sherpa-onnx-offline.cc
3.6 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
// sherpa-onnx/csrc/sherpa-onnx-offline.cc
//
// Copyright (c) 2022-2023 Xiaomi Corporation
#include <stdio.h>
#include <chrono> // NOLINT
#include <string>
#include <vector>
#include "sherpa-onnx/csrc/offline-recognizer.h"
#include "sherpa-onnx/csrc/parse-options.h"
#include "sherpa-onnx/csrc/wave-reader.h"
int main(int32_t argc, char *argv[]) {
const char *kUsageMessage = R"usage(
Usage:
(1) Transducer from icefall
./bin/sherpa-onnx-offline \
--tokens=/path/to/tokens.txt \
--encoder=/path/to/encoder.onnx \
--decoder=/path/to/decoder.onnx \
--joiner=/path/to/joiner.onnx \
--num-threads=2 \
--decoding-method=greedy_search \
/path/to/foo.wav [bar.wav foobar.wav ...]
(2) Paraformer from FunASR
./bin/sherpa-onnx-offline \
--tokens=/path/to/tokens.txt \
--paraformer=/path/to/model.onnx \
--num-threads=2 \
--decoding-method=greedy_search \
/path/to/foo.wav [bar.wav foobar.wav ...]
Note: It supports decoding multiple files in batches
Default value for num_threads is 2.
Valid values for decoding_method: greedy_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";
sherpa_onnx::ParseOptions po(kUsageMessage);
sherpa_onnx::OfflineRecognizerConfig config;
config.Register(&po);
po.Read(argc, argv);
if (po.NumArgs() < 1) {
po.PrintUsage();
exit(EXIT_FAILURE);
}
fprintf(stderr, "%s\n", config.ToString().c_str());
if (!config.Validate()) {
fprintf(stderr, "Errors in config!\n");
return -1;
}
fprintf(stderr, "Creating recognizer ...\n");
sherpa_onnx::OfflineRecognizer recognizer(config);
const auto begin = std::chrono::steady_clock::now();
fprintf(stderr, "Started\n");
std::vector<std::unique_ptr<sherpa_onnx::OfflineStream>> ss;
std::vector<sherpa_onnx::OfflineStream *> ss_pointers;
float duration = 0;
for (int32_t i = 1; i <= po.NumArgs(); ++i) {
const std::string wav_filename = po.GetArg(i);
int32_t sampling_rate = -1;
bool is_ok = false;
const 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;
}
duration += samples.size() / static_cast<float>(sampling_rate);
auto s = recognizer.CreateStream();
s->AcceptWaveform(sampling_rate, samples.data(), samples.size());
ss.push_back(std::move(s));
ss_pointers.push_back(ss.back().get());
}
recognizer.DecodeStreams(ss_pointers.data(), ss_pointers.size());
const auto end = std::chrono::steady_clock::now();
fprintf(stderr, "Done!\n\n");
for (int32_t i = 1; i <= po.NumArgs(); ++i) {
fprintf(stderr, "%s\n%s\n----\n", po.GetArg(i).c_str(),
ss[i - 1]->GetResult().AsJsonString().c_str());
}
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;
}