使用IMFSinkWriter生成WMV无音频且文件过大问题求助

使用IMFSinkWriter创建包含音视频的WMV文件时，视频功能完全正常，音频写入过程无报错，但播放生成的WMV文件时没有声音。此外，用FFmpeg存储相同音视频内容生成的WMV文件约2MB，而代码生成的文件约77MB。音频存储格式为PCMWAVEFORMAT后跟随16位短整型采样数据，怀疑遗漏了IMFSinkWriter及其组件所需的参数，寻求解决建议。

相关代码片段 #

HRESULT CreateAudioSinkWriter(IMFSinkWriter *pSinkWriter,DWORD *pStreamIndexAudio,HANDLE hAudio)
{
        PCMWAVEFORMAT *pwave = NULL;
        IMFMediaType *pAudioType = NULL;
        DWORD streamIndexAudio = 0;
        HRESULT hr;
        long nChannels = 2;
        long nSamplesPerSec = 44100;
        long wBitsPerSample = 16;
        long nBlockAlign = 4;
        long nAvgBytesPerSec = 176400;

        if(hAudio != NULL) {
                if((pwave = (PCMWAVEFORMAT *)GlobalLock(hAudio)) != NULL) {
                        nChannels = pwave->wf.nChannels;
                        nSamplesPerSec = pwave->wf.nSamplesPerSec;
                        wBitsPerSample = pwave->wBitsPerSample;

                        nBlockAlign = nChannels * (wBitsPerSample / 8);
                        nAvgBytesPerSec = nBlockAlign * nSamplesPerSec;

                        GlobalUnlock(hAudio);
                }
        }

        hr = MFCreateMediaType(&pAudioType);
        hr = pAudioType->SetGUID(MF_MT_MAJOR_TYPE,MFMediaType_Audio);
        hr = pAudioType->SetGUID(MF_MT_SUBTYPE,MFAudioFormat_WMAudioV8);
        hr = pAudioType->SetUINT32(MF_MT_AUDIO_BITS_PER_SAMPLE,wBitsPerSample);
        hr = pAudioType->SetUINT32(MF_MT_AUDIO_SAMPLES_PER_SECOND,nSamplesPerSec);
        hr = pAudioType->SetUINT32(MF_MT_AUDIO_NUM_CHANNELS,nChannels);
        hr = pAudioType->SetUINT32(MF_MT_AUDIO_BLOCK_ALIGNMENT,nBlockAlign);
        hr = pAudioType->SetUINT32(MF_MT_AUDIO_AVG_BYTES_PER_SECOND,nAvgBytesPerSec);
        hr = pAudioType->SetUINT32(MF_MT_AUDIO_PREFER_WAVEFORMATEX,TRUE);
        hr = pSinkWriter->AddStream(pAudioType,&streamIndexAudio);

        *pStreamIndexAudio = streamIndexAudio;

        SafeRelease(pAudioType);

        return(hr);
}

HRESULT CreateSinkWriter(const WCHAR *outputFile,IMFSinkWriter **ppWriter,DWORD *pStreamIndex,DWORD *pStreamIndexAudio,long width,long height,long rate,HANDLE hAudio) 
{
        *ppWriter = NULL;
        *pStreamIndex = NULL;
        *pStreamIndexAudio = NULL;

        IMFSinkWriter *pSinkWriter=NULL;
        IMFAttributes *pAttributes=NULL;
        HRESULT hr;
 
        hr = MFCreateAttributes(&pAttributes,2);
    hr = pAttributes->SetUINT32(MF_READWRITE_ENABLE_HARDWARE_TRANSFORMS,TRUE);
    hr = pAttributes->SetUINT32(MF_SINK_WRITER_DISABLE_THROTTLING,TRUE);
        hr = MFCreateSinkWriterFromURL(outputFile,NULL,pAttributes,&pSinkWriter);
        hr = CreateVideoSinkWriter(pSinkWriter,pStreamIndex,width,height,rate);
        hr = CreateAudioSinkWriter(pSinkWriter,pStreamIndexAudio,hAudio);
        hr = pSinkWriter->BeginWriting();       

        *ppWriter = pSinkWriter;
        (*ppWriter)->AddRef();

        SafeRelease(pAttributes);
        SafeRelease(pSinkWriter);
      
        return(hr);
}


HRESULT WriteAudioData(IMFSinkWriter *pWriter,DWORD streamIndex,HANDLE hAudio)
{
        HRESULT hr = S_OK;
        IMFSample *pSample=NULL;
        IMFMediaBuffer *pBuffer=NULL;
        DWORD sampleSize = GlobalSize(hAudio) - sizeof(PCMWAVEFORMAT);
        PCMWAVEFORMAT *pwave;
        LONGLONG duration;
        BYTE *pData;
        char *ps;
        DWORD chunkSize = 8192;
        LONGLONG sampleTime = 0;

        if((ps = (char *)GlobalLock(hAudio)) == NULL) return(E_OUTOFMEMORY);

        pwave = (PCMWAVEFORMAT *)ps;
        ps += sizeof(PCMWAVEFORMAT);

        while(sampleSize > 0) {
                DWORD currentChunkSize = min(chunkSize,sampleSize);

                duration = (LONGLONG)((double)currentChunkSize / (pwave->wf.nSamplesPerSec * pwave->wf.nBlockAlign) * 10000000);
                
                hr = MFCreateMemoryBuffer(currentChunkSize,&pBuffer);
                hr = pBuffer->Lock(&pData,NULL,NULL);
                
                memcpy(pData,ps,currentChunkSize);

                hr = pBuffer->Unlock();
                hr = pBuffer->SetCurrentLength(currentChunkSize);
                hr = MFCreateSample(&pSample);
                hr = pSample->AddBuffer(pBuffer);
        hr = pSample->SetSampleTime(sampleTime);
        hr = pSample->SetSampleDuration(duration);
        hr = pWriter->WriteSample(streamIndex,pSample);
                
                ps += currentChunkSize;
                sampleSize -= currentChunkSize;
                sampleTime += duration;

                SafeRelease(pBuffer);
                SafeRelease(pSample);
        }

        SafeRelease(pBuffer);
        SafeRelease(pSample);
        GlobalUnlock(hAudio);
        return(hr);
}

解决建议 #

无声音问题修复 #

• 音频媒体类型不匹配：当前代码设置音频子类型为MFAudioFormat_WMAudioV8（WMA编码格式），但写入的是原始PCM数据，格式不兼容导致播放器无法解析。应将子类型改为PCM对应的MFAudioFormat_PCM：

  hr = pAudioType->SetGUID(MF_MT_SUBTYPE, MFAudioFormat_PCM);
  

  若需编码为WMA，需额外配置编码参数并确保编码器可用，优先先匹配输入数据格式验证声音正常。
• 时间戳计算精度错误：原duration计算使用浮点运算易丢失精度，导致播放器无法正确同步音频。修正为整数运算：

  DWORD sampleCount = currentChunkSize / pwave->wf.nBlockAlign;
  duration = (LONGLONG)sampleCount * 10000000 / pwave->wf.nSamplesPerSec;
  

• 补充PCM格式关键参数：添加MF_MT_ALL_SAMPLES_INDEPENDENT参数明确为独立采样：

  hr = pAudioType->SetUINT32(MF_MT_ALL_SAMPLES_INDEPENDENT, TRUE);
  

文件体积过大问题修复 #

• 启用音频压缩编码：当前写入未压缩的原始PCM数据是体积过大的核心原因。需配置输入类型为PCM、输出类型为WMA编码，并设置编码比特率：

  // 创建输入媒体类型（匹配原始PCM数据）
  IMFMediaType *pInputType = NULL;
  hr = MFCreateMediaType(&pInputType);
  hr = pInputType->SetGUID(MF_MT_MAJOR_TYPE, MFMediaType_Audio);
  hr = pInputType->SetGUID(MF_MT_SUBTYPE, MFAudioFormat_PCM);
  hr = pInputType->SetUINT32(MF_MT_AUDIO_BITS_PER_SAMPLE, wBitsPerSample);
  hr = pInputType->SetUINT32(MF_MT_AUDIO_SAMPLES_PER_SECOND, nSamplesPerSec);
  hr = pInputType->SetUINT32(MF_MT_AUDIO_NUM_CHANNELS, nChannels);
  hr = pInputType->SetUINT32(MF_MT_AUDIO_BLOCK_ALIGNMENT, nBlockAlign);
  hr = pInputType->SetUINT32(MF_MT_AUDIO_AVG_BYTES_PER_SECOND, nAvgBytesPerSec);
  hr = pInputType->SetUINT32(MF_MT_ALL_SAMPLES_INDEPENDENT, TRUE);
  
  // 创建输出媒体类型（WMA编码）
  IMFMediaType *pOutputType = NULL;
  hr = MFCreateMediaType(&pOutputType);
  hr = pOutputType->SetGUID(MF_MT_MAJOR_TYPE, MFMediaType_Audio);
  hr = pOutputType->SetGUID(MF_MT_SUBTYPE, MFAudioFormat_WMAudioV8);
  hr = pOutputType->SetUINT32(MF_MT_AUDIO_SAMPLES_PER_SECOND, nSamplesPerSec);
  hr = pOutputType->SetUINT32(MF_MT_AUDIO_NUM_CHANNELS, nChannels);
  // 设置128kbps比特率，转换为字节每秒：128000 / 8 = 16000
  hr = pOutputType->SetUINT32(MF_MT_AUDIO_AVG_BYTES_PER_SECOND, 16000);
  
  // 添加流并设置输入输出类型
  hr = pSinkWriter->AddStream(pInputType, &streamIndexAudio);
  hr = pSinkWriter->SetInputMediaType(streamIndexAudio, pInputType, NULL);
  hr = pSinkWriter->SetOutputMediaType(streamIndexAudio, pOutputType, NULL);
  

  此配置会让SinkWriter自动将PCM编码为WMA，大幅降低文件体积。
• 检查视频压缩配置：若视频未使用WMV压缩格式（如MFVideoFormat_WMV3），也会导致体积过大，需同步确认视频媒体类型设置。

其他关键注意事项 #

• 严格检查HRESULT返回值：即使无报错，调试时需确保每一步操作的返回值均为S_OK，排查隐性错误。
• 写入完成后调用Finalize：所有数据写入完毕后，调用pSinkWriter->Finalize()确保数据完全写入文件：

  hr = pSinkWriter->Finalize();
  

内容的提问来源于stack exchange，提问作者Jack Waterhouse