Add TTS API and examples for Dart (#1010)

This commit is contained in:
Fangjun Kuang
2024-06-15 14:30:36 +08:00
committed by GitHub
parent e3077670c6
commit e52d32b95b
20 changed files with 874 additions and 0 deletions

View File

@@ -8,6 +8,7 @@ export 'src/offline_stream.dart';
export 'src/online_recognizer.dart';
export 'src/online_stream.dart';
export 'src/speaker_identification.dart';
export 'src/tts.dart';
export 'src/vad.dart';
export 'src/wave_reader.dart';
export 'src/wave_writer.dart';

View File

@@ -2,6 +2,55 @@
import 'dart:ffi';
import 'package:ffi/ffi.dart';
final class SherpaOnnxOfflineTtsVitsModelConfig extends Struct {
external Pointer<Utf8> model;
external Pointer<Utf8> lexicon;
external Pointer<Utf8> tokens;
external Pointer<Utf8> dataDir;
@Float()
external double noiseScale;
@Float()
external double noiseScaleW;
@Float()
external double lengthScale;
external Pointer<Utf8> dictDir;
}
final class SherpaOnnxOfflineTtsModelConfig extends Struct {
external SherpaOnnxOfflineTtsVitsModelConfig vits;
@Int32()
external int numThreads;
@Int32()
external int debug;
external Pointer<Utf8> provider;
}
final class SherpaOnnxOfflineTtsConfig extends Struct {
external SherpaOnnxOfflineTtsModelConfig model;
external Pointer<Utf8> ruleFsts;
@Int32()
external int maxNumSenetences;
external Pointer<Utf8> ruleFars;
}
final class SherpaOnnxGeneratedAudio extends Struct {
external Pointer<Float> samples;
@Int32()
external int n;
@Int32()
external int sampleRate;
}
final class SherpaOnnxFeatureConfig extends Struct {
@Int32()
external int sampleRate;
@@ -218,6 +267,8 @@ final class SherpaOnnxSpeakerEmbeddingExtractorConfig extends Struct {
external Pointer<Utf8> provider;
}
final class SherpaOnnxOfflineTts extends Opaque {}
final class SherpaOnnxCircularBuffer extends Opaque {}
final class SherpaOnnxVoiceActivityDetector extends Opaque {}
@@ -234,6 +285,60 @@ final class SherpaOnnxSpeakerEmbeddingExtractor extends Opaque {}
final class SherpaOnnxSpeakerEmbeddingManager extends Opaque {}
typedef SherpaOnnxCreateOfflineTtsNative = Pointer<SherpaOnnxOfflineTts>
Function(Pointer<SherpaOnnxOfflineTtsConfig>);
typedef SherpaOnnxCreateOfflineTts = SherpaOnnxCreateOfflineTtsNative;
typedef SherpaOnnxDestroyOfflineTtsNative = Void Function(
Pointer<SherpaOnnxOfflineTts>);
typedef SherpaOnnxDestroyOfflineTts = void Function(
Pointer<SherpaOnnxOfflineTts>);
typedef SherpaOnnxOfflineTtsSampleRateNative = Int32 Function(
Pointer<SherpaOnnxOfflineTts>);
typedef SherpaOnnxOfflineTtsSampleRate = int Function(
Pointer<SherpaOnnxOfflineTts>);
typedef SherpaOnnxOfflineTtsNumSpeakersNative = Int32 Function(
Pointer<SherpaOnnxOfflineTts>);
typedef SherpaOnnxOfflineTtsNumSpeakers = int Function(
Pointer<SherpaOnnxOfflineTts>);
typedef SherpaOnnxOfflineTtsGenerateNative = Pointer<SherpaOnnxGeneratedAudio>
Function(Pointer<SherpaOnnxOfflineTts>, Pointer<Utf8>, Int32, Float);
typedef SherpaOnnxOfflineTtsGenerate = Pointer<SherpaOnnxGeneratedAudio>
Function(Pointer<SherpaOnnxOfflineTts>, Pointer<Utf8>, int, double);
typedef SherpaOnnxDestroyOfflineTtsGeneratedAudioNative = Void Function(
Pointer<SherpaOnnxGeneratedAudio>);
typedef SherpaOnnxDestroyOfflineTtsGeneratedAudio = void Function(
Pointer<SherpaOnnxGeneratedAudio>);
typedef SherpaOnnxGeneratedAudioCallbackNative = Void Function(
Pointer<Float>, Int32);
typedef SherpaOnnxOfflineTtsGenerateWithCallbackNative
= Pointer<SherpaOnnxGeneratedAudio> Function(
Pointer<SherpaOnnxOfflineTts>,
Pointer<Utf8>,
Int32,
Float,
Pointer<NativeFunction<SherpaOnnxGeneratedAudioCallbackNative>>);
typedef SherpaOnnxOfflineTtsGenerateWithCallback
= Pointer<SherpaOnnxGeneratedAudio> Function(
Pointer<SherpaOnnxOfflineTts>,
Pointer<Utf8>,
int,
double,
Pointer<NativeFunction<SherpaOnnxGeneratedAudioCallbackNative>>);
typedef CreateOfflineRecognizerNative = Pointer<SherpaOnnxOfflineRecognizer>
Function(Pointer<SherpaOnnxOfflineRecognizerConfig>);
@@ -608,6 +713,16 @@ typedef SherpaOnnxFreeWaveNative = Void Function(Pointer<SherpaOnnxWave>);
typedef SherpaOnnxFreeWave = void Function(Pointer<SherpaOnnxWave>);
class SherpaOnnxBindings {
static SherpaOnnxCreateOfflineTts? createOfflineTts;
static SherpaOnnxDestroyOfflineTts? destroyOfflineTts;
static SherpaOnnxOfflineTtsSampleRate? offlineTtsSampleRate;
static SherpaOnnxOfflineTtsNumSpeakers? offlineTtsNumSpeakers;
static SherpaOnnxOfflineTtsGenerate? offlineTtsGenerate;
static SherpaOnnxDestroyOfflineTtsGeneratedAudio?
destroyOfflineTtsGeneratedAudio;
static SherpaOnnxOfflineTtsGenerateWithCallback?
offlineTtsGenerateWithCallback;
static CreateOfflineRecognizer? createOfflineRecognizer;
static DestroyOfflineRecognizer? destroyOfflineRecognizer;
static CreateOfflineStream? createOfflineStream;
@@ -740,6 +855,43 @@ class SherpaOnnxBindings {
static SherpaOnnxFreeWave? freeWave;
static void init(DynamicLibrary dynamicLibrary) {
createOfflineTts ??= dynamicLibrary
.lookup<NativeFunction<SherpaOnnxCreateOfflineTtsNative>>(
'SherpaOnnxCreateOfflineTts')
.asFunction();
destroyOfflineTts ??= dynamicLibrary
.lookup<NativeFunction<SherpaOnnxDestroyOfflineTtsNative>>(
'SherpaOnnxDestroyOfflineTts')
.asFunction();
offlineTtsSampleRate ??= dynamicLibrary
.lookup<NativeFunction<SherpaOnnxOfflineTtsSampleRateNative>>(
'SherpaOnnxOfflineTtsSampleRate')
.asFunction();
offlineTtsNumSpeakers ??= dynamicLibrary
.lookup<NativeFunction<SherpaOnnxOfflineTtsNumSpeakersNative>>(
'SherpaOnnxOfflineTtsNumSpeakers')
.asFunction();
offlineTtsGenerate ??= dynamicLibrary
.lookup<NativeFunction<SherpaOnnxOfflineTtsGenerateNative>>(
'SherpaOnnxOfflineTtsGenerate')
.asFunction();
destroyOfflineTtsGeneratedAudio ??= dynamicLibrary
.lookup<
NativeFunction<
SherpaOnnxDestroyOfflineTtsGeneratedAudioNative>>(
'SherpaOnnxDestroyOfflineTtsGeneratedAudio')
.asFunction();
offlineTtsGenerateWithCallback ??= dynamicLibrary
.lookup<NativeFunction<SherpaOnnxOfflineTtsGenerateWithCallbackNative>>(
'SherpaOnnxOfflineTtsGenerateWithCallback')
.asFunction();
createOfflineRecognizer ??= dynamicLibrary
.lookup<NativeFunction<CreateOfflineRecognizerNative>>(
'CreateOfflineRecognizer')

View File

@@ -0,0 +1,195 @@
// Copyright (c) 2024 Xiaomi Corporation
import 'dart:convert';
import 'dart:ffi';
import 'dart:typed_data';
import 'package:ffi/ffi.dart';
import './sherpa_onnx_bindings.dart';
class OfflineTtsVitsModelConfig {
const OfflineTtsVitsModelConfig({
required this.model,
this.lexicon = '',
required this.tokens,
this.dataDir = '',
this.noiseScale = 0.667,
this.noiseScaleW = 0.8,
this.lengthScale = 1.0,
this.dictDir = '',
});
@override
String toString() {
return 'OfflineTtsVitsModelConfig(model: $model, lexicon: $lexicon, tokens: $tokens, dataDir: $dataDir, noiseScale: $noiseScale, noiseScaleW: $noiseScaleW, lengthScale: $lengthScale, dictDir: $dictDir)';
}
final String model;
final String lexicon;
final String tokens;
final String dataDir;
final double noiseScale;
final double noiseScaleW;
final double lengthScale;
final String dictDir;
}
class OfflineTtsModelConfig {
const OfflineTtsModelConfig({
required this.vits,
this.numThreads = 1,
this.debug = true,
this.provider = 'cpu',
});
@override
String toString() {
return 'OfflineTtsModelConfig(vits: $vits, numThreads: $numThreads, debug: $debug, provider: $provider)';
}
final OfflineTtsVitsModelConfig vits;
final int numThreads;
final bool debug;
final String provider;
}
class OfflineTtsConfig {
const OfflineTtsConfig({
required this.model,
this.ruleFsts = '',
this.maxNumSenetences = 1,
this.ruleFars = '',
});
@override
String toString() {
return 'OfflineTtsConfig(model: $model, ruleFsts: $ruleFsts, maxNumSenetences: $maxNumSenetences, ruleFars: $ruleFars)';
}
final OfflineTtsModelConfig model;
final String ruleFsts;
final int maxNumSenetences;
final String ruleFars;
}
class GeneratedAudio {
GeneratedAudio({
required this.samples,
required this.sampleRate,
});
final Float32List samples;
final int sampleRate;
}
class OfflineTts {
OfflineTts._({required this.ptr, required this.config});
/// The user is responsible to call the OfflineTts.free()
/// method of the returned instance to avoid memory leak.
factory OfflineTts(OfflineTtsConfig config) {
final c = calloc<SherpaOnnxOfflineTtsConfig>();
c.ref.model.vits.model = config.model.vits.model.toNativeUtf8();
c.ref.model.vits.lexicon = config.model.vits.lexicon.toNativeUtf8();
c.ref.model.vits.tokens = config.model.vits.tokens.toNativeUtf8();
c.ref.model.vits.dataDir = config.model.vits.dataDir.toNativeUtf8();
c.ref.model.vits.noiseScale = config.model.vits.noiseScale;
c.ref.model.vits.noiseScaleW = config.model.vits.noiseScaleW;
c.ref.model.vits.lengthScale = config.model.vits.lengthScale;
c.ref.model.vits.dictDir = config.model.vits.dictDir.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();
c.ref.ruleFsts = config.ruleFsts.toNativeUtf8();
c.ref.maxNumSenetences = config.maxNumSenetences;
c.ref.ruleFars = config.ruleFars.toNativeUtf8();
final ptr = SherpaOnnxBindings.createOfflineTts?.call(c) ?? nullptr;
calloc.free(c.ref.ruleFars);
calloc.free(c.ref.ruleFsts);
calloc.free(c.ref.model.provider);
calloc.free(c.ref.model.vits.dictDir);
calloc.free(c.ref.model.vits.dataDir);
calloc.free(c.ref.model.vits.tokens);
calloc.free(c.ref.model.vits.lexicon);
calloc.free(c.ref.model.vits.model);
return OfflineTts._(ptr: ptr, config: config);
}
void free() {
SherpaOnnxBindings.destroyOfflineTts?.call(ptr);
ptr = nullptr;
}
GeneratedAudio generate(
{required String text, int sid = 0, double speed = 1.0}) {
final Pointer<Utf8> textPtr = text.toNativeUtf8();
final p =
SherpaOnnxBindings.offlineTtsGenerate?.call(ptr, textPtr, sid, speed) ??
nullptr;
calloc.free(textPtr);
if (p == nullptr) {
return GeneratedAudio(samples: Float32List(0), sampleRate: 0);
}
final samples = p.ref.samples.asTypedList(p.ref.n);
final sampleRate = p.ref.sampleRate;
final newSamples = Float32List.fromList(samples);
SherpaOnnxBindings.destroyOfflineTtsGeneratedAudio?.call(p);
return GeneratedAudio(samples: newSamples, sampleRate: sampleRate);
}
GeneratedAudio generateWithCallback(
{required String text,
int sid = 0,
double speed = 1.0,
required void Function(Float32List samples) callback}) {
// see
// https://github.com/dart-lang/sdk/issues/54276#issuecomment-1846109285
// https://stackoverflow.com/questions/69537440/callbacks-in-dart-dartffi-only-supports-calling-static-dart-functions-from-nat
// https://github.com/dart-lang/sdk/blob/main/tests/ffi/isolate_local_function_callbacks_test.dart#L46
final wrapper =
NativeCallable<SherpaOnnxGeneratedAudioCallbackNative>.isolateLocal(
(Pointer<Float> samples, int n) {
final s = samples.asTypedList(n);
final newSamples = Float32List.fromList(s);
callback(newSamples);
});
final Pointer<Utf8> textPtr = text.toNativeUtf8();
final p = SherpaOnnxBindings.offlineTtsGenerateWithCallback
?.call(ptr, textPtr, sid, speed, wrapper.nativeFunction) ??
nullptr;
calloc.free(textPtr);
wrapper.close();
if (p == nullptr) {
return GeneratedAudio(samples: Float32List(0), sampleRate: 0);
}
final samples = p.ref.samples.asTypedList(p.ref.n);
final sampleRate = p.ref.sampleRate;
final newSamples = Float32List.fromList(samples);
SherpaOnnxBindings.destroyOfflineTtsGeneratedAudio?.call(p);
return GeneratedAudio(samples: newSamples, sampleRate: sampleRate);
}
int get sampleRate =>
SherpaOnnxBindings.offlineTtsSampleRate?.call(this.ptr) ?? 0;
int get numSpeakers =>
SherpaOnnxBindings.offlineTtsNumSpeakers?.call(this.ptr) ?? 0;
Pointer<SherpaOnnxOfflineTts> ptr;
OfflineTtsConfig config;
}