Question

I have a code that plays mono audio event (short beeps at various frequencies).
I create an AudioOutputUnit, stop it, and whenever I need to play the audio. I start it. When I’ve played it for the required time, I stop it.

Sounds simple enough.

However, AudioOutputUnitStart will take usually 180ms to return on my iPhone 4S (with iOS 5.1), this is way too much.

Here is the creation/initialisation of the AudioOutputUnit

void createAOU()
{
    m_init = false;
    // find the default playback output unit
    AudioComponentDescription defaultOutputDescription;
    defaultOutputDescription.componentType = kAudioUnitType_Output;
    defaultOutputDescription.componentSubType = kAudioUnitSubType_RemoteIO;
    defaultOutputDescription.componentManufacturer = kAudioUnitManufacturer_Apple;
    defaultOutputDescription.componentFlags = 0;
    defaultOutputDescription.componentFlagsMask = 0;

    // Get the default playback output unit
    AudioComponent defaultOutput = AudioComponentFindNext(NULL, &defaultOutputDescription);
    if (defaultOutput == NULL)
        return;

    OSErr err = AudioComponentInstanceNew(defaultOutput, &m_toneUnit);
    if (err != noErr)
        return;

    // single channel, floating point, linear PCM
    AudioStreamBasicDescription streamFormat;
    streamFormat.mSampleRate = m_framerate;
    streamFormat.mFormatID = kAudioFormatLinearPCM;
    streamFormat.mFormatFlags =
    kLinearPCMFormatFlagIsFloat |
    kAudioFormatFlagsNativeEndian | kLinearPCMFormatFlagIsPacked;
    streamFormat.mBytesPerPacket = sizeof(float);
    streamFormat.mFramesPerPacket = 1;
    streamFormat.mBytesPerFrame = sizeof(float);
    streamFormat.mChannelsPerFrame = 1;
    streamFormat.mBitsPerChannel = sizeof(float) * 8;
    err = AudioUnitSetProperty (m_toneUnit,
                                kAudioUnitProperty_StreamFormat,
                                kAudioUnitScope_Input,
                                0,
                                &streamFormat,
                                sizeof(AudioStreamBasicDescription));
    if (err != noErr)
        return;

    // Attach callback to default output
    AURenderCallbackStruct input;
    input.inputProc = RenderTone;
    input.inputProcRefCon = this;
    err = AudioUnitSetProperty(m_toneUnit, 
                               kAudioUnitProperty_SetRenderCallback, 
                               kAudioUnitScope_Input,
                               0, 
                               &input, 
                               sizeof(input));
    if (err != noErr)
        return;

    float aBufferLength = 0.001; // In seconds
    AudioSessionSetProperty(kAudioSessionProperty_PreferredHardwareIOBufferDuration, 
                            sizeof(aBufferLength), &aBufferLength);

    err = AudioUnitInitialize(m_toneUnit);
    if (err != noErr)
        return;
    reset();
    m_init = true;
}

To start it, I would simply call:

OSErr err = AudioOutputUnitStart(m_toneUnit);

When done:

OSErr err = AudioOutputUnitStop(m_toneUnit);

I’ve tried various things:
making the kAudioSessionProperty_PreferredHardwareIOBufferDuration property to a very small value (like 5ms)
Reduce the framerate (tried 48kHz, 44.1kHz, 8kHz)

No matter what I do…
First time the audiounit is started, AudioOutputUnitStart only returns after 160-180ms. If I stop and start it right away, then it only takes about 5m which is much more acceptable.

Latency in my application is rather important, and 180ms is definitely not acceptable.

Any suggestions?
I’ve seen a few people asking similar questions, but they usually never got an answer.

I’m hoping this time things will be different 🙂

Answer 1

The way to deal with Audio Unit start-up latency is to start the Audio Unit ahead of the time needed, and not stop it. Instead, just configure the audio session for short buffers, and then fill the Audio Unit render callback buffers with silence until it’s time to fill them with your desired sound. Then after your sound, go back to filling the buffers with silence instead of stopping.