+    var enableEndpoint: Boolean = true,
+    var decodingMethod: String = "greedy_search",
+    var maxActivePaths: Int = 4,
 )
 
 class SherpaOnnx(
@@ -49,12 +51,14 @@ class SherpaOnnx(
     }
 
 
-    fun decodeSamples(samples: FloatArray) =
+    fun acceptWaveform(samples: FloatArray, sampleRate: Int) =
-        decodeSamples(ptr, samples, sampleRate = config.featConfig.sampleRate)
+        acceptWaveform(ptr, samples, sampleRate)
 
     fun inputFinished() = inputFinished(ptr)
     fun reset() = reset(ptr)
+    fun decode() = decode(ptr)
     fun isEndpoint(): Boolean = isEndpoint(ptr)
+    fun isReady(): Boolean = isReady(ptr)
 
     val text: String
         get() = getText(ptr)
@@ -66,11 +70,13 @@ class SherpaOnnx(
         config: OnlineRecognizerConfig,
     ): Long
 
-    private external fun decodeSamples(ptr: Long, samples: FloatArray, sampleRate: Float)
+    private external fun acceptWaveform(ptr: Long, samples: FloatArray, sampleRate: Int)
     private external fun inputFinished(ptr: Long)
     private external fun getText(ptr: Long): String
     private external fun reset(ptr: Long)
+    private external fun decode(ptr: Long)
     private external fun isEndpoint(ptr: Long): Boolean
+    private external fun isReady(ptr: Long): Boolean
 
     companion object {
         init {
@@ -79,7 +85,7 @@ class SherpaOnnx(
     }
 }
 
-fun getFeatureConfig(sampleRate: Float, featureDim: Int): FeatureConfig {
+fun getFeatureConfig(sampleRate: Int, featureDim: Int): FeatureConfig {
     return FeatureConfig(sampleRate=sampleRate, featureDim=featureDim)
 }
 

ios-swift/SherpaOnnx/SherpaOnnx/ViewController.swift

@@ -23,10 +23,10 @@ extension AVAudioPCMBuffer {
 class ViewController: UIViewController {
     @IBOutlet weak var resultLabel: UILabel!
     @IBOutlet weak var recordBtn: UIButton!
-    
+
     var audioEngine: AVAudioEngine? = nil
     var recognizer: SherpaOnnxRecognizer! = nil
-    
+
     /// It saves the decoded results so far
     var sentences: [String] = [] {
         didSet {
@@ -42,7 +42,7 @@ class ViewController: UIViewController {
         if sentences.isEmpty {
             return "0: \(lastSentence.lowercased())"
         }
-        
+
         let start = max(sentences.count - maxSentence, 0)
         if lastSentence.isEmpty {
             return sentences.enumerated().map { (index, s) in "\(index): \(s.lowercased())" }[start...]
@@ -52,23 +52,23 @@ class ViewController: UIViewController {
                 .joined(separator: "\n") + "\n\(sentences.count): \(lastSentence.lowercased())"
         }
     }
-    
+
     func updateLabel() {
         DispatchQueue.main.async {
             self.resultLabel.text = self.results
         }
     }
-    
+
     override func viewDidLoad() {
         super.viewDidLoad()
         // Do any additional setup after loading the view.
-        
+
         resultLabel.text = "ASR with Next-gen Kaldi\n\nSee https://github.com/k2-fsa/sherpa-onnx\n\nPress the Start button to run!"
         recordBtn.setTitle("Start", for: .normal)
         initRecognizer()
         initRecorder()
     }
-    
+
     @IBAction func onRecordBtnClick(_ sender: UIButton) {
         if recordBtn.currentTitle == "Start" {
             startRecorder()
@@ -78,30 +78,32 @@ class ViewController: UIViewController {
             recordBtn.setTitle("Start", for: .normal)
         }
     }
-    
+
     func initRecognizer() {
         // Please select one model that is best suitable for you.
         //
         // You can also modify Model.swift to add new pre-trained models from
         // https://k2-fsa.github.io/sherpa/ncnn/pretrained_models/index.html
-        
+
         let modelConfig = getBilingualStreamZhEnZipformer20230220()
-        
+
         let featConfig = sherpaOnnxFeatureConfig(
             sampleRate: 16000,
             featureDim: 80)
-        
+
         var config = sherpaOnnxOnlineRecognizerConfig(
             featConfig: featConfig,
             modelConfig: modelConfig,
             enableEndpoint: true,
             rule1MinTrailingSilence: 2.4,
             rule2MinTrailingSilence: 0.8,
-            rule3MinUtteranceLength: 30
+            rule3MinUtteranceLength: 30,
+            decodingMethod: "greedy_search",
+            maxActivePaths: 4
         )
         recognizer = SherpaOnnxRecognizer(config: &config)
     }
-    
+
     func initRecorder() {
         print("init recorder")
         audioEngine = AVAudioEngine()
@@ -112,9 +114,9 @@ class ViewController: UIViewController {
             commonFormat: .pcmFormatFloat32,
             sampleRate: 16000, channels: 1,
             interleaved: false)!
-        
+
         let converter = AVAudioConverter(from: inputFormat!, to: outputFormat)!
-        
+
         inputNode!.installTap(
             onBus: bus,
             bufferSize: 1024,
@@ -122,34 +124,34 @@ class ViewController: UIViewController {
         ) {
             (buffer: AVAudioPCMBuffer, when: AVAudioTime) in
             var newBufferAvailable = true
-            
+
             let inputCallback: AVAudioConverterInputBlock = {
                 inNumPackets, outStatus in
                 if newBufferAvailable {
                     outStatus.pointee = .haveData
                     newBufferAvailable = false
-                    
+
                     return buffer
                 } else {
                     outStatus.pointee = .noDataNow
                     return nil
                 }
             }
-            
+
             let convertedBuffer = AVAudioPCMBuffer(
                 pcmFormat: outputFormat,
                 frameCapacity:
                     AVAudioFrameCount(outputFormat.sampleRate)
                 * buffer.frameLength
                 / AVAudioFrameCount(buffer.format.sampleRate))!
-            
+
             var error: NSError?
             let _ = converter.convert(
                 to: convertedBuffer,
                 error: &error, withInputFrom: inputCallback)
-            
+
             // TODO(fangjun): Handle status != haveData
-            
+
             let array = convertedBuffer.array()
             if !array.isEmpty {
                 self.recognizer.acceptWaveform(samples: array)
@@ -158,13 +160,13 @@ class ViewController: UIViewController {
                 }
                 let isEndpoint = self.recognizer.isEndpoint()
                 let text = self.recognizer.getResult().text
-                
+
                 if !text.isEmpty && self.lastSentence != text {
                     self.lastSentence = text
                     self.updateLabel()
                     print(text)
                 }
-                
+
                 if isEndpoint {
                     if !text.isEmpty {
                         let tmp = self.lastSentence
@@ -175,13 +177,13 @@ class ViewController: UIViewController {
                 }
             }
         }
-        
+
     }
-    
+
     func startRecorder() {
         lastSentence = ""
         sentences = []
-        
+
         do {
             try self.audioEngine?.start()
         } catch let error as NSError {
@@ -189,7 +191,7 @@ class ViewController: UIViewController {
         }
         print("started")
     }
-    
+
     func stopRecorder() {
         audioEngine?.stop()
         print("stopped")

sherpa-onnx/c-api/c-api.cc

@@ -76,7 +76,7 @@ SherpaOnnxOnlineStream *CreateOnlineStream(
 
 void DestoryOnlineStream(SherpaOnnxOnlineStream *stream) { delete stream; }
 
-void AcceptWaveform(SherpaOnnxOnlineStream *stream, float sample_rate,
+void AcceptWaveform(SherpaOnnxOnlineStream *stream, int32_t sample_rate,
                     const float *samples, int32_t n) {
   stream->impl->AcceptWaveform(sample_rate, samples, n);
 }

sherpa-onnx/c-api/c-api.h

@@ -120,7 +120,7 @@ void DestoryOnlineStream(SherpaOnnxOnlineStream *stream);
 /// @param samples A pointer to a 1-D array containing audio samples.
 ///                The range of samples has to be normalized to [-1, 1].
 /// @param n  Number of elements in the samples array.
-void AcceptWaveform(SherpaOnnxOnlineStream *stream, float sample_rate,
+void AcceptWaveform(SherpaOnnxOnlineStream *stream, int32_t sample_rate,
                     const float *samples, int32_t n);
 
 /// Return 1 if there are enough number of feature frames for decoding.

sherpa-onnx/csrc/features.cc

@@ -48,7 +48,7 @@ class FeatureExtractor::Impl {
     fbank_ = std::make_unique<knf::OnlineFbank>(opts_);
   }
 
-  void AcceptWaveform(float sampling_rate, const float *waveform, int32_t n) {
+  void AcceptWaveform(int32_t sampling_rate, const float *waveform, int32_t n) {
     std::lock_guard<std::mutex> lock(mutex_);
     fbank_->AcceptWaveform(sampling_rate, waveform, n);
   }
@@ -107,7 +107,7 @@ FeatureExtractor::FeatureExtractor(const FeatureExtractorConfig &config /*={}*/)
 
 FeatureExtractor::~FeatureExtractor() = default;
 
-void FeatureExtractor::AcceptWaveform(float sampling_rate,
+void FeatureExtractor::AcceptWaveform(int32_t sampling_rate,
                                       const float *waveform, int32_t n) {
   impl_->AcceptWaveform(sampling_rate, waveform, n);
 }

sherpa-onnx/csrc/features.h

@@ -14,7 +14,7 @@
 namespace sherpa_onnx {
 
 struct FeatureExtractorConfig {
-  float sampling_rate = 16000;
+  int32_t sampling_rate = 16000;
   int32_t feature_dim = 80;
   int32_t max_feature_vectors = -1;
 
@@ -34,7 +34,7 @@ class FeatureExtractor {
      @param waveform Pointer to a 1-D array of size n
      @param n Number of entries in waveform
    */
-  void AcceptWaveform(float sampling_rate, const float *waveform, int32_t n);
+  void AcceptWaveform(int32_t sampling_rate, const float *waveform, int32_t n);
 
   /**
    * InputFinished() tells the class you won't be providing any

sherpa-onnx/csrc/sherpa-onnx-microphone.cc

@@ -112,7 +112,7 @@ for a list of pre-trained models to download.
 
   param.suggestedLatency = info->defaultLowInputLatency;
   param.hostApiSpecificStreamInfo = nullptr;
-  const float sample_rate = 16000;
+  float sample_rate = 16000;
 
   PaStream *stream;
   PaError err =

sherpa-onnx/csrc/sherpa-onnx.cc

@@ -61,7 +61,7 @@ for a list of pre-trained models to download.
 
   sherpa_onnx::OnlineRecognizer recognizer(config);
 
-  float expected_sampling_rate = config.feat_config.sampling_rate;
+  int32_t expected_sampling_rate = config.feat_config.sampling_rate;
 
   bool is_ok = false;
   std::vector<float> samples =
@@ -72,7 +72,7 @@ for a list of pre-trained models to download.
     return -1;
   }
 
-  float duration = samples.size() / expected_sampling_rate;
+  float duration = samples.size() / static_cast<float>(expected_sampling_rate);
 
   fprintf(stderr, "wav filename: %s\n", wav_filename.c_str());
   fprintf(stderr, "wav duration (s): %.3f\n", duration);

sherpa-onnx/jni/jni.cc

@@ -40,19 +40,18 @@ class SherpaOnnx {
             mgr,
 #endif
             config),
-        stream_(recognizer_.CreateStream()),
+        stream_(recognizer_.CreateStream()) {
-        tail_padding_(16000 * 0.32, 0) {
   }
 
-  void DecodeSamples(float sample_rate, const float *samples, int32_t n) const {
+  void AcceptWaveform(int32_t sample_rate, const float *samples,
+                      int32_t n) const {
     stream_->AcceptWaveform(sample_rate, samples, n);
-    Decode();
   }
 
   void InputFinished() const {
-    stream_->AcceptWaveform(16000, tail_padding_.data(), tail_padding_.size());
+    std::vector<float> tail_padding(16000 * 0.32, 0);
+    stream_->AcceptWaveform(16000, tail_padding.data(), tail_padding.size());
     stream_->InputFinished();
-    Decode();
   }
 
   const std::string GetText() const {
@@ -62,19 +61,15 @@ class SherpaOnnx {
 
   bool IsEndpoint() const { return recognizer_.IsEndpoint(stream_.get()); }
 
+  bool IsReady() const { return recognizer_.IsReady(stream_.get()); }
+
   void Reset() const { return recognizer_.Reset(stream_.get()); }
 
- private:
+  void Decode() const { recognizer_.DecodeStream(stream_.get()); }
-  void Decode() const {
-    while (recognizer_.IsReady(stream_.get())) {
-      recognizer_.DecodeStream(stream_.get());
-    }
-  }
 
  private:
   sherpa_onnx::OnlineRecognizer recognizer_;
   std::unique_ptr<sherpa_onnx::OnlineStream> stream_;
-  std::vector<float> tail_padding_;
 };
 
 static OnlineRecognizerConfig GetConfig(JNIEnv *env, jobject config) {
@@ -86,14 +81,24 @@ static OnlineRecognizerConfig GetConfig(JNIEnv *env, jobject config) {
   // https://docs.oracle.com/javase/7/docs/technotes/guides/jni/spec/types.html
   // https://courses.cs.washington.edu/courses/cse341/99wi/java/tutorial/native1.1/implementing/field.html
 
+  //---------- decoding ----------
+  fid = env->GetFieldID(cls, "decodingMethod", "Ljava/lang/String;");
+  jstring s = (jstring)env->GetObjectField(config, fid);
+  const char *p = env->GetStringUTFChars(s, nullptr);
+  ans.decoding_method = p;
+  env->ReleaseStringUTFChars(s, p);
+
+  fid = env->GetFieldID(cls, "maxActivePaths", "I");
+  ans.max_active_paths = env->GetIntField(config, fid);
+
   //---------- feat config ----------
   fid = env->GetFieldID(cls, "featConfig",
                         "Lcom/k2fsa/sherpa/onnx/FeatureConfig;");
   jobject feat_config = env->GetObjectField(config, fid);
   jclass feat_config_cls = env->GetObjectClass(feat_config);
 
-  fid = env->GetFieldID(feat_config_cls, "sampleRate", "F");
+  fid = env->GetFieldID(feat_config_cls, "sampleRate", "I");
-  ans.feat_config.sampling_rate = env->GetFloatField(feat_config, fid);
+  ans.feat_config.sampling_rate = env->GetIntField(feat_config, fid);
 
   fid = env->GetFieldID(feat_config_cls, "featureDim", "I");
   ans.feat_config.feature_dim = env->GetIntField(feat_config, fid);
@@ -153,8 +158,8 @@ static OnlineRecognizerConfig GetConfig(JNIEnv *env, jobject config) {
   jclass model_config_cls = env->GetObjectClass(model_config);
 
   fid = env->GetFieldID(model_config_cls, "encoder", "Ljava/lang/String;");
-  jstring s = (jstring)env->GetObjectField(model_config, fid);
+  s = (jstring)env->GetObjectField(model_config, fid);
-  const char *p = env->GetStringUTFChars(s, nullptr);
+  p = env->GetStringUTFChars(s, nullptr);
   ans.model_config.encoder_filename = p;
   env->ReleaseStringUTFChars(s, p);
 
@@ -198,6 +203,7 @@ JNIEXPORT jlong JNICALL Java_com_k2fsa_sherpa_onnx_SherpaOnnx_new(
 #endif
 
   auto config = sherpa_onnx::GetConfig(env, _config);
+  SHERPA_ONNX_LOGE("config:\n%s", config.ToString().c_str());
   auto model = new sherpa_onnx::SherpaOnnx(
 #if __ANDROID_API__ >= 9
       mgr,
@@ -220,6 +226,13 @@ JNIEXPORT void JNICALL Java_com_k2fsa_sherpa_onnx_SherpaOnnx_reset(
   model->Reset();
 }
 
+SHERPA_ONNX_EXTERN_C
+JNIEXPORT bool JNICALL Java_com_k2fsa_sherpa_onnx_SherpaOnnx_isReady(
+    JNIEnv *env, jobject /*obj*/, jlong ptr) {
+  auto model = reinterpret_cast<sherpa_onnx::SherpaOnnx *>(ptr);
+  return model->IsReady();
+}
+
 SHERPA_ONNX_EXTERN_C
 JNIEXPORT bool JNICALL Java_com_k2fsa_sherpa_onnx_SherpaOnnx_isEndpoint(
     JNIEnv *env, jobject /*obj*/, jlong ptr) {
@@ -228,15 +241,22 @@ JNIEXPORT bool JNICALL Java_com_k2fsa_sherpa_onnx_SherpaOnnx_isEndpoint(
 }
 
 SHERPA_ONNX_EXTERN_C
-JNIEXPORT void JNICALL Java_com_k2fsa_sherpa_onnx_SherpaOnnx_decodeSamples(
+JNIEXPORT void JNICALL Java_com_k2fsa_sherpa_onnx_SherpaOnnx_decode(
+    JNIEnv *env, jobject /*obj*/, jlong ptr) {
+  auto model = reinterpret_cast<sherpa_onnx::SherpaOnnx *>(ptr);
+  model->Decode();
+}
+
+SHERPA_ONNX_EXTERN_C
+JNIEXPORT void JNICALL Java_com_k2fsa_sherpa_onnx_SherpaOnnx_acceptWaveform(
     JNIEnv *env, jobject /*obj*/, jlong ptr, jfloatArray samples,
-    jfloat sample_rate) {
+    jint sample_rate) {
   auto model = reinterpret_cast<sherpa_onnx::SherpaOnnx *>(ptr);
 
   jfloat *p = env->GetFloatArrayElements(samples, nullptr);
   jsize n = env->GetArrayLength(samples);
 
-  model->DecodeSamples(sample_rate, p, n);
+  model->AcceptWaveform(sample_rate, p, n);
 
   env->ReleaseFloatArrayElements(samples, p, JNI_ABORT);
 }

swift-api-examples/SherpaOnnx.swift

@@ -62,11 +62,15 @@ func sherpaOnnxOnlineRecognizerConfig(
     enableEndpoint: Bool = false,
     rule1MinTrailingSilence: Float = 2.4,
     rule2MinTrailingSilence: Float = 1.2,
-    rule3MinUtteranceLength: Float = 30
+    rule3MinUtteranceLength: Float = 30,
+    decodingMethod: String = "greedy_search",
+    maxActivePaths: Int = 4
 ) ->  SherpaOnnxOnlineRecognizerConfig{
   return SherpaOnnxOnlineRecognizerConfig(
     feat_config: featConfig,
     model_config: modelConfig,
+    decoding_method: toCPointer(decodingMethod),
+    max_active_paths: Int32(maxActivePaths),
     enable_endpoint: enableEndpoint ? 1 : 0,
     rule1_min_trailing_silence: rule1MinTrailingSilence,
     rule2_min_trailing_silence: rule2MinTrailingSilence,
@@ -128,12 +132,12 @@ class SherpaOnnxRecognizer {
   /// Decode wave samples.
   ///
   /// - Parameters:
-  ///   - samples: Audio samples normalzed to the range [-1, 1]
+  ///   - samples: Audio samples normalized to the range [-1, 1]
   ///   - sampleRate: Sample rate of the input audio samples. Must match
   ///                 the one expected by the model. It must be 16000 for
   ///                 models from icefall.
-  func acceptWaveform(samples: [Float], sampleRate: Float = 16000) {
+  func acceptWaveform(samples: [Float], sampleRate: Int = 16000) {
-    AcceptWaveform(stream, sampleRate, samples, Int32(samples.count))
+    AcceptWaveform(stream, Int32(sampleRate), samples, Int32(samples.count))
   }
 
   func isReady() -> Bool {

swift-api-examples/decode-file.swift

@@ -32,7 +32,9 @@ func run() {
   var config = sherpaOnnxOnlineRecognizerConfig(
       featConfig: featConfig,
       modelConfig: modelConfig,
-      enableEndpoint: false
+      enableEndpoint: false,
+      decodingMethod: "modified_beam_search",
+      maxActivePaths: 4
   )