speech_recognition/FunASR
1
0
Fork 0
mirror of https://github.com/modelscope/FunASR synced 2025-09-15 14:48:36 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

adapted pcm to 8k

Browse Source
This commit is contained in:
雾聪 2023-12-13 13:57:59 +08:00
parent a0048dc766
commit de87e1d180
13 changed files with 38 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
runtime/onnxruntime/bin/funasr-onnx-2pass-rtf.cpp
View File

@ -43,7 +43,7 @@ void GetValue(TCLAP::ValueArg<std::string>& value_arg, string key, std::map<std:
LOG(INFO)<< key << " : " << value_arg.getValue();
}
void runReg(FUNASR_HANDLE tpass_handle, std::vector<int> chunk_size, vector<string> wav_list, vector<string> wav_ids,
void runReg(FUNASR_HANDLE tpass_handle, std::vector<int> chunk_size, vector<string> wav_list, vector<string> wav_ids, int audio_fs,
float* total_length, long* total_time, int core_id, ASR_TYPE asr_mode_, string nn_hotwords_) {
struct timeval start, end;
@ -59,7 +59,7 @@ void runReg(FUNASR_HANDLE tpass_handle, std::vector<int> chunk_size, vector<stri
// warm up
for (size_t i = 0; i < 2; i++)
{
int32_t sampling_rate_ = 16000;
int32_t sampling_rate_ = audio_fs;
funasr::Audio audio(1);
if(is_target_file(wav_list[0].c_str(), "wav")){
if(!audio.LoadWav2Char(wav_list[0].c_str(), &sampling_rate_)){
@ -104,7 +104,7 @@ void runReg(FUNASR_HANDLE tpass_handle, std::vector<int> chunk_size, vector<stri
if (i >= wav_list.size()) {
break;
}
int32_t sampling_rate_ = 16000;
int32_t sampling_rate_ = audio_fs;
funasr::Audio audio(1);
if(is_target_file(wav_list[i].c_str(), "wav")){
if(!audio.LoadWav2Char(wav_list[i].c_str(), &sampling_rate_)){
@ -220,6 +220,7 @@ int main(int argc, char** argv)
TCLAP::ValueArg<std::int32_t> onnx_thread("", "model-thread-num", "onnxruntime SetIntraOpNumThreads", false, 1, "int32_t");
TCLAP::ValueArg<std::int32_t> thread_num_("", THREAD_NUM, "multi-thread num for rtf", false, 1, "int32_t");
TCLAP::ValueArg<std::string> wav_path("", WAV_PATH, "the input could be: wav_path, e.g.: asr_example.wav; pcm_path, e.g.: asr_example.pcm; wav.scp, kaldi style wav list (wav_id \t wav_path)", true, "", "string");
TCLAP::ValueArg<std::int32_t> audio_fs("", AUDIO_FS, "the sample rate of audio", false, 16000, "int32_t");
TCLAP::ValueArg<std::string> hotword("", HOTWORD, "the hotword file, one hotword perline, Format: Hotword Weight (could be: 阿里巴巴 20)", false, "", "string");
cmd.add(offline_model_dir);
@ -231,6 +232,7 @@ int main(int argc, char** argv)
cmd.add(punc_quant);
cmd.add(itn_dir);
cmd.add(wav_path);
cmd.add(audio_fs);
cmd.add(asr_mode);
cmd.add(onnx_thread);
cmd.add(thread_num_);
@ -319,7 +321,7 @@ int main(int argc, char** argv)
int rtf_threds = thread_num_.getValue();
for (int i = 0; i < rtf_threds; i++)
{
threads.emplace_back(thread(runReg, tpass_hanlde, chunk_size, wav_list, wav_ids, &total_length, &total_time, i, (ASR_TYPE)asr_mode_, nn_hotwords_));
threads.emplace_back(thread(runReg, tpass_hanlde, chunk_size, wav_list, wav_ids, audio_fs.getValue(), &total_length, &total_time, i, (ASR_TYPE)asr_mode_, nn_hotwords_));
}
for (auto& thread : threads)

4
runtime/onnxruntime/bin/funasr-onnx-2pass.cpp
View File

@ -55,6 +55,7 @@ int main(int argc, char** argv)
TCLAP::ValueArg<std::int32_t> onnx_thread("", "model-thread-num", "onnxruntime SetIntraOpNumThreads", false, 1, "int32_t");
TCLAP::ValueArg<std::string> wav_path("", WAV_PATH, "the input could be: wav_path, e.g.: asr_example.wav; pcm_path, e.g.: asr_example.pcm; wav.scp, kaldi style wav list (wav_id \t wav_path)", true, "", "string");
TCLAP::ValueArg<std::int32_t> audio_fs("", AUDIO_FS, "the sample rate of audio", false, 16000, "int32_t");
TCLAP::ValueArg<std::string> hotword("", HOTWORD, "the hotword file, one hotword perline, Format: Hotword Weight (could be: 阿里巴巴 20)", false, "", "string");
cmd.add(offline_model_dir);
@ -66,6 +67,7 @@ int main(int argc, char** argv)
cmd.add(punc_quant);
cmd.add(itn_dir);
cmd.add(wav_path);
cmd.add(audio_fs);
cmd.add(asr_mode);
cmd.add(onnx_thread);
cmd.add(hotword);
@ -154,7 +156,7 @@ int main(int argc, char** argv)
auto& wav_file = wav_list[i];
auto& wav_id = wav_ids[i];
int32_t sampling_rate_ = 16000;
int32_t sampling_rate_ = audio_fs.getValue();
funasr::Audio audio(1);
if(is_target_file(wav_file.c_str(), "wav")){
if(!audio.LoadWav2Char(wav_file.c_str(), &sampling_rate_)){

10
runtime/onnxruntime/bin/funasr-onnx-offline-rtf.cpp
View File

@ -29,7 +29,7 @@ using namespace std;
std::atomic<int> wav_index(0);
std::mutex mtx;
void runReg(FUNASR_HANDLE asr_handle, vector<string> wav_list, vector<string> wav_ids,
void runReg(FUNASR_HANDLE asr_handle, vector<string> wav_list, vector<string> wav_ids, int audio_fs,
float* total_length, long* total_time, int core_id, float glob_beam = 3.0f, float lat_beam = 3.0f, float am_sc = 10.0f,
int fst_inc_wts = 20, string hotword_path = "") {
@ -55,7 +55,7 @@ void runReg(FUNASR_HANDLE asr_handle, vector<string> wav_list, vector<string> wa
for (size_t i = 0; i < 1; i++)
{
FunOfflineReset(asr_handle, decoder_handle);
FUNASR_RESULT result=FunOfflineInfer(asr_handle, wav_list[0].c_str(), RASR_NONE, NULL, hotwords_embedding, 16000, true, decoder_handle);
FUNASR_RESULT result=FunOfflineInfer(asr_handle, wav_list[0].c_str(), RASR_NONE, NULL, hotwords_embedding, audio_fs, true, decoder_handle);
if(result){
FunASRFreeResult(result);
}
@ -69,7 +69,7 @@ void runReg(FUNASR_HANDLE asr_handle, vector<string> wav_list, vector<string> wa
}
gettimeofday(&start, NULL);
FUNASR_RESULT result=FunOfflineInfer(asr_handle, wav_list[i].c_str(), RASR_NONE, NULL, hotwords_embedding, 16000, true, decoder_handle);
FUNASR_RESULT result=FunOfflineInfer(asr_handle, wav_list[i].c_str(), RASR_NONE, NULL, hotwords_embedding, audio_fs, true, decoder_handle);
gettimeofday(&end, NULL);
seconds = (end.tv_sec - start.tv_sec);
@ -138,6 +138,7 @@ int main(int argc, char *argv[])
TCLAP::ValueArg<std::string> itn_dir("", ITN_DIR, "the itn model(fst) path, which contains zh_itn_tagger.fst and zh_itn_verbalizer.fst", false, "", "string");
TCLAP::ValueArg<std::string> wav_path("", WAV_PATH, "the input could be: wav_path, e.g.: asr_example.wav; pcm_path, e.g.: asr_example.pcm; wav.scp, kaldi style wav list (wav_id \t wav_path)", true, "", "string");
TCLAP::ValueArg<std::int32_t> audio_fs("", AUDIO_FS, "the sample rate of audio", false, 16000, "int32_t");
TCLAP::ValueArg<std::int32_t> thread_num("", THREAD_NUM, "multi-thread num for rtf", true, 0, "int32_t");
TCLAP::ValueArg<std::string> hotword("", HOTWORD, "the hotword file, one hotword perline, Format: Hotword Weight (could be: 阿里巴巴 20)", false, "", "string");
@ -155,6 +156,7 @@ int main(int argc, char *argv[])
cmd.add(hotword);
cmd.add(fst_inc_wts);
cmd.add(wav_path);
cmd.add(audio_fs);
cmd.add(thread_num);
cmd.parse(argc, argv);
@ -234,7 +236,7 @@ int main(int argc, char *argv[])
}
for (int i = 0; i < rtf_threds; i++)
{
threads.emplace_back(thread(runReg, asr_handle, wav_list, wav_ids, &total_length, &total_time, i, glob_beam, lat_beam, am_sc, value_bias, hotword_path));
threads.emplace_back(thread(runReg, asr_handle, wav_list, wav_ids, audio_fs.getValue(), &total_length, &total_time, i, glob_beam, lat_beam, am_sc, value_bias, hotword_path));
}
for (auto& thread : threads)

4
runtime/onnxruntime/bin/funasr-onnx-offline-vad.cpp
View File

@ -68,10 +68,12 @@ int main(int argc, char *argv[])
TCLAP::ValueArg<std::string> quantize("", QUANTIZE, "true (Default), load the model of model.onnx in model_dir. If set true, load the model of model_quant.onnx in model_dir", false, "true", "string");
TCLAP::ValueArg<std::string> wav_path("", WAV_PATH, "the input could be: wav_path, e.g.: asr_example.wav; pcm_path, e.g.: asr_example.pcm; wav.scp, kaldi style wav list (wav_id \t wav_path)", true, "", "string");
TCLAP::ValueArg<std::int32_t> audio_fs("", AUDIO_FS, "the sample rate of audio", false, 16000, "int32_t");
cmd.add(model_dir);
cmd.add(quantize);
cmd.add(wav_path);
cmd.add(audio_fs);
cmd.parse(argc, argv);
std::map<std::string, std::string> model_path;
@ -131,7 +133,7 @@ int main(int argc, char *argv[])
auto& wav_file = wav_list[i];
auto& wav_id = wav_ids[i];
gettimeofday(&start, NULL);
FUNASR_RESULT result=FsmnVadInfer(vad_hanlde, wav_file.c_str(), NULL, 16000);
FUNASR_RESULT result=FsmnVadInfer(vad_hanlde, wav_file.c_str(), NULL, audio_fs.getValue());
gettimeofday(&end, NULL);
seconds = (end.tv_sec - start.tv_sec);
taking_micros += ((seconds * 1000000) + end.tv_usec) - (start.tv_usec);

4
runtime/onnxruntime/bin/funasr-onnx-offline.cpp
View File

@ -57,6 +57,7 @@ int main(int argc, char** argv)
TCLAP::ValueArg<std::int32_t> fst_inc_wts("", FST_INC_WTS, "the fst hotwords incremental bias", false, 20, "int32_t");
TCLAP::ValueArg<std::string> itn_dir("", ITN_DIR, "the itn model(fst) path, which contains zh_itn_tagger.fst and zh_itn_verbalizer.fst", false, "", "string");
TCLAP::ValueArg<std::string> wav_path("", WAV_PATH, "the input could be: wav_path, e.g.: asr_example.wav; pcm_path, e.g.: asr_example.pcm; wav.scp, kaldi style wav list (wav_id \t wav_path)", true, "", "string");
TCLAP::ValueArg<std::int32_t> audio_fs("", AUDIO_FS, "the sample rate of audio", false, 16000, "int32_t");
TCLAP::ValueArg<std::string> hotword("", HOTWORD, "the hotword file, one hotword perline, Format: Hotword Weight (could be: 阿里巴巴 20)", false, "", "string");
cmd.add(model_dir);
@ -72,6 +73,7 @@ int main(int argc, char** argv)
cmd.add(am_scale);
cmd.add(fst_inc_wts);
cmd.add(wav_path);
cmd.add(audio_fs);
cmd.add(hotword);
cmd.parse(argc, argv);
@ -157,7 +159,7 @@ int main(int argc, char** argv)
auto& wav_file = wav_list[i];
auto& wav_id = wav_ids[i];
gettimeofday(&start, NULL);
FUNASR_RESULT result=FunOfflineInfer(asr_hanlde, wav_file.c_str(), RASR_NONE, NULL, hotwords_embedding, 16000, true, decoder_handle);
FUNASR_RESULT result=FunOfflineInfer(asr_hanlde, wav_file.c_str(), RASR_NONE, NULL, hotwords_embedding, audio_fs.getValue(), true, decoder_handle);
gettimeofday(&end, NULL);
seconds = (end.tv_sec - start.tv_sec);
taking_micros += ((seconds * 1000000) + end.tv_usec) - (start.tv_usec);

4
runtime/onnxruntime/bin/funasr-onnx-online-asr.cpp
View File

@ -49,10 +49,12 @@ int main(int argc, char *argv[])
TCLAP::ValueArg<std::string> quantize("", QUANTIZE, "true (Default), load the model of model.onnx in model_dir. If set true, load the model of model_quant.onnx in model_dir", false, "true", "string");
TCLAP::ValueArg<std::string> wav_path("", WAV_PATH, "the input could be: wav_path, e.g.: asr_example.wav; pcm_path, e.g.: asr_example.pcm; wav.scp, kaldi style wav list (wav_id \t wav_path)", true, "", "string");
TCLAP::ValueArg<std::int32_t> audio_fs("", AUDIO_FS, "the sample rate of audio", false, 16000, "int32_t");
cmd.add(model_dir);
cmd.add(quantize);
cmd.add(wav_path);
cmd.add(audio_fs);
cmd.parse(argc, argv);
std::map<std::string, std::string> model_path;
@ -143,7 +145,7 @@ int main(int argc, char *argv[])
is_final = false;
}
gettimeofday(&start, NULL);
FUNASR_RESULT result = FunASRInferBuffer(online_handle, speech_buff+sample_offset, step, RASR_NONE, NULL, is_final, 16000);
FUNASR_RESULT result = FunASRInferBuffer(online_handle, speech_buff+sample_offset, step, RASR_NONE, NULL, is_final, audio_fs.getValue());
gettimeofday(&end, NULL);
seconds = (end.tv_sec - start.tv_sec);
taking_micros += ((seconds * 1000000) + end.tv_usec) - (start.tv_usec);

10
runtime/onnxruntime/bin/funasr-onnx-online-rtf.cpp
View File

@ -38,7 +38,7 @@ bool is_target_file(const std::string& filename, const std::string target) {
return (extension == target);
}
void runReg(FUNASR_HANDLE asr_handle, vector<string> wav_list, vector<string> wav_ids,
void runReg(FUNASR_HANDLE asr_handle, vector<string> wav_list, vector<string> wav_ids, int audio_fs,
float* total_length, long* total_time, int core_id) {
struct timeval start, end;
@ -84,7 +84,7 @@ void runReg(FUNASR_HANDLE asr_handle, vector<string> wav_list, vector<string> wa
} else {
is_final = false;
}
FUNASR_RESULT result = FunASRInferBuffer(online_handle, speech_buff+sample_offset, step, RASR_NONE, NULL, is_final, 16000);
FUNASR_RESULT result = FunASRInferBuffer(online_handle, speech_buff+sample_offset, step, RASR_NONE, NULL, is_final, audio_fs);
if (result)
{
FunASRFreeResult(result);
@ -131,7 +131,7 @@ void runReg(FUNASR_HANDLE asr_handle, vector<string> wav_list, vector<string> wa
is_final = false;
}
gettimeofday(&start, NULL);
FUNASR_RESULT result = FunASRInferBuffer(online_handle, speech_buff+sample_offset, step, RASR_NONE, NULL, is_final, 16000);
FUNASR_RESULT result = FunASRInferBuffer(online_handle, speech_buff+sample_offset, step, RASR_NONE, NULL, is_final, audio_fs);
gettimeofday(&end, NULL);
seconds = (end.tv_sec - start.tv_sec);
long taking_micros = ((seconds * 1000000) + end.tv_usec) - (start.tv_usec);
@ -186,6 +186,7 @@ int main(int argc, char *argv[])
TCLAP::ValueArg<std::string> punc_quant("", PUNC_QUANT, "true (Default), load the model of model.onnx in punc_dir. If set true, load the model of model_quant.onnx in punc_dir", false, "true", "string");
TCLAP::ValueArg<std::string> wav_path("", WAV_PATH, "the input could be: wav_path, e.g.: asr_example.wav; pcm_path, e.g.: asr_example.pcm; wav.scp, kaldi style wav list (wav_id \t wav_path)", true, "", "string");
TCLAP::ValueArg<std::int32_t> audio_fs("", AUDIO_FS, "the sample rate of audio", false, 16000, "int32_t");
TCLAP::ValueArg<std::int32_t> thread_num("", THREAD_NUM, "multi-thread num for rtf", true, 0, "int32_t");
cmd.add(model_dir);
@ -195,6 +196,7 @@ int main(int argc, char *argv[])
cmd.add(punc_dir);
cmd.add(punc_quant);
cmd.add(wav_path);
cmd.add(audio_fs);
cmd.add(thread_num);
cmd.parse(argc, argv);
@ -260,7 +262,7 @@ int main(int argc, char *argv[])
int rtf_threds = thread_num.getValue();
for (int i = 0; i < rtf_threds; i++)
{
threads.emplace_back(thread(runReg, asr_handle, wav_list, wav_ids, &total_length, &total_time, i));
threads.emplace_back(thread(runReg, asr_handle, wav_list, wav_ids, audio_fs.getValue(), &total_length, &total_time, i));
}
for (auto& thread : threads)

4
runtime/onnxruntime/bin/funasr-onnx-online-vad.cpp
View File

@ -75,10 +75,12 @@ int main(int argc, char *argv[])
TCLAP::ValueArg<std::string> quantize("", QUANTIZE, "false (Default), load the model of model.onnx in model_dir. If set true, load the model of model_quant.onnx in model_dir", false, "true", "string");
TCLAP::ValueArg<std::string> wav_path("", WAV_PATH, "the input could be: wav_path, e.g.: asr_example.wav; pcm_path, e.g.: asr_example.pcm; wav.scp, kaldi style wav list (wav_id \t wav_path)", true, "", "string");
TCLAP::ValueArg<std::int32_t> audio_fs("", AUDIO_FS, "the sample rate of audio", false, 16000, "int32_t");
cmd.add(model_dir);
cmd.add(quantize);
cmd.add(wav_path);
cmd.add(audio_fs);
cmd.parse(argc, argv);
std::map<std::string, std::string> model_path;
@ -170,7 +172,7 @@ int main(int argc, char *argv[])
is_final = false;
}
gettimeofday(&start, NULL);
FUNASR_RESULT result = FsmnVadInferBuffer(online_hanlde, speech_buff+sample_offset, step, NULL, is_final, 16000);
FUNASR_RESULT result = FsmnVadInferBuffer(online_hanlde, speech_buff+sample_offset, step, NULL, is_final, audio_fs.getValue());
gettimeofday(&end, NULL);
seconds = (end.tv_sec - start.tv_sec);
taking_micros += ((seconds * 1000000) + end.tv_usec) - (start.tv_usec);

1
runtime/onnxruntime/include/com-define.h
View File

@ -34,6 +34,7 @@ namespace funasr {
#define THREAD_NUM "thread-num"
#define PORT_ID "port-id"
#define HOTWORD_SEP " "
#define AUDIO_FS "audio-fs"
// #define VAD_MODEL_PATH "vad-model"
// #define VAD_CMVN_PATH "vad-cmvn"

2
runtime/onnxruntime/src/audio.cpp
View File

@ -245,7 +245,7 @@ void Audio::WavResample(int32_t sampling_rate, const float *waveform,
{
LOG(INFO) << "Creating a resampler:\n"
<< " in_sample_rate: "<< sampling_rate << "\n"
<< " output_sample_rate: " << static_cast<int32_t>(MODEL_SAMPLE_RATE);
<< " output_sample_rate: " << static_cast<int32_t>(dest_sample_rate);
float min_freq =
std::min<int32_t>(sampling_rate, dest_sample_rate);
float lowpass_cutoff = 0.99 * 0.5 * min_freq;

2
runtime/onnxruntime/src/paraformer.cpp
View File

@ -466,7 +466,7 @@ string Paraformer::Forward(float* din, int len, bool input_finished, const std::
int32_t in_feat_dim = fbank_opts_.mel_opts.num_bins;
std::vector<std::vector<float>> asr_feats;
FbankKaldi(MODEL_SAMPLE_RATE, din, len, asr_feats);
FbankKaldi(asr_sample_rate, din, len, asr_feats);
if(asr_feats.size() == 0){
return "";
}

2
runtime/websocket/bin/websocket-server-2pass.cpp
View File

@ -232,7 +232,7 @@ void WebSocketServer::on_open(websocketpp::connection_hdl hdl) {
data_msg->msg["wav_name"] = "wav-default-id";
data_msg->msg["mode"] = "2pass";
data_msg->msg["itn"] = true;
data_msg->msg["audio_fs"] = 16000;
data_msg->msg["audio_fs"] = 16000; // default is 16k
data_msg->msg["access_num"] = 0; // the number of access for this object, when it is 0, we can free it saftly
data_msg->msg["is_eof"]=false; // if this connection is closed
data_msg->punc_cache =

2
runtime/websocket/bin/websocket-server.cpp
View File

@ -144,7 +144,7 @@ void WebSocketServer::on_open(websocketpp::connection_hdl hdl) {
data_msg->msg["wav_format"] = "pcm";
data_msg->msg["wav_name"] = "wav-default-id";
data_msg->msg["itn"] = true;
data_msg->msg["audio_fs"] = 16000;
data_msg->msg["audio_fs"] = 16000; // default is 16k
data_msg->msg["access_num"] = 0; // the number of access for this object, when it is 0, we can free it saftly
data_msg->msg["is_eof"]=false;
FUNASR_DEC_HANDLE decoder_handle =