构建简单的智能客服系统（三）——基于 UniMRCP 实现讯飞 TTS MRCP Server

article/2025/8/22 12:44:47

MRCP V2
上一篇我们封装了科大讯飞接口实现了 UniMRCP ASR Plugin，这篇文章我们再简单说一下 TTS 的实现。

MRCP plugin

以下内容请参考《基于 UniMRCP 实现讯飞 ASR MRCP Server》：

UniMRCP 的编译、安装运行
UniMRCP plugin 的加载、调用流程
UniMRCP plugin 的新建
讯飞 SDK 的导入

调用讯飞 API 实现 plugin

引用头文件

#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <unistd.h>#include "qtts.h"
#include "msp_cmn.h"
#include "msp_errors.h"
#include "mpf_buffer.h"

channel 新增变量

新增 buffer 用于存放 TTS 转化后的语音：

struct xfyun_synth_channel_t {
...mpf_buffer_t          *audio_buffer;
}

因为编写 ASR plugin 的时候我们已经调用过，这里可以省略。

语音合成

跟 ASR 不同，TTS 的请求一下子就发送过来，没有一个长时间处理语音流的过程。所以我们把 session 的创建销毁直接放在一个处理过程中即可：

static apt_bool_t xfyun_synth_text_to_speech(const char* src_text, const char* params, mpf_buffer_t *buffer) {int ret = -1;const char*  sessionID = NULL;int synth_status = MSP_TTS_FLAG_STILL_HAVE_DATA;unsigned int audio_len = 0;sessionID = QTTSSessionBegin(params, &ret);if (MSP_SUCCESS != ret){apt_log(APT_LOG_MARK, APT_PRIO_WARNING,"[xfyun] QTTSSessionBegin failed, error code: %d.", ret);return FALSE;}ret = QTTSTextPut(sessionID, src_text, (unsigned int)strlen(src_text), NULL);if (MSP_SUCCESS != ret){apt_log(APT_LOG_MARK, APT_PRIO_WARNING,"[xfyun] QTTSTextPut failed, error code: %d.",ret);QTTSSessionEnd(sessionID, "TextPutError");return FALSE;}apt_log(APT_LOG_MARK, APT_PRIO_WARNING,"[xfyun] 正在合成 ...");while (1) {/* 获取合成音频 */const void* data = QTTSAudioGet(sessionID, &audio_len, &synth_status, &ret);if (MSP_SUCCESS != ret)break;if (NULL != data){mpf_buffer_audio_write(buffer, data, audio_len);}if (MSP_TTS_FLAG_DATA_END == synth_status)break;usleep(150*1000); //防止频繁占用CPU}if (MSP_SUCCESS != ret){apt_log(APT_LOG_MARK, APT_PRIO_WARNING,"[xfyun] QTTSAudioGet failed, error code: %d.",ret);QTTSSessionEnd(sessionID, "AudioGetError");return FALSE;}/* 合成完毕 */ret = QTTSSessionEnd(sessionID, "Normal");if (MSP_SUCCESS != ret){apt_log(APT_LOG_MARK, APT_PRIO_WARNING,"[xfyun] QTTSSessionEnd failed, error code: %d.",ret);return FALSE;}return TRUE;
}

xfyun_synth_channel_request_dispatch处理 SYNTHESIZER_SPEAK 消息时调用转换接口。转换完成后，调用如下接口触发媒体流事件：

1	mpf_buffer_event_write(synth_channel->audio_buffer, MEDIA_FRAME_TYPE_EVENT);

传递语音流

xfyun_synth_stream_read中读取 audio_buffer 中的语音流发送给客户端：

static apt_bool_t xfyun_synth_stream_read(mpf_audio_stream_t *stream, mpf_frame_t *frame)
{
...mpf_buffer_frame_read(synth_channel->audio_buffer,frame);
...
}