sherpa-onnx.cc
3.9 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
132
133
// sherpa-onnx/csrc/sherpa-onnx.cc
//
// Copyright (c) 2022-2023 Xiaomi Corporation
#include <stdio.h>
#include <chrono> // NOLINT
#include <string>
#include <vector>
#include "sherpa-onnx/csrc/online-recognizer.h"
#include "sherpa-onnx/csrc/online-stream.h"
#include "sherpa-onnx/csrc/symbol-table.h"
#include "sherpa-onnx/csrc/wave-reader.h"
// TODO(fangjun): Use ParseOptions as we are getting more args
int main(int32_t argc, char *argv[]) {
if (argc < 6 || argc > 9) {
const char *usage = R"usage(
Usage:
./bin/sherpa-onnx \
/path/to/tokens.txt \
/path/to/encoder.onnx \
/path/to/decoder.onnx \
/path/to/joiner.onnx \
/path/to/foo.wav [num_threads [decoding_method [/path/to/rnn_lm.onnx]]]
Default value for num_threads is 2.
Valid values for decoding_method: greedy_search (default), modified_beam_search.
foo.wav should be of single channel, 16-bit PCM encoded wave file; its
sampling rate can be arbitrary and does not need to be 16kHz.
Please refer to
https://k2-fsa.github.io/sherpa/onnx/pretrained_models/index.html
for a list of pre-trained models to download.
)usage";
fprintf(stderr, "%s\n", usage);
return 0;
}
sherpa_onnx::OnlineRecognizerConfig config;
config.model_config.tokens = argv[1];
config.model_config.debug = false;
config.model_config.encoder_filename = argv[2];
config.model_config.decoder_filename = argv[3];
config.model_config.joiner_filename = argv[4];
std::string wav_filename = argv[5];
config.model_config.num_threads = 2;
if (argc == 7 && atoi(argv[6]) > 0) {
config.model_config.num_threads = atoi(argv[6]);
}
if (argc == 8) {
config.decoding_method = argv[7];
}
if (argc == 9) {
config.lm_config.model = argv[8];
}
config.max_active_paths = 4;
fprintf(stderr, "%s\n", config.ToString().c_str());
if (!config.Validate()) {
fprintf(stderr, "Errors in config!\n");
return -1;
}
sherpa_onnx::OnlineRecognizer recognizer(config);
int32_t sampling_rate = -1;
bool is_ok = false;
std::vector<float> samples =
sherpa_onnx::ReadWave(wav_filename, &sampling_rate, &is_ok);
if (!is_ok) {
fprintf(stderr, "Failed to read %s\n", wav_filename.c_str());
return -1;
}
fprintf(stderr, "sampling rate of input file: %d\n", sampling_rate);
float duration = samples.size() / static_cast<float>(sampling_rate);
fprintf(stderr, "wav filename: %s\n", wav_filename.c_str());
fprintf(stderr, "wav duration (s): %.3f\n", duration);
auto begin = std::chrono::steady_clock::now();
fprintf(stderr, "Started\n");
auto s = recognizer.CreateStream();
s->AcceptWaveform(sampling_rate, samples.data(), samples.size());
std::vector<float> tail_paddings(static_cast<int>(0.5 * sampling_rate));
// Note: We can call AcceptWaveform() multiple times.
s->AcceptWaveform(sampling_rate, tail_paddings.data(), tail_paddings.size());
// Call InputFinished() to indicate that no audio samples are available
s->InputFinished();
while (recognizer.IsReady(s.get())) {
recognizer.DecodeStream(s.get());
}
std::string text = recognizer.GetResult(s.get()).AsJsonString();
fprintf(stderr, "Done!\n");
fprintf(stderr, "Recognition result for %s:\n%s\n", wav_filename.c_str(),
text.c_str());
auto end = std::chrono::steady_clock::now();
float elapsed_seconds =
std::chrono::duration_cast<std::chrono::milliseconds>(end - begin)
.count() /
1000.;
fprintf(stderr, "num threads: %d\n", config.model_config.num_threads);
fprintf(stderr, "decoding method: %s\n", config.decoding_method.c_str());
if (config.decoding_method == "modified_beam_search") {
fprintf(stderr, "max active paths: %d\n", config.max_active_paths);
}
fprintf(stderr, "Elapsed seconds: %.3f s\n", elapsed_seconds);
float rtf = elapsed_seconds / duration;
fprintf(stderr, "Real time factor (RTF): %.3f / %.3f = %.3f\n",
elapsed_seconds, duration, rtf);
return 0;
}