Add Dart API for speech enhancement GTCRN models (#1993)

This commit is contained in:
Fangjun Kuang
2025-03-12 12:39:08 +08:00
committed by GitHub
parent c3b009988b
commit fd78a482df
16 changed files with 435 additions and 0 deletions

View File

@@ -9,6 +9,8 @@
## Pure dart-examples
Hint: All of the following functions can be used in Flutter, even if some of them are only provided in pure dart api examples.
| Functions | URL | Supported Platforms|
|---|---|---|
|Speaker diarization| [Address](https://github.com/k2-fsa/sherpa-onnx/tree/master/dart-api-examples/speaker-diarization)| macOS, Windows, Linux|
@@ -20,3 +22,5 @@
|Speaker identification and verification| [Address](https://github.com/k2-fsa/sherpa-onnx/tree/master/dart-api-examples/speaker-identification)| macOS, Windows, Linux|
|Audio tagging| [Address](https://github.com/k2-fsa/sherpa-onnx/tree/master/dart-api-examples/audio-tagging)| macOS, Windows, Linux|
|Keyword spotter| [Address](https://github.com/k2-fsa/sherpa-onnx/tree/master/dart-api-examples/keyword-spotter)| macOS, Windows, Linux|
|Add punctuions| [Address](https://github.com/k2-fsa/sherpa-onnx/tree/master/dart-api-examples/add-punctuations)| macOS, Windows, Linux|
|Speech enhancement/denoising| [Address](https://github.com/k2-fsa/sherpa-onnx/tree/master/dart-api-examples/speech-enhancement-gtcrn)| macOS, Windows, Linux|

View File

@@ -8,6 +8,7 @@ export 'src/keyword_spotter.dart';
export 'src/offline_punctuation.dart';
export 'src/offline_recognizer.dart';
export 'src/offline_speaker_diarization.dart';
export 'src/offline_speech_denoiser.dart';
export 'src/offline_stream.dart';
export 'src/online_punctuation.dart';
export 'src/online_recognizer.dart';

View File

@@ -0,0 +1,169 @@
// Copyright (c) 2025 Xiaomi Corporation
import 'dart:ffi';
import 'dart:typed_data';
import 'package:ffi/ffi.dart';
import './sherpa_onnx_bindings.dart';
class OfflineSpeechDenoiserGtcrnModelConfig {
const OfflineSpeechDenoiserGtcrnModelConfig({
this.model = '',
});
factory OfflineSpeechDenoiserGtcrnModelConfig.fromJson(
Map<String, dynamic> json) {
return OfflineSpeechDenoiserGtcrnModelConfig(
model: json['model'] as String? ?? '',
);
}
@override
String toString() {
return 'OfflineSpeechDenoiserGtcrnModelConfig(model: $model)';
}
Map<String, dynamic> toJson() => {
'model': model,
};
final String model;
}
class OfflineSpeechDenoiserModelConfig {
const OfflineSpeechDenoiserModelConfig({
this.gtcrn = const OfflineSpeechDenoiserGtcrnModelConfig(),
this.numThreads = 1,
this.debug = true,
this.provider = 'cpu',
});
factory OfflineSpeechDenoiserModelConfig.fromJson(Map<String, dynamic> json) {
return OfflineSpeechDenoiserModelConfig(
gtcrn: json['gtcrn'] != null
? OfflineSpeechDenoiserGtcrnModelConfig.fromJson(
json['gtcrn'] as Map<String, dynamic>)
: const OfflineSpeechDenoiserGtcrnModelConfig(),
numThreads: json['numThreads'] as int? ?? 1,
debug: json['debug'] as bool? ?? true,
provider: json['provider'] as String? ?? 'cpu',
);
}
@override
String toString() {
return 'OfflineSpeechDenoiserModelConfig(gtcrn: $gtcrn, numThreads: $numThreads, debug: $debug, provider: $provider)';
}
Map<String, dynamic> toJson() => {
'gtcrn': gtcrn.toJson(),
'numThreads': numThreads,
'debug': debug,
'provider': provider,
};
final OfflineSpeechDenoiserGtcrnModelConfig gtcrn;
final int numThreads;
final bool debug;
final String provider;
}
class OfflineSpeechDenoiserConfig {
const OfflineSpeechDenoiserConfig({
this.model = const OfflineSpeechDenoiserModelConfig(),
});
factory OfflineSpeechDenoiserConfig.fromJson(Map<String, dynamic> json) {
return OfflineSpeechDenoiserConfig(
model: json['model'] != null
? OfflineSpeechDenoiserModelConfig.fromJson(
json['model'] as Map<String, dynamic>)
: const OfflineSpeechDenoiserModelConfig(),
);
}
@override
String toString() {
return 'OfflineSpeechDenoiserConfig(model: $model)';
}
Map<String, dynamic> toJson() => {
'model': model.toJson(),
};
final OfflineSpeechDenoiserModelConfig model;
}
class DenoisedAudio {
DenoisedAudio({
required this.samples,
required this.sampleRate,
});
final Float32List samples;
final int sampleRate;
}
class OfflineSpeechDenoiser {
OfflineSpeechDenoiser.fromPtr({required this.ptr, required this.config});
OfflineSpeechDenoiser._({required this.ptr, required this.config});
/// The user is responsible to call the OfflineSpeechDenoiser.free()
/// method of the returned instance to avoid memory leak.
factory OfflineSpeechDenoiser(OfflineSpeechDenoiserConfig config) {
final c = calloc<SherpaOnnxOfflineSpeechDenoiserConfig>();
c.ref.model.gtcrn.model = config.model.gtcrn.model.toNativeUtf8();
c.ref.model.numThreads = config.model.numThreads;
c.ref.model.debug = config.model.debug ? 1 : 0;
c.ref.model.provider = config.model.provider.toNativeUtf8();
final ptr =
SherpaOnnxBindings.sherpaOnnxCreateOfflineSpeechDenoiser?.call(c) ??
nullptr;
calloc.free(c.ref.model.provider);
calloc.free(c.ref.model.gtcrn.model);
return OfflineSpeechDenoiser._(ptr: ptr, config: config);
}
void free() {
SherpaOnnxBindings.sherpaOnnxDestroyOfflineSpeechDenoiser?.call(ptr);
ptr = nullptr;
}
DenoisedAudio run({required Float32List samples, required int sampleRate}) {
final n = samples.length;
final Pointer<Float> psamples = calloc<Float>(n);
final pList = psamples.asTypedList(n);
pList.setAll(0, samples);
final p = SherpaOnnxBindings.sherpaOnnxOfflineSpeechDenoiserRun
?.call(ptr, psamples, n, sampleRate) ??
nullptr;
calloc.free(psamples);
if (p == nullptr) {
return DenoisedAudio(samples: Float32List(0), sampleRate: 0);
}
final denoisedSamples = p.ref.samples.asTypedList(p.ref.n);
final denoisedSampleRate = p.ref.sampleRate;
final newSamples = Float32List.fromList(denoisedSamples);
SherpaOnnxBindings.sherpaOnnxDestroyDenoisedAudio?.call(p);
return DenoisedAudio(samples: newSamples, sampleRate: denoisedSampleRate);
}
int get sampleRate =>
SherpaOnnxBindings.sherpaOnnxOfflineSpeechDenoiserGetSampleRate
?.call(ptr) ??
0;
Pointer<SherpaOnnxOfflineSpeechDenoiser> ptr;
OfflineSpeechDenoiserConfig config;
}

View File

@@ -2,6 +2,36 @@
import 'dart:ffi';
import 'package:ffi/ffi.dart';
final class SherpaOnnxOfflineSpeechDenoiserGtcrnModelConfig extends Struct {
external Pointer<Utf8> model;
}
final class SherpaOnnxOfflineSpeechDenoiserModelConfig extends Struct {
external SherpaOnnxOfflineSpeechDenoiserGtcrnModelConfig gtcrn;
@Int32()
external int numThreads;
@Int32()
external int debug;
external Pointer<Utf8> provider;
}
final class SherpaOnnxOfflineSpeechDenoiserConfig extends Struct {
external SherpaOnnxOfflineSpeechDenoiserModelConfig model;
}
final class SherpaOnnxDenoisedAudio extends Struct {
external Pointer<Float> samples;
@Int32()
external int n;
@Int32()
external int sampleRate;
}
final class SherpaOnnxSpeakerEmbeddingExtractorConfig extends Struct {
external Pointer<Utf8> model;
@@ -517,6 +547,41 @@ final class SherpaOnnxOfflineSpeakerDiarization extends Opaque {}
final class SherpaOnnxOfflineSpeakerDiarizationResult extends Opaque {}
final class SherpaOnnxOfflineSpeechDenoiser extends Opaque {}
typedef SherpaOnnxCreateOfflineSpeechDenoiserNative
= Pointer<SherpaOnnxOfflineSpeechDenoiser> Function(
Pointer<SherpaOnnxOfflineSpeechDenoiserConfig>);
typedef SherpaOnnxCreateOfflineSpeechDenoiser
= SherpaOnnxCreateOfflineSpeechDenoiserNative;
typedef SherpaOnnxDestroyOfflineSpeechDenoiserNative = Void Function(
Pointer<SherpaOnnxOfflineSpeechDenoiser>);
typedef SherpaOnnxDestroyOfflineSpeechDenoiser = void Function(
Pointer<SherpaOnnxOfflineSpeechDenoiser>);
typedef SherpaOnnxOfflineSpeechDenoiserGetSampleRateNative = Int32 Function(
Pointer<SherpaOnnxOfflineSpeechDenoiser>);
typedef SherpaOnnxOfflineSpeechDenoiserGetSampleRate = int Function(
Pointer<SherpaOnnxOfflineSpeechDenoiser>);
typedef SherpaOnnxOfflineSpeechDenoiserRunNative
= Pointer<SherpaOnnxDenoisedAudio> Function(
Pointer<SherpaOnnxOfflineSpeechDenoiser>, Pointer<Float>, Int32, Int32);
typedef SherpaOnnxOfflineSpeechDenoiserRun
= Pointer<SherpaOnnxDenoisedAudio> Function(
Pointer<SherpaOnnxOfflineSpeechDenoiser>, Pointer<Float>, int, int);
typedef SherpaOnnxDestroyDenoisedAudioNative = Void Function(
Pointer<SherpaOnnxDenoisedAudio>);
typedef SherpaOnnxDestroyDenoisedAudio = void Function(
Pointer<SherpaOnnxDenoisedAudio>);
typedef SherpaOnnxCreateOfflineSpeakerDiarizationNative
= Pointer<SherpaOnnxOfflineSpeakerDiarization> Function(
Pointer<SherpaOnnxOfflineSpeakerDiarizationConfig>);
@@ -1172,6 +1237,17 @@ typedef SherpaOnnxFreeWaveNative = Void Function(Pointer<SherpaOnnxWave>);
typedef SherpaOnnxFreeWave = void Function(Pointer<SherpaOnnxWave>);
class SherpaOnnxBindings {
static SherpaOnnxCreateOfflineSpeechDenoiser?
sherpaOnnxCreateOfflineSpeechDenoiser;
static SherpaOnnxDestroyOfflineSpeechDenoiser?
sherpaOnnxDestroyOfflineSpeechDenoiser;
static SherpaOnnxOfflineSpeechDenoiserGetSampleRate?
sherpaOnnxOfflineSpeechDenoiserGetSampleRate;
static SherpaOnnxOfflineSpeechDenoiserRun? sherpaOnnxOfflineSpeechDenoiserRun;
static SherpaOnnxDestroyDenoisedAudio? sherpaOnnxDestroyDenoisedAudio;
static SherpaOnnxCreateOfflineSpeakerDiarization?
sherpaOnnxCreateOfflineSpeakerDiarization;
static SherpaOnnxDestroyOfflineSpeakerDiarization?
@@ -1370,6 +1446,33 @@ class SherpaOnnxBindings {
static SherpaOnnxFreeWave? freeWave;
static void init(DynamicLibrary dynamicLibrary) {
sherpaOnnxCreateOfflineSpeechDenoiser ??= dynamicLibrary
.lookup<NativeFunction<SherpaOnnxCreateOfflineSpeechDenoiserNative>>(
'SherpaOnnxCreateOfflineSpeechDenoiser')
.asFunction();
sherpaOnnxDestroyOfflineSpeechDenoiser ??= dynamicLibrary
.lookup<NativeFunction<SherpaOnnxDestroyOfflineSpeechDenoiserNative>>(
'SherpaOnnxDestroyOfflineSpeechDenoiser')
.asFunction();
sherpaOnnxOfflineSpeechDenoiserGetSampleRate ??= dynamicLibrary
.lookup<
NativeFunction<
SherpaOnnxOfflineSpeechDenoiserGetSampleRateNative>>(
'SherpaOnnxOfflineSpeechDenoiserGetSampleRate')
.asFunction();
sherpaOnnxOfflineSpeechDenoiserRun ??= dynamicLibrary
.lookup<NativeFunction<SherpaOnnxOfflineSpeechDenoiserRunNative>>(
'SherpaOnnxOfflineSpeechDenoiserRun')
.asFunction();
sherpaOnnxDestroyDenoisedAudio ??= dynamicLibrary
.lookup<NativeFunction<SherpaOnnxDestroyDenoisedAudioNative>>(
'SherpaOnnxDestroyDenoisedAudio')
.asFunction();
sherpaOnnxCreateOfflineSpeakerDiarization ??= dynamicLibrary
.lookup<
NativeFunction<