vad-with-paraformer.dart
3.4 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
// Copyright (c) 2024 Xiaomi Corporation
import 'dart:io';
import 'dart:typed_data';
import 'package:args/args.dart';
import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;
import './init.dart';
void main(List<String> arguments) async {
await initSherpaOnnx();
final parser = ArgParser()
..addOption('silero-vad', help: 'Path to silero_vad.onnx')
..addOption('model', help: 'Path to the paraformer model')
..addOption('tokens', help: 'Path to tokens.txt')
..addOption('input-wav', help: 'Path to input.wav to transcribe');
final res = parser.parse(arguments);
if (res['silero-vad'] == null ||
res['model'] == null ||
res['tokens'] == null ||
res['input-wav'] == null) {
print(parser.usage);
exit(1);
}
final sileroVad = res['silero-vad'] as String;
final model = res['model'] as String;
final tokens = res['tokens'] as String;
final inputWav = res['input-wav'] as String;
final paraformer = sherpa_onnx.OfflineParaformerModelConfig(
model: model,
);
final modelConfig = sherpa_onnx.OfflineModelConfig(
paraformer: paraformer,
tokens: tokens,
debug: true,
numThreads: 1,
modelType: 'paraformer',
);
final config = sherpa_onnx.OfflineRecognizerConfig(model: modelConfig);
final recognizer = sherpa_onnx.OfflineRecognizer(config);
final sileroVadConfig = sherpa_onnx.SileroVadModelConfig(
model: sileroVad,
minSilenceDuration: 0.25,
minSpeechDuration: 0.5,
);
final vadConfig = sherpa_onnx.VadModelConfig(
sileroVad: sileroVadConfig,
numThreads: 1,
debug: true,
);
final vad = sherpa_onnx.VoiceActivityDetector(
config: vadConfig, bufferSizeInSeconds: 10);
final waveData = sherpa_onnx.readWave(inputWav);
int numSamples = waveData.samples.length;
int numIter = numSamples ~/ vadConfig.sileroVad.windowSize;
for (int i = 0; i != numIter; ++i) {
int start = i * vadConfig.sileroVad.windowSize;
vad.acceptWaveform(Float32List.sublistView(
waveData.samples, start, start + vadConfig.sileroVad.windowSize));
while (!vad.isEmpty()) {
final stream = recognizer.createStream();
final segment = vad.front();
stream.acceptWaveform(
samples: segment.samples, sampleRate: waveData.sampleRate);
recognizer.decode(stream);
final result = recognizer.getResult(stream);
final startTime = segment.start * 1.0 / waveData.sampleRate;
final duration = segment.samples.length * 1.0 / waveData.sampleRate;
final stopTime = startTime + duration;
if (result.text != '') {
print(
'${startTime.toStringAsPrecision(4)} -- ${stopTime.toStringAsPrecision(4)}: ${result.text}');
}
stream.free();
vad.pop();
}
}
vad.flush();
while (!vad.isEmpty()) {
final stream = recognizer.createStream();
final segment = vad.front();
stream.acceptWaveform(
samples: segment.samples, sampleRate: waveData.sampleRate);
recognizer.decode(stream);
final result = recognizer.getResult(stream);
final startTime = segment.start * 1.0 / waveData.sampleRate;
final duration = segment.samples.length * 1.0 / waveData.sampleRate;
final stopTime = startTime + duration;
if (result.text != '') {
print(
'${startTime.toStringAsPrecision(4)} -- ${stopTime.toStringAsPrecision(4)}: ${result.text}');
}
stream.free();
vad.pop();
}
vad.free();
recognizer.free();
}