利用Waveform函数播放PCM数据流

前言

由于近期需要做一些语音合成的工作，因此，需要进行对语音合成的数据进行实时播放，到网上找了一下资料，参考MSDN的相关说明，写下了如下一个PCM播放数据流的类，多说无益，直接上码：

头文件 pcmspeaker.h

#pragma once #include <Windows.h>  #include "mmsystem.h"  #pragma comment(lib, "winmm.lib")  #define DEF_MAX_BUFFER_SIZE (1024 * 16) #define DEF_MAX_BUFFER_COUNT 16 class CPcmSpeaker { public:  CPcmSpeaker(int bufferSize = DEF_MAX_BUFFER_SIZE, int bufferCnt = DEF_MAX_BUFFER_COUNT);  ~CPcmSpeaker();  int init(int channels, int samplePerSec, int bitsPerSample);  //添加PCM音频数据，等待播放  int toSpeaker(const void *data, int len, int timeout = INFINITE);  int clearPcmData(); private:  typedef struct  {   WAVEHDR header;   char *data;  }WaveHeadandData;  int m_maxBufferSize;  int m_maxBufferCnt;  WaveHeadandData *m_headAndDatas;  static void CALLBACK waveOutProc(HWAVEOUT hwo, UINT uMsg, DWORD dwInstance, DWORD dwParam1, DWORD dwParam2);  int writeToWave(const void *data, int len);  int pcmtoWave(const void *data, int len, int timeout = INFINITE);  // 公共信息  WAVEFORMATEX m_waveFormat;  HWAVEOUT m_hWaveOut; // WAVEOUT句柄  HANDLE m_hBufferEvent;  CRITICAL_SECTION m_BufferOpCriticalSection; };  

实现文件 pcmspeaker.cpp

#include "PcmSpeaker.h" CPcmSpeaker::CPcmSpeaker(int bufferSize, int bufferCnt) {  m_hWaveOut = NULL;  m_hBufferEvent = CreateEvent(NULL, FALSE, FALSE, NULL);  InitializeCriticalSection(&m_BufferOpCriticalSection);  //申请内存  m_headAndDatas = new WaveHeadandData[bufferCnt];  for (int i = 0; i < bufferCnt; i++)  {   memset(&m_headAndDatas[i].header, 0, sizeof(WAVEHDR));   m_headAndDatas[i].header.dwFlags = WHDR_DONE;   m_headAndDatas[i].data = new char[bufferSize];  }  m_maxBufferSize = bufferSize;  m_maxBufferCnt = bufferCnt; } CPcmSpeaker::~CPcmSpeaker() {  //关闭Wave  if (m_hWaveOut != NULL)  {   clearPcmData();   waveOutClose(m_hWaveOut);   m_hWaveOut = NULL;  }  //关闭一些句柄  CloseHandle(m_hBufferEvent);  //删除临界区  DeleteCriticalSection(&m_BufferOpCriticalSection);  //释放内存  for (int i = 0; i < m_maxBufferCnt; i++)   delete[] m_headAndDatas[i].data;  delete[] m_headAndDatas; } int CPcmSpeaker::init(int channels, int samplePerSec, int bitsPerSample) {  if (m_hWaveOut != NULL) {   return 0;// 已经进行了初始化  }  // 第一步: 获取waveformat信息  m_waveFormat.wFormatTag = WAVE_FORMAT_PCM;  m_waveFormat.nChannels = channels;  m_waveFormat.wBitsPerSample = bitsPerSample;  m_waveFormat.nSamplesPerSec = samplePerSec;  m_waveFormat.nBlockAlign =   m_waveFormat.nChannels * m_waveFormat.wBitsPerSample / 8;  m_waveFormat.nAvgBytesPerSec =   m_waveFormat.nSamplesPerSec * m_waveFormat.nBlockAlign;  m_waveFormat.cbSize = sizeof(m_waveFormat);  MMRESULT ret = waveOutOpen(NULL, WAVE_MAPPER, &m_waveFormat,   NULL, NULL, WAVE_FORMAT_QUERY);  if (MMSYSERR_NOERROR != ret) {   return -1;  }  // 第二步: 获取WAVEOUT句柄  ret = waveOutOpen(&m_hWaveOut, WAVE_MAPPER, &m_waveFormat,   (DWORD_PTR)waveOutProc, (DWORD_PTR)this, CALLBACK_FUNCTION);  if (MMSYSERR_NOERROR != ret) {   return -1;  }  return 0; } void CALLBACK CPcmSpeaker::waveOutProc(HWAVEOUT hwo, UINT uMsg, DWORD dwInstance, DWORD dwParam1, DWORD dwParam2) {  CPcmSpeaker *render = (CPcmSpeaker *)dwInstance;  //WAVEHDR *header = (WAVEHDR *)dwParam1;  int i = 0;  switch (uMsg)  {  case WOM_DONE:   EnterCriticalSection(&render->m_BufferOpCriticalSection);   SetEvent(render->m_hBufferEvent);   LeaveCriticalSection(&render->m_BufferOpCriticalSection);   break;  case WOM_CLOSE:   i = 1;   break;  case WOM_OPEN:   i = 2;   break;  } } int CPcmSpeaker::clearPcmData() {  if (m_hWaveOut != NULL)  {   EnterCriticalSection(&m_BufferOpCriticalSection);   for (int i = 0; i < m_maxBufferCnt; i++)   {    if (m_headAndDatas[i].header.dwFlags & WHDR_PREPARED) //有数据被Prepered     waveOutUnprepareHeader(m_hWaveOut, &m_headAndDatas[i].header, sizeof(WAVEHDR));   }   waveOutReset(m_hWaveOut);   LeaveCriticalSection(&m_BufferOpCriticalSection);  }  return 0; } int CPcmSpeaker::writeToWave(const void *data, int len) {  MMRESULT mmres;  int i;  EnterCriticalSection(&m_BufferOpCriticalSection);  for (i = 0; i < m_maxBufferCnt; i++)   if (m_headAndDatas[i].header.dwFlags & WHDR_DONE)   {    //查看是否需要释放之前已经Prepared资源    if (m_headAndDatas[i].header.dwFlags & WHDR_PREPARED) //有数据被Prepered     waveOutUnprepareHeader(m_hWaveOut, &m_headAndDatas[i].header, sizeof(WAVEHDR));    //写入新的数据到音频缓冲区       memcpy(m_headAndDatas[i].data, data, len);    m_headAndDatas[i].header.lpData = m_headAndDatas[i].data;    m_headAndDatas[i].header.dwBufferLength = len;    m_headAndDatas[i].header.dwFlags = 0;    mmres = waveOutPrepareHeader(m_hWaveOut, &m_headAndDatas[i].header, sizeof(WAVEHDR));    if (MMSYSERR_NOERROR == mmres)     mmres = waveOutWrite(m_hWaveOut, &m_headAndDatas[i].header, sizeof(WAVEHDR));    break;   }  LeaveCriticalSection(&m_BufferOpCriticalSection);  if (i == m_maxBufferCnt)   return -2;  return (mmres == MMSYSERR_NOERROR) ? 0 : -1; } //添加PCM音频数据，等待播放 int CPcmSpeaker::pcmtoWave(const void *data, int len, int timeout) {  int res;  if (len > m_maxBufferSize)   return -1;  res = writeToWave(data, len);  //缓冲区已满，需要等待  if (res == -2)  {       if (WAIT_OBJECT_0 == WaitForSingleObject(m_hBufferEvent, timeout))    res = writeToWave(data, len);  }  return res; } int CPcmSpeaker::toSpeaker(const void *data, int len, int timeout) {  int res;  int n, l, ptr;  //对大数据做分段处理  n = len / m_maxBufferSize;  l = len % m_maxBufferSize;  ptr = 0;  for (int i = 0; i < n; i++)  {   res = pcmtoWave(((char *)data) + ptr, m_maxBufferSize, timeout);   ptr += m_maxBufferSize;   if (res != 0)    return -1;  }  return pcmtoWave(((char *)data) + ptr, l, timeout); }  

用法

用法非常简单，如下：

1. 定义实例： CPcmSpeaker ps;
2. 初始化参数： ps.init(1, 16000, 16); 三个参数分别为：通道数，采样速率，单次采样数据位
3. 填PCM数据到喇叭： ps.toSpeaker(data, data_len); 两个参数分别为PCM数据指针和数据长度。

而外说明

微软的这个Waveform相关的函数，感觉比较原始，用的时候，需要如下注意事项：

• 如MSDN所言， waveOutProc 中不能调用任何 Waveform 相关函数，原文如下：

• CPcmSpeaker 的构造函数定义为： CPcmSpeaker(int bufferSize = DEF_MAX_BUFFER_SIZE, int bufferCnt = DEF_MAX_BUFFER_COUNT) ，其有两个有默认值的参数，分别为每次写入系统音频缓冲区的数据的最大大小，以及 CPcmSpeaker 自己的缓冲区个数， bufferCnt 不要太小(最好大于2，根据具体的情况设置大小，建议值为32)，否则会出现卡顿现象。

• toSpeaker 函数，带有第三个参数（默认为 INFINITE ） timeout ,表示函数调用超时时间。换句话说，该函数在某种程度上是阻塞式的，即，如果写入的太快，使得系统来不及播放数据，导致CPcmSpeaker类内部的缓冲区已经满了，那么 toSpeaker 函数将会等待有新的缓冲区数据被播放后，腾出空间后，才返回，当然如果你不想死等，可以设置一个超时值，超时后，也会返回。