Search code examples
objective-csignal-processingcore-audioaccelerate-framework

De-interleave and interleave buffer with vDSP_ctoz() and vDSP_ztoz()?


How do I de-interleave the float *newAudio into float *channel1 and float* channel2 and interleave it back into newAudio?

Novocaine *audioManager = [Novocaine audioManager];

__block float *channel1;
__block float *channel2;
[audioManager setInputBlock:^(float *newAudio, UInt32 numSamples, UInt32 numChannels) {
  // Audio comes in interleaved, so, 
  // if numChannels = 2, newAudio[0] is channel 1, newAudio[1] is channel 2, newAudio[2] is channel 1, etc. 

      // Deinterleave with vDSP_ctoz()/vDSP_ztoz(); and fill channel1 and channel2
      // ... processing on channel1 & channel2
      // Interleave channel1 and channel2 with vDSP_ctoz()/vDSP_ztoz(); to newAudio
}];

What would these two lines of code look like? I don't understand the syntax of ctoz/ztoz.


Solution

  • What I do in Novocaine's accessory classes, like the Ringbuffer, for de-interleaving:

    float zero = 0.0;  
    vDSP_vsadd(data, numChannels, &zero, leftSampleData, 1, numFrames);   
    vDSP_vsadd(data+1, numChannels, &zero, rightSampleData, 1, numFrames);  
    

    for interleaving:

    float zero = 0.0;  
    vDSP_vsadd(leftSampleData, 1, &zero, data, numChannels, numFrames);   
    vDSP_vsadd(rightSampleData, 1, &zero, data+1, numChannels, numFrames);  
    

    The more general way to do things is to have an array of arrays, like

    int maxNumChannels = 2; 
    int maxNumFrames = 1024;
    float **arrays = (float **)calloc(maxNumChannels, sizeof(float *));  
    for (int i=0; i < maxNumChannels; ++i) {  
        arrays[i] = (float *)calloc(maxNumFrames, sizeof(float));
    }
    
    [[Novocaine audioManager] setInputBlock:^(float *data, UInt32 numFrames, UInt32 numChannels) {
        float zero = 0.0;
        for (int iChannel = 0; iChannel < numChannels; ++iChannel) {
            vDSP_vsadd(data, numChannels, &zero, arrays[iChannel], 1, numFrames);
        }
    }];
    

    which is what I use internally a lot in the RingBuffer accessory classes for Novocaine. I timed the speed of vDSP_vsadd versus memcpy, and (very, very surprisingly), there's no speed difference.

    Of course, you can always just use a ring buffer, and save yourself the hassle

    #import "RingBuffer.h"
    
    int maxNumFrames = 4096
    int maxNumChannels = 2
    RingBuffer *ringBuffer = new RingBuffer(maxNumFrames, maxNumChannels)
    
    [[Novocaine audioManager] setInputBlock:^(float *data, UInt32 numFrames, UInt32 numChannels) {
        ringBuffer->AddNewInterleavedFloatData(data, numFrames, numChannels);
    }];
    
    [[Novocaine audioManager] setOuputBlock:^(float *data, UInt32 numFrames, UInt32 numChannels) {
        ringBuffer->FetchInterleavedData(data, numFrames, numChannels);
    }];
    

    Hope that helps.