Sending wav over TCP, serializing problem?

[elementalsc]

Hi!

I'm working on a voice chat built with boost::asio and when I'm on the receiving side, I hear a serie of quick "click!" instead of my original wav. I think the issue might be with how I'm marshalling/unmarshalling the data.

Could I have your opinion on this?

Here's how I split it when sending the samples:

  // Chopping SoundBuffer into buffer-length char[] filled into mByteBufferQueue
  Int16 * mSoundBufferRaw = const_cast<sf::Int16 *>(mSoundBuffer.getSamples());

  while(wBufferIndex < mSoundBuffer.getSampleCount())
  {
 
    // ... some calculation for mBlocksToConvert  

    // create char[] buffer
    for(unsigned int i = 0; i < mBlocksToConvert; ++i)
    {
      mByteBuffer[2*i]   = 0xFF & static_cast<char>(mSoundBufferRaw[wBufferIndex] >> 8);
      mByteBuffer[2*i+1] = 0xFF & static_cast<char>(mSoundBufferRaw[wBufferIndex]);
      ++wBufferIndex;
    }

    // fill queue
    mByteBufferQueue.push_back(mByteBuffer);
  }
 

This is how it's received and played:
part 1...

      std::vector<sf::Int16> wSamples;

      for (unsigned int i = 0; i < iBytesReceived; i += 2) {
        wSamples.push_back(static_cast<sf::Int16>( (mData[i] << 8) | (0x00FF & mData[i+1]) ));
      }

      mServerPtr->pushSamples(std::move(wSamples));
 

part 2...

    if((mAudioStream.getStatus() != AudioStream::Playing) && !mSamplesQueue.empty())
   {
      if(mSoundBuffer.loadFromSamples(
        reinterpret_cast<sf::Int16*>(mSamplesQueue.front().data()),
        mSamplesQueue.front().size(), 2, 48000)
      )
      {
        mAudioStream.load(mSoundBuffer);
        mAudioStream.play();
        mSamplesQueue.pop_front();
      }
    }
 

[eXpl0it3r]

Get out Wireshark or log on both sides to a file and you'll see the difference. What's the point of complicating things instead of simply packing the bytes?

[elementalsc]

What's the point of complicating things instead of simply packing the bytes?

What do you mean by "simply packing the bytes"? I though that splitting int16 into two bytes and streaming them was simple enough... but I guess I'm missing something. How would you do it?

[Laurent]

What's the point of complicating things instead of simply packing the bytes?

Byte order?

Could I have your opinion on this?

I don't think it's the data, but rather how you handle it in mAudioStream. Would be interesting to see the implementation.

[elementalsc]

I took the Custom stream demo from the tutorial, it seems to work fine when loading and playing a buffer in the same program.

Maybe it's not working in the receiving end... other that the serializing thing, I was starting to ask myself if i was correctly "appending" SoundBuffer. I feel like I might simply be playing the same sample over and over.

    if((mAudioStream.getStatus() != AudioStream::Playing) && !mSamplesQueue.empty()) {
      if(mSoundBuffer.loadFromSamples(
        mSamplesQueue.front(),
        BUFFER_SIZE/2, 1, 44100) )
      {
        mAudioStream.load(mSoundBuffer);
        mAudioStream.play();
        mSamplesQueue.pop_front();
      }
    }
 

[Laurent]

I took the Custom stream demo from the tutorial

It's not meant for real streaming, it's just a dummy example that shows what to implement in a derived sf::SoundStream. The "voip" example from the SFML SDK is a much better implementation of what you want to do.