How to manually convert 8.24-bit deinterleaved lpcm to 16-bit lpcm?

I have a chunk of data (void*) which is 2 ch, 44100 Hz, 'lpcm' 8.24-bit li开发者_StackOverflowttle-endian signed integer, deinterleaved. I need to record that chunk to a file as 2 ch, 44100 Hz, 'lpcm' 16-bit little-endian signed integer.

How do I convert data? I can imagine I need to do something like this:

uint dataByteSize = sizeof(UInt32) * samplesCount;
UInt32* source = ...;
UInt32* dest = (UInt32*)malloc(dataByteSize);
for (int i = 0; i < samplesCount; ++i) {
    UInt32 sourceSample = source[i];
    UInt32 destSample = sourceSample>>24;
    dest[i] = destSample;
}

But how do I convert deinterleaved to interleaved?

Ok, I've spent some time investigating the issue and realized that question contains too few information to be answered =) So heres the deal:

First, about non-interleaved: I initially thought that it would look like this: l1 l2 l3 l4...ln r1 r2 r3 r4...rn But it turned out that in my data right channel was just missing. It turned out that it wasn't a non-interleaved data, it was just a plain mono data. And yes, it should always be multiple buffers in case data is actually non-interleaved. If it's interleaved, it should be l1 r1 l2 r2 l3 r3 l4 r4...

Second, about actual transformation: it all depends on the range of samples. In my case (and in any case where core audio is involved if I'm correct) fixed-point 8.24 values should range between (-1, 1), while 16-bit signed values should range between (-32768, 32767). So 8.24 value will always have its first 8 bits set either to 0 (in case it is positive) or to 1 (in case it is negative). This first 8 bits should be removed (preserving sign ofc). Also you can remove as many trailing bits as you want - it'll just reduce quility of the sound, but it wont ruin the sound. In case of converting to 16 bits signed format, bits 8-22 (15 bits that is) will actually contain the data we need to use for SInt16. Bit 7 can be used as the sign bit. So to convert 8.24 to SInt16 you just need to shift 9 bits right (9 because you need to preserve the sign) and cast to SInt16

11111111 10110110 11101110 10000011 - > 11111111 11111111 (11011011 01110111)
00000000 01101111 00000000 11000001 - > 00000000 00000000 (00110111 10000000)

That's it. Nothing more then iterating through array and shifting bits right. Hope that's gonna save someone couple of hours.

I have read the following clip in audiograph https://github.com/tkzic/audiograph

/* convert sample vector from fixed point 8.24 to SInt16 */
void fixedPointToSInt16( SInt32 * source, SInt16 * target, int length ) {    
    int i;

    for(i = 0;i < length; i++ ) {
        target[i] =  (SInt16) (source[i] >> 9);        
    }    
}

Best description can be found at http://lists.apple.com/archives/coreaudio-api/2011/Feb/msg00083.html

So,

8.24 numbers are interpreted as ranging from -128.000000000000 to +127.999999940393

But!

The convention in iOS/CoreAudio is to treat -1.000000000000 to +0.999969482421875 as non-clipped values that do not exceed Full Scale for the 16-bit analog audio converters.

Ok?

I tested the popular method with shifting 9 bits and for some reason it does not work for me, as I further use the result to encode to ogg. The resulted ogg was verry noisy. What did work is this function, based on a method I found in audiograph https://github.com/tkzic/audiograph

void ConvertInputToInt16(AudioStreamBasicDescription inFormat, void *buf, void *outputBuf, size_t capacity) {
    AudioConverterRef converter;
    OSStatus err;

    size_t bytesPerSample = sizeof(SInt16);
    AudioStreamBasicDescription outFormat = {0};
    outFormat.mFormatID = kAudioFormatLinearPCM;
    outFormat.mFormatFlags = kAudioFormatFlagIsSignedInteger | kAudioFormatFlagIsPacked;
    outFormat.mBitsPerChannel = 8 * bytesPerSample;
    outFormat.mFramesPerPacket = 1;
    outFormat.mChannelsPerFrame = 1;
    outFormat.mBytesPerPacket = bytesPerSample * outFormat.mFramesPerPacket;
    outFormat.mBytesPerFrame = bytesPerSample * outFormat.mChannelsPerFrame;
    outFormat.mSampleRate = inFormat.mSampleRate;

    NSLog(@"description for in format: %@", descriptionForAudioFormat(inFormat));
    NSLog(@"description for out format: %@", descriptionForAudioFormat(outFormat));

    UInt32 inSize = capacity*sizeof(SInt32);
    UInt32 outSize = capacity*sizeof(SInt16);

    // this is the famed audio converter

    err = AudioConverterNew(&inFormat, &outFormat, &converter);
    if(noErr != err) {
        NSLog(@"error in audioConverterNew: %d", (int)err);
    }


    err = AudioConverterConvertBuffer(converter, inSize, buf, &outSize, outputBuf);
    if(noErr != err) {
        NSLog(@"error in audioConverterConvertBuffer: %d", err);
    }

}