ROS 科大讯飞语音（三）识别篇
环境准备
sudo apt-get install libasound2-dev
- 将demo里面的call.bnf拷贝到/Robot/voice/call.bnf
编写节点
语音识别节点代码
- 在xf_voice/src文件夹下新建xf_asr.cpp文件，并复制以下代码。
- 注意，请将appid改为你自己在官网申请的。
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <errno.h>
#include <unistd.h>
#include "ros/ros.h"
#include "std_msgs/String.h"
#include "pthread.h"
#include "/Robot/voice/inc/qisr.h"
#include "/Robot/voice/inc/msp_cmn.h"
#include "/Robot/voice/inc/msp_errors.h"
#include <stdio.h>
#include <stdlib.h>
#include <alsa/asoundlib.h>
#define SAMPLE_RATE 16000
#define CHANNLE 1
#define FRAMES_SIZE 3200
#define FORMAT SND_PCM_FORMAT_S16_LE
#define PER_SAMPLE 2
#define DEVICE "default"
char data[10000];
int flag_begin=0;
int flag_understand=0;
int flag_unknow=0;
/* wav音频头部格式 */
typedef struct _wave_pcm_hdr
{
char riff[4]; // = "RIFF"
int size_8; // = FileSize - 8
char wave[4]; // = "WAVE"
char fmt[4]; // = "fmt "
int fmt_size; // = 下一个结构体的大小 : 16
short int format_tag; // = PCM : 1
short int channels; // = 通道数 : 1
int samples_per_sec; // = 采样率 : 8000 | 6000 | 11025 | 16000
int avg_bytes_per_sec; // = 每秒字节数 : samples_per_sec * bits_per_sample / 8
short int block_align; // = 每采样点字节数 : wBitsPerSample / 8
short int bits_per_sample; // = 量化比特数: 8 | 16
char data[4]; // = "data";
int data_size; // = 纯数据长度 : FileSize - 44
} wave_pcm_hdr;
/* 默认wav音频头部数据 */
wave_pcm_hdr default_wav_hdr =
{
{ 'R', 'I', 'F', 'F' },
0,
{'W', 'A', 'V', 'E'},
{'f', 'm', 't', ' '},
16,
1,
1,
16000,
32000,
2,
16,
{'d', 'a', 't', 'a'},
0
};
#define SAMPLE_RATE_16K (16000)
#define SAMPLE_RATE_8K (8000)
#define MAX_GRAMMARID_LEN (32)
#define MAX_PARAMS_LEN (1024)
const char * ASR_RES_PATH = "fo|/Robot/voice/bin/msc/res/asr/common.jet"; //离线语法识别资源路径
const char * GRM_BUILD_PATH = "/Robot/voice/bin/msc/res/asr/GrmBuilld"; //构建离线语法识别网络生成数据保存路径
const char * GRM_FILE = "/Robot/voice/call.bnf"; //构建离线识别语法网络所用的语法文件
const char * listened_file ="/Robot/voice/wav/ddhgdw.pcm";
typedef struct _UserData {
int build_fini; //标识语法构建是否完成
int update_fini; //标识更新词典是否完成
int errcode; //记录语法构建或更新词典回调错误码
char grammar_id[MAX_GRAMMARID_LEN]; //保存语法构建返回的语法ID
}UserData;
UserData asr_data;
int build_grammar(UserData *udata); //构建离线识别语法网络
int run_asr(void *udata); //进行离线语法识别
int build_grm_cb(int ecode, const char *info, void *udata)
{
UserData *grm_data = (UserData *)udata;
if (NULL != grm_data) {
grm_data->build_fini = 1;
grm_data->errcode = ecode;
}
if (MSP_SUCCESS == ecode && NULL != info) {
printf("构建语法成功！ 语法ID:%s\n", info);
if (NULL != grm_data)
snprintf(grm_data->grammar_id, MAX_GRAMMARID_LEN - 1, info);
}
else
printf("构建语法失败！%d\n", ecode);
return 0;
}
int build_grammar(UserData *udata)
{
FILE *grm_file = NULL;
char *grm_content = NULL;
unsigned int grm_cnt_len = 0;
char grm_build_params[MAX_PARAMS_LEN] = {NULL};
int ret = 0;
grm_file = fopen(GRM_FILE, "rb");
if(NULL == grm_file) {
printf("打开\"%s\"文件失败！[%s]\n", GRM_FILE, strerror(errno));
return -1;
}
fseek(grm_file, 0, SEEK_END);
grm_cnt_len = ftell(grm_file);
fseek(grm_file, 0, SEEK_SET);
grm_content = (char *)malloc(grm_cnt_len + 1);
if (NULL == grm_content)
{
printf("内存分配失败!\n");
fclose(grm_file);
grm_file = NULL;
return -1;
}
fread((void*)grm_content, 1, grm_cnt_len, grm_file);
grm_content[grm_cnt_len] = '\0';
fclose(grm_file);
grm_file = NULL;
snprintf(grm_build_params, MAX_PARAMS_LEN - 1,
"engine_type = local, \
asr_res_path = %s, sample_rate = %d, \
grm_build_path = %s, ",
ASR_RES_PATH,
SAMPLE_RATE_16K,
GRM_BUILD_PATH
);
ret = QISRBuildGrammar("bnf", grm_content, grm_cnt_len, grm_build_params, build_grm_cb, udata);
free(grm_content);
grm_content = NULL;
return ret;
}
int recode_asr(void *ptr)
{
char asr_params[MAX_PARAMS_LEN] = {NULL};
const char *rec_rslt = NULL;
const char *session_id = NULL;
FILE *f_pcm = NULL;
char *pcm_data = NULL;
long pcm_count = 0;
int aud_stat = MSP_AUDIO_SAMPLE_CONTINUE;
int ep_status = MSP_EP_LOOKING_FOR_SPEECH;
int rec_status = MSP_REC_STATUS_INCOMPLETE;
int rss_status = MSP_REC_STATUS_INCOMPLETE;
int errcode = -1;
long loops;
int rc, size;
float time=5;
snd_pcm_t *handle;
snd_pcm_hw_params_t *params;
snd_pcm_uframes_t frames, ret;
char *buffer;
char *ptr_buffer;
FILE *fp = fopen("/Robot/voice/wav/listened.wav", "wb");
//离线语法识别参数设置
snprintf(asr_params, MAX_PARAMS_LEN - 1,
"engine_type = cloud,\
sub=iat,\
domain = iat,\
sch=1,\
nlp_version=2.0,\
mixed_type=delay ,\
mixed_threshold=50,\
mixed_timeout= 3000 ,\
asr_res_path = %s, sample_rate = %d, \
grm_build_path = %s, local_grammar = %s, \
result_type = json, result_encoding = UTF-8 ",
ASR_RES_PATH,
SAMPLE_RATE_16K,
GRM_BUILD_PATH,
asr_data.grammar_id
);
session_id = QISRSessionBegin(NULL, asr_params, &errcode);
printf("开始识别...\n");
if(fp == NULL){ }else(printf("open file success"));
rc = snd_pcm_open(&handle, DEVICE, SND_PCM_STREAM_CAPTURE, 0);
if (rc < 0) { } else printf("OK:before alloca\n");
snd_pcm_hw_params_alloca(¶ms);
printf("OK:after alloca\n");
rc = snd_pcm_hw_params_any(handle, params);
if (rc < 0) { } else{ printf("OK:snd_pcm_hw_params_an()\n"); }
rc = snd_pcm_hw_params_set_access(handle, params,SND_PCM_ACCESS_RW_INTERLEAVED);
if (rc < 0) { } else{ printf("OK:snd_pcm_hw_params_set_access()\n"); }
rc = snd_pcm_hw_params_set_format(handle, params, FORMAT);
if (rc < 0) { } else{ printf("OK:snd_pcm_hw_params_set_format()\n"); }
rc = snd_pcm_hw_params_set_channels(handle, params, CHANNLE);
if (rc < 0) { } else{ printf("OK:snd_pcm_hw_params_set_channels()\n"); }
rc = snd_pcm_hw_params_set_rate(handle, params,SAMPLE_RATE, 0);
if (rc < 0) { } else{ printf("OK:snd_pcm_hw_params_set_rate()\n"); }
frames = FRAMES_SIZE;
rc = snd_pcm_hw_params(handle, params);
if (rc < 0) { } else{ printf("OK:snd_pcm_hw_paraams()\n"); }
size = frames * PER_SAMPLE *CHANNLE; /* 2 bytes/sample, 1 channels */
ptr_buffer = buffer = (char *) malloc(size);
if(buffer == NULL){ }else{ printf("OK:malloc()\n"); }
loops = SAMPLE_RATE/frames*time;
fwrite(&default_wav_hdr, sizeof(default_wav_hdr) ,1, fp);
while (loops > 0)
{
loops--;
ret = snd_pcm_readi(handle, ptr_buffer, frames);
if (ret == -EPIPE) {
printf( "overrun occurred\n");
snd_pcm_prepare(handle);
} else if (ret < 0) printf("error from read: %s\n",snd_strerror(ret));
else if (ret != frames) printf( "short read, read %d frames\n", ret);
rc = fwrite(ptr_buffer, size, 1, fp);
if (rc < 0){ printf("error in write\n"); }
if (rc != 1) printf("failed to write %d bytes\n",size);
printf(">");
fflush(stdout);
if (0 == pcm_count) aud_stat = MSP_AUDIO_SAMPLE_FIRST;
else aud_stat = MSP_AUDIO_SAMPLE_CONTINUE;
errcode = QISRAudioWrite(session_id, (const void *)ptr_buffer, size, aud_stat, &ep_status, &rec_status);
if (MSP_EP_AFTER_SPEECH == ep_status)break; //检测到音频结束
}
QISRAudioWrite(session_id, (const void *)NULL, 0, MSP_AUDIO_SAMPLE_LAST, &ep_status, &rec_status);
if(buffer != NULL) free(buffer);
if(fp != NULL) fclose(fp);
snd_pcm_drain(handle);
snd_pcm_close(handle);
free(pcm_data);
pcm_data = NULL;
//获取识别结果
while (MSP_REC_STATUS_COMPLETE != rss_status && MSP_SUCCESS == errcode) {
rec_rslt = QISRGetResult(session_id, &rss_status, 0, &errcode);
printf(".");
usleep(150 * 1000);
}
printf("\n识别结束%d：\n",loops);
printf("=============================================================\n");
if (NULL != rec_rslt)
{
printf("%s\n", rec_rslt);
sprintf(data,"%s\n", rec_rslt);
flag_understand=1;
}
else
{
printf("没有识别结果！\n");
flag_unknow=1;
}
printf("=============================================================\n");
goto run_exit;
if (NULL != pcm_data) {
free(pcm_data);
pcm_data = NULL;
}
if (NULL != f_pcm) {
fclose(f_pcm);
f_pcm = NULL;
}
run_exit:
printf("exit with code :%d..\n",errcode);
QISRSessionEnd(session_id, NULL);
return errcode;
}
void wakeupcallback(const std_msgs::String::ConstPtr& msg)
{
std::cout<<"waking up"<<std::endl;
// printf("%s", *msg->data.c_str());
usleep(700*1000);
flag_begin=1;
}
int main(int argc,char **argv)
{
const char *login_config = "appid = 573bdbff"; //登录参数
int ret = 0 ;
ret = MSPLogin(NULL, NULL, login_config); //第一个参数为用户名，第二个参数为密码，传NULL即可，第三个参数是登录参数
if (MSP_SUCCESS != ret) {
printf("登录失败：%d\n", ret);
return -1;
}
printf("构建离线识别语法网络...\n");
ret = build_grammar(&asr_data); //第一次使用某语法进行识别，需要先构建语法网络，获取语法ID，之后使用此语法进行识别，无需再次构建
if (MSP_SUCCESS != ret) {
printf("构建语法调用失败！\n");
return -1;
}
while (1 != asr_data.build_fini) usleep(300 * 1000);
if (MSP_SUCCESS != asr_data.errcode) return -1;
//printf("离线识别语法网络构建完成，开始识别...\n");
if (MSP_SUCCESS != ret) {
printf("离线语法识别出错: %d \n", ret);
return -1;
}
ros::init(argc, argv, "xf_asr");
ros::NodeHandle n;
ros::Rate loop_rate(10);
ros::Subscriber sub = n.subscribe("xfwakeup", 1000, wakeupcallback);
ros::Publisher pub = n.advertise<std_msgs::String>("xfsaywords", 1000);
ros::Publisher pub2= n.advertise<std_msgs::String>("xfunderstand",10000);
while (ros::ok())
{
if(flag_begin)
{
ret = recode_asr(&asr_data);
// ret = run_asr(&asr_data);
flag_begin=0;
}
if(flag_unknow)
{
std_msgs::String msg;
std::stringstream ss;
flag_unknow=0;
ss << "对不起，我好像不明白！ ";
msg.data = ss.str();
pub.publish(msg);
}
if(flag_understand)
{
flag_understand=0;
std_msgs::String msg;
std::stringstream ss2;
ss2 << data;
msg.data = ss2.str();
pub2.publish(msg);
}
ros::spinOnce();
loop_rate.sleep();
}
MSPLogout();
printf("请按任意键退出...\n");
getchar();
}
配置CMakeLists.txt
- 在原xf_voice文件夹的CMakeLists.txt文件末尾加上以下代码
add_executable(xf_asr src/xf_asr.cpp)
target_link_libraries(xf_asr ${catkin_LIBRARIES} /Robot/voice/lib/libmsc.so -ldl -lpthread -lm -lrt -lasound)
add_dependencies(xf_asr xf_voice_generate_messages_cpp)
编译测试
编译
- 回到catkin_ws文件夹下，执行catkin_make
$ catkin_make
测试
roscore
$ cd catkin_ws/
$ source devel/setup.sh
$ rosrun xf_voice xf_tts
$ cd catkin_ws/
$ source devel/setup.sh
$ rosrun xf_voice xf_asr
$ cd catkin_ws/
$ source devel/setup.sh
$ rostopic pub /xfwakeup std_msgs/String s
效果
- 每次第四个终端发送一次唤醒信号，在第三个启动语音识别的终端中就会启动一次语音识别过程。
- 超时如果没有识别结果则返回没有识别结果！同时在第二个终端会提示“对不起，我好像不明白”，你也会听到相应的提示音。
- 如果有识别结果，则会停止识别并返回结果。同时结果会被发送到xfunderstand主题。
构建离线识别语法网络...
构建语法成功！ 语法ID:call
waking up
开始识别...
open file successOK:before alloca
OK:after alloca
OK:snd_pcm_hw_params_an()
OK:snd_pcm_hw_params_set_access()
OK:snd_pcm_hw_params_set_format()
OK:snd_pcm_hw_params_set_channels()
OK:snd_pcm_hw_params_set_rate()
OK:snd_pcm_hw_paraams()
OK:malloc()
>>>>>>>>>>>>>>>>>>>>>>>>>..
识别结束0：
=============================================================
没有识别结果！
=============================================================
exit with code :0..
waking up
开始识别...
open file successOK:before alloca
OK:after alloca
OK:snd_pcm_hw_params_an()
OK:snd_pcm_hw_params_set_access()
OK:snd_pcm_hw_params_set_format()
OK:snd_pcm_hw_params_set_channels()
OK:snd_pcm_hw_params_set_rate()
OK:snd_pcm_hw_paraams()
OK:malloc()
>>>>>>>>>>>>>>>>>>>>>>>>>....
识别结束0：
=============================================================
{"rc":0,"operation":"ANSWER","service":"chat","answer":{"text":"医院门口，一老大爷很伤心。好心人上前询问，老大爷答道：孙子生病，拉了坨便便想带来化验。用报纸包了好几层放口袋里，结果在车上被偷了……","type":"T"},"text":"给我讲一个笑话。"}
=============================================================
exit with code :0..
- 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
I heard,I will say:对不起，我好像不明白！
开始合成 ...
正在合成 ...
合成完毕