PCM Data format with fmod - difference on multiple paltforms

Question

I am writing a Unity3D plugin that reads data from an MP3 file, feeds the PCM data to Unity so that it can play it inside the engine. On iOS, I use the AVAssetReaderAudioMixOutput class to decode and read the data, and on Android/Windows, I use FMOD. I have set up a program on both Windows and iOS which uses FMOD to play back the music, just like Unity3D does. I am having trouble getting the same results on both iOS and windows, and I cant seem to find the difference in audio output settings/format that would cause the difference.

So first, these are the settings that I set for my output audio stream, which are the same settings as Unity3D uses:

FMOD_CREATESOUNDEXINFO   exinfo2;
memset(&exinfo2, 0, sizeof(FMOD_CREATESOUNDEXINFO));
exinfo2.cbsize            = sizeof(FMOD_CREATESOUNDEXINFO);             
exinfo2.decodebuffersize  = 44100;                                  
exinfo2.length            = 44100 * 1 * sizeof(float) * 100;            
exinfo2.numchannels       = 1;                                          
exinfo2.defaultfrequency  = 44100;                                  
exinfo2.format            = FMOD_SOUND_FORMAT_PCMFLOAT;         
exinfo2.pcmreadcallback   = pcmreadcallback;                          
result = system_->createStream("./1.mp3", FMOD_LOOP_NORMAL | FMOD_SOFTWARE | FMOD_OPENUSER | FMOD_CREATESTREAM, &exinfo2, &sound2_);
ERRCHECK(result);
result = system_->playSound(FMOD_CHANNEL_FREE,sound2_,false,0);

Basically, 1 channel, 32-bit floating point PCM data. This is set on both the iOS and windows playback programs. Now, on iOS, I set the AVAssetReaderAudioMixOutput audio settings like this:

NSDictionary *audioSetting = [NSDictionary dictionaryWithObjectsAndKeys:
                                  [NSNumber numberWithFloat:44100.0],AVSampleRateKey,
                                  [NSNumber numberWithInt:1],AVNumberOfChannelsKey,    //how many channels has original?
                                  [NSNumber numberWithInt:32],AVLinearPCMBitDepthKey, //was 16
                                  [NSNumber numberWithInt:kAudioFormatLinearPCM], AVFormatIDKey,
                                  [NSNumber numberWithBool:YES], AVLinearPCMIsFloatKey,  //was NO
                                  [NSNumber numberWithBool:0], AVLinearPCMIsBigEndianKey,
                                  [NSNumber numberWithBool:NO], AVLinearPCMIsNonInterleaved,
                                  [NSData data], AVChannelLayoutKey, nil];

I set the PCMIsFloatKey to 1 so that the PCM data is a floating point, I set the bit-depth to 32, 1 channel, so that everything matches the FMOD output settings.

I read the data and write it into a circular buffer:

        float* convertedBuffer = (float * ) audioBufferList.mBuffers[0].mData;
        //We don't need the audioconverter on iOS
        //Fill up the circular buffer
        for(int i = 0; i < numSamples; i++)
        {
            circularAudioBuffer_[bufferWritePosition_] = convertedBuffer[i];
            bufferWritePosition_++;
            if(bufferWritePosition_ >= circularBufferSize_)
                bufferWritePosition_ = 0;

        }

Then read the data from the buffer and write it into the audio stream in the pcmreadcallback:

float *writeBuffer = (float *)data;
for(int i = 0; i < dataLength; i++)
{
    sampleBuffer[i] = circularAudioBuffer_[bufferReadPosition_];
    bufferReadPosition_++;
    if(bufferReadPosition_ >= circularBufferSize_)
        bufferReadPosition_ = 0;
}

With this, the audio plays perfectly, and the range of values inside the circular buffer is 0.0-1.0f

Now, on windows, I initialize the sound from which I read the data like this:

exinfo.cbsize            = sizeof(FMOD_CREATESOUNDEXINFO);             
exinfo.decodebuffersize  = 44100;
exinfo.numchannels       = 1;
exinfo.defaultfrequency  = 44100;            
exinfo.format            = FMOD_SOUND_FORMAT_PCMFLOAT;  

Setting the same parameters: 1channel, 32bit floating point. I read the data and write the data in the buffer:

    FMOD_RESULT result = sound_->readData(rawBuffer, N4, &bytesRead);

    float* floatBuffer = (float*) rawBuffer;
    for(int j = 0; j < N; j++)
    {
        circularAudioBuffer_[bufferWritePosition_++] = floatBuffer[j];
        if(bufferWritePosition_ >= circularBufferSize_)
            bufferWritePosition_ = 0;
    }

Now, when I read the data, I get very high or very low floating-point values (about 1e34 or -1e33). In the test program, I can't hear anything in the output.

I can switch the input and output sound format to PCM32 and it plays fine in the test program, but can`t be read properly in Unity3D (it screeches a lot, but I can make out the song).

Can anyone help me figure this out and make it works properly using the PCMFLOAT format? thanks!

TL;DR: I can't read data from FMOD sound with PCMFLOAT format!

Answer 1

From Fmod Support: Fmod Forums The specified output format doesn't matter. The FMod codec will always return PCM16 and the iOS codec returns PCMFloats. So, I need to convert them:

(float)pcm16in[j] / 32768.0f;

In addition to this, I was (accidently) initializing the output stream with an MP3 file, which made it so that I couldn't change the output format.