Add node-addon-api for VAD (#864)

nodejs-addon-examples/.gitignore

@@ -0,0 +1 @@
+crash.log

nodejs-addon-examples/README.md

@@ -38,3 +38,12 @@ node ./test_asr_streaming_transducer.js
 
 node ./test_asr_streaming_transducer_microphone.js
 ```
+
+# VAD
+
+```bash
+wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/silero_vad.onnx
+
+node ./test_vad_microphone.js
+```
+

nodejs-addon-examples/test_asr_streaming_transducer.js

@@ -36,10 +36,10 @@ console.log('Started')
 let start = performance.now();
 const stream = recognizer.createStream();
 const wave = sherpa_onnx.readWave(waveFilename);
-stream.acceptWaveform(wave.samples, wave.sampleRate);
+stream.acceptWaveform({sampleRate: wave.sampleRate, samples: wave.samples});
 
 const tailPadding = new Float32Array(wave.sampleRate * 0.4);
-stream.acceptWaveform(tailPadding, wave.sampleRate);
+stream.acceptWaveform({samples: tailPadding, sampleRate: wave.sampleRate});
 
 while (recognizer.isReady(stream)) {
   recognizer.decode(stream);

nodejs-addon-examples/test_asr_streaming_transducer_microphone.js

@@ -60,7 +60,8 @@ const display = new sherpa_onnx.Display(50);
 ai.on('data', data => {
   const samples = new Float32Array(data.buffer);
 
-  stream.acceptWaveform(samples, recognizer.config.featConfig.sampleRate);
+  stream.acceptWaveform(
+      {sampleRate: recognizer.config.featConfig.sampleRate, samples: samples});
 
   while (recognizer.isReady(stream)) {
     recognizer.decode(stream);

nodejs-addon-examples/test_vad_microphone.js

@@ -0,0 +1,88 @@
+// Copyright (c)  2023-2024  Xiaomi Corporation (authors: Fangjun Kuang)
+
+const portAudio = require('naudiodon2');
+// console.log(portAudio.getDevices());
+
+const sherpa_onnx = require('sherpa-onnx-node');
+
+function createVad() {
+  // please download silero_vad.onnx from
+  // https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/silero_vad.onnx
+  const config = {
+    sileroVad: {
+      model: './silero_vad.onnx',
+      threshold: 0.5,
+      minSpeechDuration: 0.25,
+      minSilenceDuration: 0.5,
+      windowSize: 512,
+    },
+    sampleRate: 16000,
+    debug: true,
+    numThreads: 1,
+  };
+
+  const bufferSizeInSeconds = 60;
+
+  return new sherpa_onnx.Vad(config, bufferSizeInSeconds);
+}
+
+vad = createVad();
+
+const bufferSizeInSeconds = 30;
+const buffer =
+    new sherpa_onnx.CircularBuffer(bufferSizeInSeconds * vad.config.sampleRate);
+
+
+const ai = new portAudio.AudioIO({
+  inOptions: {
+    channelCount: 1,
+    closeOnError: true,  // Close the stream if an audio error is detected, if
+                         // set false then just log the error
+    deviceId: -1,  // Use -1 or omit the deviceId to select the default device
+    sampleFormat: portAudio.SampleFormatFloat32,
+    sampleRate: vad.config.sampleRate,
+  }
+});
+
+let printed = false;
+let index = 0;
+ai.on('data', data => {
+  const windowSize = vad.config.sileroVad.windowSize;
+  buffer.push(new Float32Array(data.buffer));
+  while (buffer.size() > windowSize) {
+    const samples = buffer.get(buffer.head(), windowSize);
+    buffer.pop(windowSize);
+    vad.acceptWaveform(samples)
+    if (vad.isDetected() && !printed) {
+      console.log(`${index}: Detected speech`)
+      printed = true;
+    }
+
+    if (!vad.isDetected()) {
+      printed = false;
+    }
+
+    while (!vad.isEmpty()) {
+      const segment = vad.front();
+      vad.pop();
+      const filename = `${index}-${
+          new Date()
+              .toLocaleTimeString('en-US', {hour12: false})
+              .split(' ')[0]}.wav`;
+      sherpa_onnx.writeWave(
+          filename,
+          {samples: segment.samples, sampleRate: vad.config.sampleRate})
+      const duration = segment.samples.length / vad.config.sampleRate;
+      console.log(`${index} End of speech. Duration: ${duration} seconds`);
+      console.log(`Saved to ${filename}`);
+      index += 1;
+    }
+  }
+});
+
+ai.on('close', () => {
+  console.log('Free resources');
+});
+
+ai.start();
+console.log('Started! Please speak')

python-api-examples/offline-tts-play.py

@@ -47,7 +47,7 @@
  --vits-model=./sherpa-onnx-vits-zh-ll/model.onnx \
  --vits-lexicon=./sherpa-onnx-vits-zh-ll/lexicon.txt \
  --vits-tokens=./sherpa-onnx-vits-zh-ll/tokens.txt \
- --tts-rule-fsts='./sherpa-onnx-vits-zh-ll/phone.fst,./sherpa-onnx-vits-zh-ll/date.fst,./sherpa-onnx-vits-zh-ll/number.fst' \
+ --tts-rule-fsts=./sherpa-onnx-vits-zh-ll/phone.fst,./sherpa-onnx-vits-zh-ll/date.fst,./sherpa-onnx-vits-zh-ll/number.fst \
  --vits-dict-dir=./sherpa-onnx-vits-zh-ll/dict \
  --sid=2 \
  --output-filename=./test-2.wav \

python-api-examples/offline-tts.py

@@ -48,7 +48,7 @@
  --vits-model=./sherpa-onnx-vits-zh-ll/model.onnx \
  --vits-lexicon=./sherpa-onnx-vits-zh-ll/lexicon.txt \
  --vits-tokens=./sherpa-onnx-vits-zh-ll/tokens.txt \
- --tts-rule-fsts='./sherpa-onnx-vits-zh-ll/phone.fst,./sherpa-onnx-vits-zh-ll/date.fst,./sherpa-onnx-vits-zh-ll/number.fst' \
+ --tts-rule-fsts=./sherpa-onnx-vits-zh-ll/phone.fst,./sherpa-onnx-vits-zh-ll/date.fst,./sherpa-onnx-vits-zh-ll/number.fst \
  --vits-dict-dir=./sherpa-onnx-vits-zh-ll/dict \
  --sid=2 \
  --output-filename=./test-2.wav \

scripts/node-addon-api/CMakeLists.txt

@@ -20,7 +20,9 @@ include_directories(${CMAKE_JS_INC})
 set(srcs
   src/sherpa-onnx-node-addon-api.cc
   src/streaming-asr.cc
+  src/vad.cc
   src/wave-reader.cc
+  src/wave-writer.cc
 )
 
 if(NOT DEFINED ENV{SHERPA_ONNX_INSTALL_DIR})

scripts/node-addon-api/lib/sherpa-onnx.js

@@ -1,8 +1,12 @@
 const addon = require('./addon.js')
 const streaming_asr = require('./streaming-asr.js');
+const vad = require('./vad.js');
 
 module.exports = {
   OnlineRecognizer: streaming_asr.OnlineRecognizer,
   readWave: addon.readWave,
+  writeWave: addon.writeWave,
   Display: streaming_asr.Display,
+  Vad: vad.Vad,
+  CircularBuffer: vad.CircularBuffer,
 }

scripts/node-addon-api/lib/streaming-asr.js

@@ -15,10 +15,11 @@ class OnlineStream {
     this.handle = handle;
   }
 
+  // obj is {samples: samples, sampleRate: sampleRate}
   // samples is a float32 array containing samples in the range [-1, 1]
-  acceptWaveform(samples, sampleRate) {
-    addon.acceptWaveformOnline(
-        this.handle, {samples: samples, sampleRate: sampleRate})
+  // sampleRate is a number
+  acceptWaveform(obj) {
+    addon.acceptWaveformOnline(this.handle, obj)
   }
 
   inputFinished() {

scripts/node-addon-api/lib/vad.js

@@ -0,0 +1,88 @@
+const addon = require('./addon.js');
+
+class CircularBuffer {
+  constructor(capacity) {
+    this.handle = addon.createCircularBuffer(capacity);
+  }
+
+  // samples is a float32 array
+  push(samples) {
+    addon.circularBufferPush(this.handle, samples);
+  }
+
+  // return a float32 array
+  get(startIndex, n) {
+    return addon.circularBufferGet(this.handle, startIndex, n);
+  }
+
+  pop(n) {
+    return addon.circularBufferPop(this.handle, n);
+  }
+
+  size() {
+    return addon.circularBufferSize(this.handle);
+  }
+
+  head() {
+    return addon.circularBufferHead(this.handle);
+  }
+
+  reset() {
+    return addon.circularBufferReset(this.handle);
+  }
+}
+
+class Vad {
+  /*
+config = {
+  sileroVad: {
+    model: "./silero_vad.onnx",
+    threshold: 0.5,
+  }
+}
+   */
+  constructor(config, bufferSizeInSeconds) {
+    this.handle =
+        addon.createVoiceActivityDetector(config, bufferSizeInSeconds);
+    this.config = config;
+  }
+
+  acceptWaveform(samples) {
+    addon.voiceActivityDetectorAcceptWaveform(this.handle, samples)
+  }
+
+  isEmpty() {
+    return addon.voiceActivityDetectorIsEmpty(this.handle)
+  }
+
+  isDetected() {
+    return addon.voiceActivityDetectorIsDetected(this.handle)
+  }
+
+  pop() {
+    addon.voiceActivityDetectorPop(this.handle)
+  }
+
+  clear() {
+    addon.VoiceActivityDetectorClearWrapper(this.handle)
+  }
+
+  /*
+{
+  samples: a 1-d float32 array,
+  start: a int32
+}
+   */
+  front() {
+    return addon.voiceActivityDetectorFront(this.handle)
+  }
+
+  reset() {
+    return addon.VoiceActivityDetectorResetWrapper(this.handle)
+  }
+}
+
+module.exports = {
+  Vad,
+  CircularBuffer,
+}

scripts/node-addon-api/src/sherpa-onnx-node-addon-api.cc

@@ -5,10 +5,14 @@
 
 void InitStreamingAsr(Napi::Env env, Napi::Object exports);
 void InitWaveReader(Napi::Env env, Napi::Object exports);
+void InitWaveWriter(Napi::Env env, Napi::Object exports);
+void InitVad(Napi::Env env, Napi::Object exports);
 
 Napi::Object Init(Napi::Env env, Napi::Object exports) {
   InitStreamingAsr(env, exports);
   InitWaveReader(env, exports);
+  InitWaveWriter(env, exports);
+  InitVad(env, exports);
 
   return exports;
 }

scripts/node-addon-api/src/streaming-asr.cc

@@ -125,8 +125,13 @@ static SherpaOnnxOnlineModelConfig GetOnlineModelConfig(Napi::Object obj) {
     config.provider = p;
   }
 
-  if (o.Has("debug") && o.Get("debug").IsNumber()) {
-    config.debug = o.Get("debug").As<Napi::Number>().Int32Value();
+  if (o.Has("debug") &&
+      (o.Get("debug").IsNumber() || o.Get("debug").IsBoolean())) {
+    if (o.Get("debug").IsBoolean()) {
+      config.debug = o.Get("debug").As<Napi::Boolean>().Value();
+    } else {
+      config.debug = o.Get("debug").As<Napi::Number>().Int32Value();
+    }
   }
 
   if (o.Has("modelType") && o.Get("modelType").IsString()) {