Add Dart API for VAD (#904)

@@ -123,6 +123,7 @@ jobs:
           pushd example/assets
 
           curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-recongition-models/3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx
+          curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/silero_vad.onnx
           git clone https://github.com/csukuangfj/sr-data
 
           rm -rf sr-data/.git

@@ -143,3 +143,5 @@ xcuserdata/
 ## Xcode 8 and earlier
 *.xcscmblueprint
 *.xccheckout
+
+flutter_jank_metrics*.json

@@ -6,6 +6,7 @@
 # switch to this directory and run
 
 wget https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-recongition-models/3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx
+wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/silero_vad.onnx
 git clone https://github.com/csukuangfj/sr-data
 
 rm -rf sr-data/.git

 // Copyright (c)  2024  Xiaomi Corporation
 import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;
 import 'package:flutter/material.dart';
+
 import "./speaker_identification_test.dart";
+import "./vad_test.dart";
 
 void main() {
   runApp(const MyApp());
@@ -51,6 +53,7 @@ class _MyHomePageState extends State<MyHomePage> {
     if (_counter <= 10) {
       sherpa_onnx.initBindings();
       await testSpeakerID();
+      // await testVad();
     }
 
     setState(() {

+// Copyright (c)  2024  Xiaomi Corporation
+import 'dart:typed_data';
+import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;
+import './utils.dart';
+
+Future<void> testVad() async {
+  final src = 'assets/silero_vad.onnx';
+  final modelPath = await copyAssetFile(src: src, dst: 'silero_vad.onnx');
+
+  final sileroVadConfig = sherpa_onnx.SileroVadModelConfig(model: modelPath);
+  final config = sherpa_onnx.VadModelConfig(
+    sileroVad: sileroVadConfig,
+    numThreads: 1,
+    debug: true,
+  );
+
+  final vad = sherpa_onnx.VoiceActivityDetector(
+      config: config, bufferSizeInSeconds: 10);
+  print('before vad.free(): ${vad.ptr}');
+  vad.free();
+  print('after vad.free(): ${vad.ptr}');
+
+  final buffer = sherpa_onnx.CircularBuffer(capacity: 16000 * 2);
+
+  final d = Float32List.fromList([0, 10, 20, 30]);
+  buffer.push(d);
+  assert(d.length == buffer.size, '${d.length} vs ${buffer.size}');
+
+  final f = Float32List.fromList([-5, 100.25, 599]);
+  buffer.push(f);
+
+  assert(buffer.size == d.length + f.length);
+  final g = buffer.get(startIndex: 0, n: 5);
+
+  assert(g.length == 5);
+  assert(g[0] == 0);
+  assert(g[1] == 10);
+  assert(g[2] == 20);
+  assert(g[3] == 30);
+  assert(g[4] == -5);
+
+  assert(buffer.size == d.length + f.length);
+
+  buffer.pop(3);
+  assert(buffer.size == d.length + f.length - 3);
+
+  final h = buffer.get(startIndex: buffer.head, n: 4);
+  assert(h.length == 4);
+  assert(h[0] == 30);
+  assert(h[1] == -5);
+  assert(h[2] == 100.25);
+  assert(h[3] == 599);
+
+  buffer.reset();
+
+  assert(buffer.size == 0);
+  assert(buffer.head == 0);
+
+  print('before free: ${buffer.ptr}');
+  buffer.free();
+  print('after free: ${buffer.ptr}');
+}

@@ -2,10 +2,11 @@
 import 'dart:io';
 import 'dart:ffi';
 
-import 'src/sherpa_onnx_bindings.dart';
-export 'src/speaker_identification.dart';
 export 'src/online_stream.dart';
+export 'src/speaker_identification.dart';
+export 'src/vad.dart';
 export 'src/wave_reader.dart';
+import 'src/sherpa_onnx_bindings.dart';
 
 final DynamicLibrary _dylib = () {
   if (Platform.isIOS) {

 // Copyright (c)  2024  Xiaomi Corporation
-import 'dart:typed_data';
 import 'dart:ffi';
+import 'dart:typed_data';
 import 'package:ffi/ffi.dart';
-import "./sherpa_onnx_bindings.dart";
+
+import './sherpa_onnx_bindings.dart';
 
 class OnlineStream {
   /// The user has to call OnlineStream.free() to avoid memory leak.

@@ -2,6 +2,47 @@
 import 'dart:ffi';
 import 'package:ffi/ffi.dart';
 
+final class SherpaOnnxSileroVadModelConfig extends Struct {
+  external Pointer<Utf8> model;
+
+  @Float()
+  external double threshold;
+
+  @Float()
+  external double minSilenceDuration;
+
+  @Float()
+  external double minSpeechDuration;
+
+  @Int32()
+  external int windowSize;
+}
+
+final class SherpaOnnxVadModelConfig extends Struct {
+  external SherpaOnnxSileroVadModelConfig sileroVad;
+
+  @Int32()
+  external int sampleRate;
+
+  @Int32()
+  external int numThreads;
+
+  external Pointer<Utf8> provider;
+
+  @Int32()
+  external int debug;
+}
+
+final class SherpaOnnxSpeechSegment extends Struct {
+  @Int32()
+  external int start;
+
+  external Pointer<Float> samples;
+
+  @Int32()
+  external int n;
+}
+
 final class SherpaOnnxWave extends Struct {
   external Pointer<Float> samples;
 
@@ -24,17 +65,136 @@ final class SherpaOnnxSpeakerEmbeddingExtractorConfig extends Struct {
   external Pointer<Utf8> provider;
 }
 
+final class SherpaOnnxCircularBuffer extends Opaque {}
+
+final class SherpaOnnxVoiceActivityDetector extends Opaque {}
+
 final class SherpaOnnxOnlineStream extends Opaque {}
 
 final class SherpaOnnxSpeakerEmbeddingExtractor extends Opaque {}
 
 final class SherpaOnnxSpeakerEmbeddingManager extends Opaque {}
 
+typedef SherpaOnnxCreateVoiceActivityDetectorNative
+    = Pointer<SherpaOnnxVoiceActivityDetector> Function(
+        Pointer<SherpaOnnxVadModelConfig>, Float);
+
+typedef SherpaOnnxCreateVoiceActivityDetector
+    = Pointer<SherpaOnnxVoiceActivityDetector> Function(
+        Pointer<SherpaOnnxVadModelConfig>, double);
+
+typedef SherpaOnnxDestroyVoiceActivityDetectorNative = Void Function(
+    Pointer<SherpaOnnxVoiceActivityDetector>);
+
+typedef SherpaOnnxDestroyVoiceActivityDetector = void Function(
+    Pointer<SherpaOnnxVoiceActivityDetector>);
+
+typedef SherpaOnnxVoiceActivityDetectorAcceptWaveformNative = Void Function(
+    Pointer<SherpaOnnxVoiceActivityDetector>, Pointer<Float>, Int32);
+
+typedef SherpaOnnxVoiceActivityDetectorAcceptWaveform = void Function(
+    Pointer<SherpaOnnxVoiceActivityDetector>, Pointer<Float>, int);
+
+typedef SherpaOnnxVoiceActivityDetectorEmptyNative = Int32 Function(
+    Pointer<SherpaOnnxVoiceActivityDetector>);
+
+typedef SherpaOnnxVoiceActivityDetectorEmpty = int Function(
+    Pointer<SherpaOnnxVoiceActivityDetector>);
+
+typedef SherpaOnnxVoiceActivityDetectorDetectedNative = Int32 Function(
+    Pointer<SherpaOnnxVoiceActivityDetector>);
+
+typedef SherpaOnnxVoiceActivityDetectorDetected = int Function(
+    Pointer<SherpaOnnxVoiceActivityDetector>);
+
+typedef SherpaOnnxVoiceActivityDetectorPopNative = Void Function(
+    Pointer<SherpaOnnxVoiceActivityDetector>);
+
+typedef SherpaOnnxVoiceActivityDetectorPop = void Function(
+    Pointer<SherpaOnnxVoiceActivityDetector>);
+
+typedef SherpaOnnxVoiceActivityDetectorClearNative = Void Function(
+    Pointer<SherpaOnnxVoiceActivityDetector>);
+
+typedef SherpaOnnxVoiceActivityDetectorClear = void Function(
+    Pointer<SherpaOnnxVoiceActivityDetector>);
+
+typedef SherpaOnnxVoiceActivityDetectorResetNative = Void Function(
+    Pointer<SherpaOnnxVoiceActivityDetector>);
+
+typedef SherpaOnnxVoiceActivityDetectorReset = void Function(
+    Pointer<SherpaOnnxVoiceActivityDetector>);
+
+typedef SherpaOnnxVoiceActivityDetectorFrontNative
+    = Pointer<SherpaOnnxSpeechSegment> Function(
+        Pointer<SherpaOnnxVoiceActivityDetector>);
+
+typedef SherpaOnnxVoiceActivityDetectorFront
+    = SherpaOnnxVoiceActivityDetectorFrontNative;
+
+typedef SherpaOnnxDestroySpeechSegmentNative = Void Function(
+    Pointer<SherpaOnnxSpeechSegment>);
+
+typedef SherpaOnnxDestroySpeechSegment = void Function(
+    Pointer<SherpaOnnxSpeechSegment>);
+
+typedef SherpaOnnxCreateCircularBufferNative = Pointer<SherpaOnnxCircularBuffer>
+    Function(Int32);
+
+typedef SherpaOnnxCreateCircularBuffer = Pointer<SherpaOnnxCircularBuffer>
+    Function(int);
+
+typedef SherpaOnnxDestroyCircularBufferNative = Void Function(
+    Pointer<SherpaOnnxCircularBuffer>);
+
+typedef SherpaOnnxDestroyCircularBuffer = void Function(
+    Pointer<SherpaOnnxCircularBuffer>);
+
+typedef SherpaOnnxCircularBufferPushNative = Void Function(
+    Pointer<SherpaOnnxCircularBuffer>, Pointer<Float>, Int32);
+
+typedef SherpaOnnxCircularBufferPush = void Function(
+    Pointer<SherpaOnnxCircularBuffer>, Pointer<Float>, int);
+
+typedef SherpaOnnxCircularBufferGetNative = Pointer<Float> Function(
+    Pointer<SherpaOnnxCircularBuffer>, Int32, Int32);
+
+typedef SherpaOnnxCircularBufferGet = Pointer<Float> Function(
+    Pointer<SherpaOnnxCircularBuffer>, int, int);
+
+typedef SherpaOnnxCircularBufferFreeNative = Void Function(Pointer<Float>);
+
+typedef SherpaOnnxCircularBufferFree = void Function(Pointer<Float>);
+
+typedef SherpaOnnxCircularBufferPopNative = Void Function(
+    Pointer<SherpaOnnxCircularBuffer>, Int32);
+
+typedef SherpaOnnxCircularBufferPop = void Function(
+    Pointer<SherpaOnnxCircularBuffer>, int);
+
+typedef SherpaOnnxCircularBufferSizeNative = Int32 Function(
+    Pointer<SherpaOnnxCircularBuffer>);
+
+typedef SherpaOnnxCircularBufferSize = int Function(
+    Pointer<SherpaOnnxCircularBuffer>);
+
+typedef SherpaOnnxCircularBufferHeadNative = Int32 Function(
+    Pointer<SherpaOnnxCircularBuffer>);
+
+typedef SherpaOnnxCircularBufferHead = int Function(
+    Pointer<SherpaOnnxCircularBuffer>);
+
+typedef SherpaOnnxCircularBufferResetNative = Void Function(
+    Pointer<SherpaOnnxCircularBuffer>);
+
+typedef SherpaOnnxCircularBufferReset = void Function(
+    Pointer<SherpaOnnxCircularBuffer>);
+
 typedef SherpaOnnxCreateSpeakerEmbeddingManagerNative
-    = Pointer<SherpaOnnxSpeakerEmbeddingManager> Function(Int32 dim);
+    = Pointer<SherpaOnnxSpeakerEmbeddingManager> Function(Int32);
 
 typedef SherpaOnnxCreateSpeakerEmbeddingManager
-    = Pointer<SherpaOnnxSpeakerEmbeddingManager> Function(int dim);
+    = Pointer<SherpaOnnxSpeakerEmbeddingManager> Function(int);
 
 typedef SherpaOnnxDestroySpeakerEmbeddingManagerNative = Void Function(
     Pointer<SherpaOnnxSpeakerEmbeddingManager>);
@@ -190,6 +350,45 @@ typedef SherpaOnnxFreeWaveNative = Void Function(Pointer<SherpaOnnxWave>);
 typedef SherpaOnnxFreeWave = void Function(Pointer<SherpaOnnxWave>);
 
 class SherpaOnnxBindings {
+  static SherpaOnnxCreateVoiceActivityDetector? createVoiceActivityDetector;
+
+  static SherpaOnnxDestroyVoiceActivityDetector? destroyVoiceActivityDetector;
+
+  static SherpaOnnxVoiceActivityDetectorAcceptWaveform?
+      voiceActivityDetectorAcceptWaveform;
+
+  static SherpaOnnxVoiceActivityDetectorEmpty? voiceActivityDetectorEmpty;
+
+  static SherpaOnnxVoiceActivityDetectorDetected? voiceActivityDetectorDetected;
+
+  static SherpaOnnxVoiceActivityDetectorPop? voiceActivityDetectorPop;
+
+  static SherpaOnnxVoiceActivityDetectorClear? voiceActivityDetectorClear;
+
+  static SherpaOnnxVoiceActivityDetectorFront? voiceActivityDetectorFront;
+
+  static SherpaOnnxDestroySpeechSegment? destroySpeechSegment;
+
+  static SherpaOnnxVoiceActivityDetectorReset? voiceActivityDetectorReset;
+
+  static SherpaOnnxCreateCircularBuffer? createCircularBuffer;
+
+  static SherpaOnnxDestroyCircularBuffer? destroyCircularBuffer;
+
+  static SherpaOnnxCircularBufferPush? circularBufferPush;
+
+  static SherpaOnnxCircularBufferGet? circularBufferGet;
+
+  static SherpaOnnxCircularBufferFree? circularBufferFree;
+
+  static SherpaOnnxCircularBufferPop? circularBufferPop;
+
+  static SherpaOnnxCircularBufferSize? circularBufferSize;
+
+  static SherpaOnnxCircularBufferHead? circularBufferHead;
+
+  static SherpaOnnxCircularBufferReset? circularBufferReset;
+
   static SherpaOnnxCreateSpeakerEmbeddingExtractor?
       createSpeakerEmbeddingExtractor;
 
@@ -252,8 +451,107 @@ class SherpaOnnxBindings {
   static SherpaOnnxFreeWave? freeWave;
 
   static void init(DynamicLibrary dynamicLibrary) {
+    createVoiceActivityDetector ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxCreateVoiceActivityDetectorNative>>(
+            'SherpaOnnxCreateVoiceActivityDetector')
+        .asFunction();
+
+    destroyVoiceActivityDetector ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxDestroyVoiceActivityDetectorNative>>(
+            'SherpaOnnxDestroyVoiceActivityDetector')
+        .asFunction();
+
+    voiceActivityDetectorAcceptWaveform ??= dynamicLibrary
+        .lookup<
+                NativeFunction<
+                    SherpaOnnxVoiceActivityDetectorAcceptWaveformNative>>(
+            'SherpaOnnxVoiceActivityDetectorAcceptWaveform')
+        .asFunction();
+
+    voiceActivityDetectorEmpty ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxVoiceActivityDetectorEmptyNative>>(
+            'SherpaOnnxVoiceActivityDetectorEmpty')
+        .asFunction();
+
+    voiceActivityDetectorDetected ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxVoiceActivityDetectorDetectedNative>>(
+            'SherpaOnnxVoiceActivityDetectorDetected')
+        .asFunction();
+
+    voiceActivityDetectorPop ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxVoiceActivityDetectorPopNative>>(
+            'SherpaOnnxVoiceActivityDetectorPop')
+        .asFunction();
+
+    voiceActivityDetectorClear ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxVoiceActivityDetectorClearNative>>(
+            'SherpaOnnxVoiceActivityDetectorClear')
+        .asFunction();
+
+    voiceActivityDetectorFront ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxVoiceActivityDetectorFrontNative>>(
+            'SherpaOnnxVoiceActivityDetectorFront')
+        .asFunction();
+
+    destroySpeechSegment ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxDestroySpeechSegmentNative>>(
+            'SherpaOnnxDestroySpeechSegment')
+        .asFunction();
+
+    voiceActivityDetectorReset ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxVoiceActivityDetectorResetNative>>(
+            'SherpaOnnxVoiceActivityDetectorReset')
+        .asFunction();
+
+    createCircularBuffer ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxCreateCircularBufferNative>>(
+            'SherpaOnnxCreateCircularBuffer')
+        .asFunction();
+
+    destroyCircularBuffer ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxDestroyCircularBufferNative>>(
+            'SherpaOnnxDestroyCircularBuffer')
+        .asFunction();
+
+    circularBufferPush ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxCircularBufferPushNative>>(
+            'SherpaOnnxCircularBufferPush')
+        .asFunction();
+
+    circularBufferGet ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxCircularBufferGetNative>>(
+            'SherpaOnnxCircularBufferGet')
+        .asFunction();
+
+    circularBufferFree ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxCircularBufferFreeNative>>(
+            'SherpaOnnxCircularBufferFree')
+        .asFunction();
+
+    circularBufferPop ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxCircularBufferPopNative>>(
+            'SherpaOnnxCircularBufferPop')
+        .asFunction();
+
+    circularBufferSize ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxCircularBufferSizeNative>>(
+            'SherpaOnnxCircularBufferSize')
+        .asFunction();
+
+    circularBufferHead ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxCircularBufferHeadNative>>(
+            'SherpaOnnxCircularBufferHead')
+        .asFunction();
+
+    circularBufferReset ??= dynamicLibrary
+        .lookup<NativeFunction<SherpaOnnxCircularBufferResetNative>>(
+            'SherpaOnnxCircularBufferReset')
+        .asFunction();
+
     createSpeakerEmbeddingExtractor ??= dynamicLibrary
-        .lookup<NativeFunction<SherpaOnnxCreateSpeakerEmbeddingExtractor>>(
+        .lookup<
+                NativeFunction<
+                    SherpaOnnxCreateSpeakerEmbeddingExtractorNative>>(
             'SherpaOnnxCreateSpeakerEmbeddingExtractor')
         .asFunction();
 

@@ -2,19 +2,20 @@
 import 'dart:ffi';
 import 'dart:typed_data';
 import 'package:ffi/ffi.dart';
-import "./sherpa_onnx_bindings.dart";
-import "./online_stream.dart";
+
+import './online_stream.dart';
+import './sherpa_onnx_bindings.dart';
 
 class SpeakerEmbeddingExtractorConfig {
   const SpeakerEmbeddingExtractorConfig(
       {required this.model,
       this.numThreads = 1,
       this.debug = true,
-      this.provider = "cpu"});
+      this.provider = 'cpu'});
 
   @override
   String toString() {
-    return "SpeakerEmbeddingExtractorConfig(model: $model, numThreads: $numThreads, debug: $debug, provider: $provider)";
+    return 'SpeakerEmbeddingExtractorConfig(model: $model, numThreads: $numThreads, debug: $debug, provider: $provider)';
   }
 
   final String model;
@@ -116,7 +117,7 @@ class SpeakerEmbeddingManager {
 
   /// Return true if added successfully; return false otherwise
   bool add({required String name, required Float32List embedding}) {
-    assert(embedding.length == this.dim, "${embedding.length} vs ${this.dim}");
+    assert(embedding.length == this.dim, '${embedding.length} vs ${this.dim}');
 
     final Pointer<Utf8> namePtr = name.toNativeUtf8();
     final int n = embedding.length;
@@ -145,7 +146,7 @@ class SpeakerEmbeddingManager {
 
     int offset = 0;
     for (final e in embeddingList) {
-      assert(e.length == this.dim, "${e.length} vs ${this.dim}");
+      assert(e.length == this.dim, '${e.length} vs ${this.dim}');
 
       pList.setAll(offset, e);
       offset += this.dim;

+// Copyright (c)  2024  Xiaomi Corporation
+import 'dart:ffi';
+import 'dart:typed_data';
+import 'package:ffi/ffi.dart';
+
+import './sherpa_onnx_bindings.dart';
+
+class SileroVadModelConfig {
+  const SileroVadModelConfig(
+      {this.model = '',
+      this.threshold = 0.5,
+      this.minSilenceDuration = 0.5,
+      this.minSpeechDuration = 0.25,
+      this.windowSize = 512});
+
+  @override
+  String toString() {
+    return 'SileroVadModelConfig(model: $model, threshold: $threshold, minSilenceDuration: $minSilenceDuration, minSpeechDuration: $minSpeechDuration, windowSize: $windowSize)';
+  }
+
+  final String model;
+  final double threshold;
+  final double minSilenceDuration;
+  final double minSpeechDuration;
+  final int windowSize;
+}
+
+class VadModelConfig {
+  VadModelConfig(
+      {this.sileroVad = const SileroVadModelConfig(),
+      this.sampleRate = 16000,
+      this.numThreads = 1,
+      this.provider = 'cpu',
+      this.debug = true});
+
+  @override
+  String toString() {
+    return 'VadModelConfig(sileroVad: $sileroVad, sampleRate: $sampleRate, numThreads: $numThreads, provider: $provider, debug: $debug)';
+  }
+
+  final SileroVadModelConfig sileroVad;
+  final int sampleRate;
+  final int numThreads;
+  final String provider;
+  final bool debug;
+}
+
+class SpeechSegment {
+  SpeechSegment({required this.samples, required this.start});
+  final Float32List samples;
+  final int start;
+}
+
+class CircularBuffer {
+  CircularBuffer._({required this.ptr});
+
+  /// The user has to invoke CircularBuffer.free() on the returned instance
+  /// to avoid memory leak.
+  factory CircularBuffer({required int capacity}) {
+    assert(capacity > 0, 'capacity is $capacity');
+    final p =
+        SherpaOnnxBindings.createCircularBuffer?.call(capacity) ?? nullptr;
+
+    return CircularBuffer._(ptr: p);
+  }
+
+  void free() {
+    SherpaOnnxBindings.destroyCircularBuffer?.call(ptr);
+    ptr = nullptr;
+  }
+
+  void push(Float32List data) {
+    final n = data.length;
+    final Pointer<Float> p = calloc<Float>(n);
+
+    final pList = p.asTypedList(n);
+    pList.setAll(0, data);
+
+    SherpaOnnxBindings.circularBufferPush?.call(this.ptr, p, n);
+
+    calloc.free(p);
+  }
+
+  Float32List get({required int startIndex, required int n}) {
+    final Pointer<Float> p =
+        SherpaOnnxBindings.circularBufferGet?.call(this.ptr, startIndex, n) ??
+            nullptr;
+
+    if (p == nullptr) {
+      return Float32List(0);
+    }
+
+    final pList = p.asTypedList(n);
+    final Float32List ans = Float32List.fromList(pList);
+
+    SherpaOnnxBindings.circularBufferFree?.call(p);
+
+    return ans;
+  }
+
+  void pop(int n) {
+    SherpaOnnxBindings.circularBufferPop?.call(this.ptr, n);
+  }
+
+  void reset() {
+    SherpaOnnxBindings.circularBufferReset?.call(this.ptr);
+  }
+
+  int get size => SherpaOnnxBindings.circularBufferSize?.call(this.ptr) ?? 0;
+  int get head => SherpaOnnxBindings.circularBufferHead?.call(this.ptr) ?? 0;
+
+  Pointer<SherpaOnnxCircularBuffer> ptr;
+}
+
+class VoiceActivityDetector {
+  VoiceActivityDetector._({required this.ptr});
+
+  // The user has to invoke VoiceActivityDetector.free() to avoid memory leak.
+  factory VoiceActivityDetector(
+      {required VadModelConfig config, required double bufferSizeInSeconds}) {
+    final c = calloc<SherpaOnnxVadModelConfig>();
+
+    final modelPtr = config.sileroVad.model.toNativeUtf8();
+    c.ref.sileroVad.model = modelPtr;
+
+    c.ref.sileroVad.threshold = config.sileroVad.threshold;
+    c.ref.sileroVad.minSilenceDuration = config.sileroVad.minSilenceDuration;
+    c.ref.sileroVad.minSpeechDuration = config.sileroVad.minSpeechDuration;
+    c.ref.sileroVad.windowSize = config.sileroVad.windowSize;
+
+    c.ref.sampleRate = config.sampleRate;
+    c.ref.numThreads = config.numThreads;
+
+    final providerPtr = config.provider.toNativeUtf8();
+    c.ref.provider = providerPtr;
+
+    c.ref.debug = config.debug ? 1 : 0;
+
+    final ptr = SherpaOnnxBindings.createVoiceActivityDetector
+            ?.call(c, bufferSizeInSeconds) ??
+        nullptr;
+
+    calloc.free(providerPtr);
+    calloc.free(modelPtr);
+    calloc.free(c);
+
+    return VoiceActivityDetector._(ptr: ptr);
+  }
+
+  void free() {
+    SherpaOnnxBindings.destroyVoiceActivityDetector?.call(ptr);
+    ptr = nullptr;
+  }
+
+  void acceptWaveform(Float32List samples) {
+    final n = samples.length;
+    final Pointer<Float> p = calloc<Float>(n);
+
+    final pList = p.asTypedList(n);
+    pList.setAll(0, samples);
+
+    SherpaOnnxBindings.voiceActivityDetectorAcceptWaveform
+        ?.call(this.ptr, p, n);
+
+    calloc.free(p);
+  }
+
+  bool isEmpty() {
+    final int empty =
+        SherpaOnnxBindings.voiceActivityDetectorEmpty?.call(this.ptr) ?? 0;
+
+    return empty == 1;
+  }
+
+  bool isDetected() {
+    final int detected =
+        SherpaOnnxBindings.voiceActivityDetectorDetected?.call(this.ptr) ?? 0;
+
+    return detected == 1;
+  }
+
+  void pop() {
+    SherpaOnnxBindings.voiceActivityDetectorPop?.call(this.ptr);
+  }
+
+  void clear() {
+    SherpaOnnxBindings.voiceActivityDetectorClear?.call(this.ptr);
+  }
+
+  SpeechSegment front() {
+    final Pointer<SherpaOnnxSpeechSegment> segment =
+        SherpaOnnxBindings.voiceActivityDetectorFront?.call(this.ptr) ??
+            nullptr;
+    if (segment == nullptr) {
+      return SpeechSegment(samples: Float32List(0), start: 0);
+    }
+
+    final sampleList = segment.ref.samples.asTypedList(segment.ref.n);
+    final start = segment.ref.start;
+
+    final samples = Float32List.fromList(sampleList);
+
+    SherpaOnnxBindings.destroySpeechSegment?.call(segment);
+
+    return SpeechSegment(samples: samples, start: start);
+  }
+
+  void reset() {
+    SherpaOnnxBindings.voiceActivityDetectorReset?.call(this.ptr);
+  }
+
+  Pointer<SherpaOnnxVoiceActivityDetector> ptr;
+}

@@ -2,7 +2,8 @@
 import 'dart:ffi';
 import 'dart:typed_data';
 import 'package:ffi/ffi.dart';
-import "./sherpa_onnx_bindings.dart";
+
+import './sherpa_onnx_bindings.dart';
 
 class WaveData {
   WaveData({required this.samples, required this.sampleRate});