zipformer-transducer.dart
3.8 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
119
120
121
122
123
124
125
126
127
128
129
130
131
// 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('encoder', help: 'Path to the encoder model')
..addOption('decoder', help: 'Path to decoder model')
..addOption('joiner', help: 'Path to joiner 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['encoder'] == null ||
res['decoder'] == null ||
res['joiner'] == null ||
res['tokens'] == null ||
res['input-wav'] == null) {
print(parser.usage);
exit(1);
}
// create VAD
final sileroVad = res['silero-vad'] as String;
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);
// create zipformer transducer recognizer
final encoder = res['encoder'] as String;
final decoder = res['decoder'] as String;
final joiner = res['joiner'] as String;
final tokens = res['tokens'] as String;
final inputWav = res['input-wav'] as String;
final transducer = sherpa_onnx.OfflineTransducerModelConfig(
encoder: encoder,
decoder: decoder,
joiner: joiner,
);
final modelConfig = sherpa_onnx.OfflineModelConfig(
transducer: transducer,
tokens: tokens,
debug: true,
numThreads: 1,
);
final config = sherpa_onnx.OfflineRecognizerConfig(model: modelConfig);
final recognizer = sherpa_onnx.OfflineRecognizer(config);
final waveData = sherpa_onnx.readWave(inputWav);
if (waveData.sampleRate != 16000) {
print('Only 16000 Hz is supported. Given: ${waveData.sampleRate}');
exit(1);
}
int numSamples = waveData.samples.length;
int numIter = numSamples ~/ vadConfig.sileroVad.windowSize;
for (int i = 0; i != numIter; ++i) {
int start = i * vadConfig.sileroVad.windowSize;
vad.acceptWaveform(Float32List.sublistView(
waveData.samples, start, start + vadConfig.sileroVad.windowSize));
if (vad.isDetected()) {
while (!vad.isEmpty()) {
final samples = vad.front().samples;
final startTime = vad.front().start.toDouble() / waveData.sampleRate;
final endTime =
startTime + samples.length.toDouble() / waveData.sampleRate;
final stream = recognizer.createStream();
stream.acceptWaveform(
samples: samples, sampleRate: waveData.sampleRate);
recognizer.decode(stream);
final result = recognizer.getResult(stream);
stream.free();
print(
'${startTime.toStringAsPrecision(5)} -- ${endTime.toStringAsPrecision(5)} : ${result.text}');
vad.pop();
}
}
}
vad.flush();
while (!vad.isEmpty()) {
final samples = vad.front().samples;
final startTime = vad.front().start.toDouble() / waveData.sampleRate;
final endTime = startTime + samples.length.toDouble() / waveData.sampleRate;
final stream = recognizer.createStream();
stream.acceptWaveform(samples: samples, sampleRate: waveData.sampleRate);
recognizer.decode(stream);
final result = recognizer.getResult(stream);
stream.free();
print(
'${startTime.toStringAsPrecision(5)} -- ${endTime.toStringAsPrecision(5)} : ${result.text}');
vad.pop();
}
vad.free();
recognizer.free();
}