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

update files.

Browse Source
This commit is contained in:
mayong 2023-03-07 12:15:07 +08:00
parent 37aec4ec07
commit 39529df088
10 changed files with 554 additions and 53 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
funasr/runtime/onnxruntime/CMakeSettings.json
View File

@ -19,8 +19,26 @@
"cmakeCommandArgs": "",
"buildCommandArgs": "",
"ctestCommandArgs": "",
"inheritEnvironments": [ "msvc_x64_x64" ],
"variables": []
"inheritEnvironments": [ "msvc_x64_x64" ]
},
{
"name": "Linux-GCC-Debug",
"generator": "Unix Makefiles",
"configurationType": "Debug",
"cmakeExecutable": "cmake",
"remoteCopySourcesExclusionList": [ ".vs", ".git", "out" ],
"cmakeCommandArgs": "-DONNXRUNTIME_DIR=/data/linux/thirdpart/onnxruntime-linux-x64-1.14.1",
"buildCommandArgs": "",
"ctestCommandArgs": "",
"inheritEnvironments": [ "linux_x64" ],
"remoteMachineName": "${defaultRemoteMachineName}",
"remoteCMakeListsRoot": "$HOME/.vs/${projectDirName}/${workspaceHash}/src",
"remoteBuildRoot": "$HOME/.vs/${projectDirName}/${workspaceHash}/out/build/${name}",
"remoteInstallRoot": "$HOME/.vs/${projectDirName}/${workspaceHash}/out/install/${name}",
"remoteCopySources": true,
"rsyncCommandArgs": "-t --delete",
"remoteCopyBuildOutput": false,
"remoteCopySourcesMethod": "rsync"
}
]
}

7
funasr/runtime/onnxruntime/include/Audio.h
View File

@ -43,11 +43,16 @@ class Audio {
Audio(int data_type, int size);
~Audio();
void disp();
bool loadwav(const char *filename);
bool loadwav(const char* filename);
bool loadwav(const char* buf, int nLen);
bool loadpcmwav(const char* buf, int nFileLen);
bool loadpcmwav(const char* filename);
int fetch_chunck(float *&dout, int len);
int fetch(float *&dout, int &len, int &flag);
void padding();
void split();
int get_queue_size() { return (int)frame_queue.size(); }
};
#endif

95
funasr/runtime/onnxruntime/include/librapidasrapi.h Normal file
View File

@ -0,0 +1,95 @@
#pragma once
#ifdef WIN32
#ifdef _RPASR_API_EXPORT
#define _RAPIDASRAPI __declspec(dllexport)
#else
#define _RAPIDASRAPI __declspec(dllimport)
#endif
#else
#define _RAPIDASRAPI
#endif
#ifndef _WIN32
#define RPASR_CALLBCK_PREFIX __attribute__((__stdcall__))
#else
#define RPASR_CALLBCK_PREFIX __stdcall
#endif
#ifdef __cplusplus
extern "C" {
#endif
typedef void* RPASR_HANDLE;
typedef void* RPASR_RESULT;
typedef unsigned char RPASR_BOOL;
#define RPASR_TRUE 1
#define RPASR_FALSE 0
#define QM_DEFAULT_THREAD_NUM 4
typedef enum
{
RASR_NONE=-1,
RASRM_CTC_GREEDY_SEARCH=0,
RASRM_CTC_RPEFIX_BEAM_SEARCH = 1,
RASRM_ATTENSION_RESCORING = 2,
}RPASR_MODE;
typedef enum {
RPASR_MODEL_PADDLE = 0,
RPASR_MODEL_PADDLE_2 = 1,
RPASR_MODEL_K2 = 2,
RPASR_MODEL_PARAFORMER = 3,
}RPASR_MODEL_TYPE;
typedef void (* QM_CALLBACK)(int nCurStep, int nTotal); // nTotal: total steps; nCurStep: Current Step.
// APIs for qmasr
_RAPIDASRAPI RPASR_HANDLE RapidAsrInit(const char* szModelDir, int nThread);
// if not give a fnCallback ,it should be NULL
_RAPIDASRAPI RPASR_RESULT RapidAsrRecogBuffer(RPASR_HANDLE handle, const char* szBuf, int nLen, RPASR_MODE Mode, QM_CALLBACK fnCallback);
_RAPIDASRAPI RPASR_RESULT RapidAsrRecogPCMBuffer(RPASR_HANDLE handle, const char* szBuf, int nLen, RPASR_MODE Mode, QM_CALLBACK fnCallback);
_RAPIDASRAPI RPASR_RESULT RapidAsrRecogPCMFile(RPASR_HANDLE handle, const char* szFileName, RPASR_MODE Mode, QM_CALLBACK fnCallback);
_RAPIDASRAPI RPASR_RESULT RapidAsrRecogFile(RPASR_HANDLE handle, const char* szWavfile, RPASR_MODE Mode, QM_CALLBACK fnCallback);
_RAPIDASRAPI const char* RapidAsrGetResult(RPASR_RESULT Result,int nIndex);
_RAPIDASRAPI const int RapidAsrGetRetNumber(RPASR_RESULT Result);
_RAPIDASRAPI void RapidAsrFreeResult(RPASR_RESULT Result);
_RAPIDASRAPI void RapidAsrUninit(RPASR_HANDLE Handle);
#ifdef __cplusplus
}
#endif

211
funasr/runtime/onnxruntime/src/Audio.cpp
View File

@ -25,8 +25,7 @@ class AudioWindow {
out_idx = 1;
sum = 0;
};
~AudioWindow()
{
~AudioWindow(){
free(window);
};
int put(int val)
@ -102,6 +101,11 @@ Audio::~Audio()
{
if (speech_buff != NULL) {
free(speech_buff);
}
if (speech_data != NULL) {
free(speech_data);
}
}
@ -115,9 +119,11 @@ void Audio::disp()
bool Audio::loadwav(const char *filename)
{
if (speech_data != NULL) {
free(speech_data);
}
if (speech_buff != NULL) {
free(speech_buff);
free(speech_data);
}
offset = 0;
@ -133,28 +139,191 @@ bool Audio::loadwav(const char *filename)
speech_len = (nFileLen - 44) / 2;
speech_align_len = (int)(ceil((float)speech_len / align_size) * align_size);
speech_buff = (int16_t *)malloc(sizeof(int16_t) * speech_align_len);
memset(speech_buff, 0, sizeof(int16_t) * speech_align_len);
int ret = fread(speech_buff, sizeof(int16_t), speech_len, fp);
fclose(fp);
speech_data = (float *)malloc(sizeof(float) * speech_align_len);
memset(speech_data, 0, sizeof(float) * speech_align_len);
int i;
float scale = 1;
if (speech_buff)
{
memset(speech_buff, 0, sizeof(int16_t) * speech_align_len);
int ret = fread(speech_buff, sizeof(int16_t), speech_len, fp);
fclose(fp);
if (data_type == 1) {
scale = 32768;
speech_data = (float*)malloc(sizeof(float) * speech_align_len);
memset(speech_data, 0, sizeof(float) * speech_align_len);
int i;
float scale = 1;
if (data_type == 1) {
scale = 32768;
}
for (i = 0; i < speech_len; i++) {
speech_data[i] = (float)speech_buff[i] / scale;
}
AudioFrame* frame = new AudioFrame(speech_len);
frame_queue.push(frame);
return true;
}
for (i = 0; i < speech_len; i++) {
speech_data[i] = (float)speech_buff[i] / scale;
}
AudioFrame *frame = new AudioFrame(speech_len);
frame_queue.push(frame);
return true;
else
return false;
}
bool Audio::loadwav(const char* buf, int nFileLen)
{
if (speech_data != NULL) {
free(speech_data);
}
if (speech_buff != NULL) {
free(speech_buff);
}
offset = 0;
size_t nOffset = 0;
#define WAV_HEADER_SIZE 44
speech_len = (nFileLen - WAV_HEADER_SIZE) / 2;
speech_align_len = (int)(ceil((float)speech_len / align_size) * align_size);
speech_buff = (int16_t*)malloc(sizeof(int16_t) * speech_align_len);
if (speech_buff)
{
memset(speech_buff, 0, sizeof(int16_t) * speech_align_len);
memcpy((void*)speech_buff, (const void*)(buf + WAV_HEADER_SIZE), speech_len * sizeof(int16_t));
speech_data = (float*)malloc(sizeof(float) * speech_align_len);
memset(speech_data, 0, sizeof(float) * speech_align_len);
int i;
float scale = 1;
if (data_type == 1) {
scale = 32768;
}
for (i = 0; i < speech_len; i++) {
speech_data[i] = (float)speech_buff[i] / scale;
}
return true;
}
else
return false;
}
bool Audio::loadpcmwav(const char* buf, int nBufLen)
{
if (speech_data != NULL) {
free(speech_data);
}
if (speech_buff != NULL) {
free(speech_buff);
}
offset = 0;
size_t nOffset = 0;
#define WAV_HEADER_SIZE 44
speech_len = nBufLen / 2;
speech_align_len = (int)(ceil((float)speech_len / align_size) * align_size);
speech_buff = (int16_t*)malloc(sizeof(int16_t) * speech_align_len);
if (speech_buff)
{
memset(speech_buff, 0, sizeof(int16_t) * speech_align_len);
memcpy((void*)speech_buff, (const void*)buf, speech_len * sizeof(int16_t));
speech_data = (float*)malloc(sizeof(float) * speech_align_len);
memset(speech_data, 0, sizeof(float) * speech_align_len);
int i;
float scale = 1;
if (data_type == 1) {
scale = 32768;
}
for (i = 0; i < speech_len; i++) {
speech_data[i] = (float)speech_buff[i] / scale;
}
return true;
}
else
return false;
}
bool Audio::loadpcmwav(const char* filename)
{
if (speech_data != NULL) {
free(speech_data);
}
if (speech_buff != NULL) {
free(speech_buff);
}
offset = 0;
FILE* fp;
fp = fopen(filename, "rb");
if (fp == nullptr)
return false;
fseek(fp, 0, SEEK_END);
uint32_t nFileLen = ftell(fp);
fseek(fp, 0, SEEK_SET);
speech_len = (nFileLen) / 2;
speech_align_len = (int)(ceil((float)speech_len / align_size) * align_size);
speech_buff = (int16_t*)malloc(sizeof(int16_t) * speech_align_len);
if (speech_buff)
{
memset(speech_buff, 0, sizeof(int16_t) * speech_align_len);
int ret = fread(speech_buff, sizeof(int16_t), speech_len, fp);
fclose(fp);
speech_data = (float*)malloc(sizeof(float) * speech_align_len);
memset(speech_data, 0, sizeof(float) * speech_align_len);
int i;
float scale = 1;
if (data_type == 1) {
scale = 32768;
}
for (i = 0; i < speech_len; i++) {
speech_data[i] = (float)speech_buff[i] / scale;
}
AudioFrame* frame = new AudioFrame(speech_len);
frame_queue.push(frame);
return true;
}
else
return false;
}
int Audio::fetch_chunck(float *&dout, int len)
{
if (offset >= speech_align_len) {
@ -163,7 +332,7 @@ int Audio::fetch_chunck(float *&dout, int len)
} else if (offset == speech_align_len - len) {
dout = speech_data + offset;
offset = speech_align_len;
// 临时解决
// 临时解决
AudioFrame *frame = frame_queue.front();
frame_queue.pop();
delete frame;

2
funasr/runtime/onnxruntime/src/CMakeLists.txt
View File

@ -18,7 +18,7 @@ if(WIN32)
endif()
target_include_directories(rapidasr PUBLIC ${CMAKE_SOURCE_DIR}/win/include )
target_compile_definitions(rapidasr PUBLIC -D_RPASR_API_EXPORT)
else()
set(EXTRA_LIBS fftw3f webrtcvad pthread)

11
funasr/runtime/onnxruntime/src/commonfunc.h
View File

@ -1,7 +1,18 @@
#pragma once
typedef struct
{
std::string msg;
}RPASR_RECOG_RESULT;
#ifdef _WIN32
#include <codecvt>
inline std::wstring string2wstring(const std::string& str, const std::string& locale)
{
typedef std::codecvt_byname<wchar_t, char, std::mbstate_t> F;

204
funasr/runtime/onnxruntime/src/librapidasrapi.cpp Normal file
View File

@ -0,0 +1,204 @@
#include "precomp.h"
#ifdef __cplusplus
// void __attribute__ ((visibility ("default"))) fun();
extern "C" {
#endif
// APIs for qmasr
_RAPIDASRAPI RPASR_HANDLE RapidAsrInit(const char* szModelDir, int nThreadNum)
{
#ifdef NDEBUG
QMLIC_BOOL bMatched = QmLicCheckValid(QLFM_ASR);
if (!bMatched) {
return nullptr;
}
#endif
Model* mm = create_model(szModelDir, nThreadNum);
return mm;
}
_RAPIDASRAPI RPASR_RESULT RapidAsrRecogBuffer(RPASR_HANDLE handle, const char* szBuf, int nLen, RPASR_MODE Mode, QM_CALLBACK fnCallback)
{
Model* pRecogObj = (Model*)handle;
if (!pRecogObj)
return nullptr;
Audio audio(1);
audio.loadwav(szBuf,nLen);
audio.split();
float* buff;
int len;
int flag=0;
RPASR_RECOG_RESULT* pResult = new RPASR_RECOG_RESULT;
int nStep = 0;
int nTotal = audio.get_queue_size();
while (audio.fetch(buff, len, flag) > 0) {
pRecogObj->reset();
string msg = pRecogObj->forward(buff, len, flag);
pResult->msg += msg;
nStep++;
if (fnCallback)
fnCallback(nStep, nTotal);
}
return pResult;
}
_RAPIDASRAPI RPASR_RESULT RapidAsrRecogPCMBuffer(RPASR_HANDLE handle, const char* szBuf, int nLen, RPASR_MODE Mode, QM_CALLBACK fnCallback)
{
Model* pRecogObj = (Model*)handle;
if (!pRecogObj)
return nullptr;
Audio audio(1);
audio.loadpcmwav(szBuf, nLen);
audio.split();
float* buff;
int len;
int flag = 0;
RPASR_RECOG_RESULT* pResult = new RPASR_RECOG_RESULT;
int nStep = 0;
int nTotal = audio.get_queue_size();
while (audio.fetch(buff, len, flag) > 0) {
pRecogObj->reset();
string msg = pRecogObj->forward(buff, len, flag);
pResult->msg += msg;
nStep++;
if (fnCallback)
fnCallback(nStep, nTotal);
}
return pResult;
}
_RAPIDASRAPI RPASR_RESULT RapidAsrRecogPCMFile(RPASR_HANDLE handle, const char* szFileName, RPASR_MODE Mode, QM_CALLBACK fnCallback)
{
Model* pRecogObj = (Model*)handle;
if (!pRecogObj)
return nullptr;
Audio audio(1);
audio.loadpcmwav(szFileName);
audio.split();
float* buff;
int len;
int flag = 0;
RPASR_RECOG_RESULT* pResult = new RPASR_RECOG_RESULT;
int nStep = 0;
int nTotal = audio.get_queue_size();
while (audio.fetch(buff, len, flag) > 0) {
pRecogObj->reset();
string msg = pRecogObj->forward(buff, len, flag);
pResult->msg += msg;
nStep++;
if (fnCallback)
fnCallback(nStep, nTotal);
}
return pResult;
}
_RAPIDASRAPI RPASR_RESULT RapidAsrRecogFile(RPASR_HANDLE handle, const char* szWavfile, RPASR_MODE Mode, QM_CALLBACK fnCallback)
{
Model* pRecogObj = (Model*)handle;
if (!pRecogObj)
return nullptr;
Audio audio(1);
if(!audio.loadwav(szWavfile))
return nullptr;
audio.split();
float* buff;
int len;
int flag = 0;
int nStep = 0;
int nTotal = audio.get_queue_size();
RPASR_RECOG_RESULT* pResult = new RPASR_RECOG_RESULT;
while (audio.fetch(buff, len, flag) > 0) {
pRecogObj->reset();
string msg = pRecogObj->forward(buff, len, flag);
pResult->msg+= msg;
nStep++;
if (fnCallback)
fnCallback(nStep, nTotal);
}
return pResult;
}
_RAPIDASRAPI const int RapidAsrGetRetNumber(RPASR_RESULT Result)
{
if (!Result)
return 0;
return 1;
}
_RAPIDASRAPI const char* RapidAsrGetResult(RPASR_RESULT Result,int nIndex)
{
RPASR_RECOG_RESULT * pResult = (RPASR_RECOG_RESULT*)Result;
if(!pResult)
return nullptr;
return pResult->msg.c_str();
}
_RAPIDASRAPI void RapidAsrFreeResult(RPASR_RESULT Result)
{
if (Result)
{
delete (RPASR_RECOG_RESULT*)Result;
}
}
_RAPIDASRAPI void RapidAsrUninit(RPASR_HANDLE handle)
{
Model* pRecogObj = (Model*)handle;
if (!pRecogObj)
return;
delete pRecogObj;
}
#ifdef __cplusplus
}
#endif

3
funasr/runtime/onnxruntime/src/precomp.h
View File

@ -41,9 +41,10 @@ using namespace std;
#include "FeatureExtract.h"
#include "FeatureQueue.h"
#include "SpeechWrap.h"
#include <Audio.h>
#include "Model.h"
#include "paraformer_onnx.h"
#include "librapidasrapi.h"
using namespace paraformer;

2
funasr/runtime/onnxruntime/tester/CMakeLists.txt
View File

@ -15,6 +15,6 @@ include_directories(${CMAKE_SOURCE_DIR}/include)
set(EXECNAME "tester")
add_executable(${EXECNAME} "tester.cpp")
target_link_libraries(${EXECNAME} PUBLIC onnxruntime ${EXTRA_LIBS})
target_link_libraries(${EXECNAME} PUBLIC ${EXTRA_LIBS})

50
funasr/runtime/onnxruntime/tester/tester.cpp
View File

@ -1,12 +1,13 @@
#include <iostream>
#ifndef _WIN32
#include <sys/time.h>
#else
#include <win_func.h>
#endif
#include <Audio.h>
#include <Model.h>
#include "librapidasrapi.h"
#include <iostream>
using namespace std;
@ -21,52 +22,49 @@ int main(int argc, char *argv[])
struct timeval start, end;
gettimeofday(&start, NULL);
int nThreadNum = 4;
Model* mm = create_model(argv[1], nThreadNum);
if (!mm)
RPASR_HANDLE AsrHanlde=RapidAsrInit(argv[1], nThreadNum);
if (!AsrHanlde)
{
printf("Cannot load ASR Model from: %s, there must be files model.onnx and vocab.txt", argv[1]);
exit(-1);
}
Audio audio(0);
if (!audio.loadwav(argv[2]))
{
printf("cannot load %s\n", argv[2]);
return -1;
}
audio.disp();
gettimeofday(&end, NULL);
long seconds = (end.tv_sec - start.tv_sec);
long micros = ((seconds * 1000000) + end.tv_usec) - (start.tv_usec);
printf("Model initialization takes %lfs.\n", (double)micros / 1000000);
audio.split();
setbuf(stdout, NULL);
cout << "Result: \"";
gettimeofday(&start, NULL);
float *buff;
int len;
int flag;
while (audio.fetch(buff, len, flag) > 0) {
mm->reset();
string msg = mm->forward(buff, len, flag);
cout << msg;
}
RPASR_RESULT Result=RapidAsrRecogPCMFile(AsrHanlde, argv[2], RASR_NONE, NULL);
gettimeofday(&end, NULL);
cout << "\"." << endl;
if (Result)
{
string msg = RapidAsrGetResult(Result, 0);
cout << msg << endl;
cout << "\"." << endl;
RapidAsrFreeResult(Result);
}
else
{
cout <<("no return data!");
}
seconds = (end.tv_sec - start.tv_sec);
long taking_micros = ((seconds * 1000000) + end.tv_usec) - (start.tv_usec);
printf("Model inference takes %lfs.\n", (double)micros / 1000000);
printf("Model inference RTF: %04lf.\n", (double)taking_micros/micros );
delete mm;
RapidAsrUninit(AsrHanlde);
return 0;
}