Why might my AudioQueueOutputCallback not be called? - iphone

I'm using the Audio Queue Services API to play audio streamed from a server over a TCP socket connection on an iPhone. I can play the buffers that were filled from the socket connection, I just cannot seem to make my AudioQueue call my AudioQueueOutputCallback function, and I'm out of ideas.
High level design
Data is passed to the player from the socket connection, and written
immediately into circular buffers in memory.
As AudioQueueBuffers become available, data is copied from the circular buffers into the
available AudioQueueBuffer, which is immediately re-queued. (Or would be, if my callback happened)
What happens
The buffers are all filled and enqueued successfully, and I hear the audio stream clearly. For testing, I use a large number of buffers (15) and all of them play through seamlessly, but the AudioQueueOutputCallback is never called, so I never re-queue any of those buffers, despite the fact that everything seems to be working perfectly. If I don't wait for my callback, assuming it will never be called, and instead drive the enqueueing of buffers based on the data as it is written, I can play the audio stream indefinitely, reusing and re-enqueueing buffers as if they had been explicitly returned to me by the callback. It is that fact: that I can play the stream perfectly while reusing buffers as needed, that confuses me the most. Why isn't the callback being called?
Possibly Relevant Code
The format of the stream is 16 bit linear PCM, 8 kHz, Mono:
_streamDescription.mSampleRate = 8000.0f;
_streamDescription.mFormatID = kAudioFormatLinearPCM;
_streamDescription.mBytesPerPacket = 2;
_streamDescription.mFramesPerPacket = 1;
_streamDescription.mBytesPerFrame = sizeof(AudioSampleType);
_streamDescription.mChannelsPerFrame = 1;
_streamDescription.mBitsPerChannel = 8 * sizeof(AudioSampleType)
_streamDescription.mReserved = 0;
_streamDescription.mFormatFlags = (kLinearPCMFormatFlagIsBigEndian |
kLinearPCMFormatFlagIsPacked);
My prototype and implementation of the callback are as follows. Nothing fancy, and pretty much identical to every example I've seen so far:
// Prototype, declared above the class's #implementation
void AQBufferCallback(void* inUserData, AudioQueueRef inAudioQueue, AudioQueueBufferRef inAudioQueueBuffer);
// Definition at the bottom of the file.
void AQBufferCallback(void* inUserData, AudioQueueRef inAudioQueue, AudioQueueBufferRef inAudioQueueBuffer) {
printf("callback\n");
[(MyAudioPlayer *)inUserData audioQueue:inAudioQueue didAquireBufferForReuse:inAudioQueueBuffer];
}
I create the AudioQueue like this:
OSStatus status = 0;
status = AudioQueueNewOutput(&_streamDescription,
AQBufferCallback, // <-- Doesn't work...
self,
CFRunLoopGetCurrent(),
kCFRunLoopCommonModes,
0,
&_audioQueue);
if (status) {
// This is not called...
NSLog(#"Error creating new audio output queue: %#", [MyAudioPlayer stringForOSStatus:status]);
return;
}
And I enqueue buffers like this. At this point, it is known that the local buffer contains the correct amount of data for copying:
memcpy(aqBuffer->mAudioData, localBuffer, kAQBufferSize);
aqBuffer->mAudioDataByteSize = kAQBufferSize;
OSStatus status = AudioQueueEnqueueBuffer(_audioQueue, aqBuffer, 0, NULL);
if (status) {
// This is also not called.
NSLog(#"Error enqueueing buffer %#", [MyAudioPlayer stringForOSStatus:status]);
}
Please save me.

Is this executed on the main thread or a background thread? probably not good if CFRunLoopGetCurrent() returns a run loop of a thread that could disappear (thread pool etc) or is a run loop that don't care about kCFRunLoopCommonModes.
Try to change CFRunLoopGetCurrent() to CFRunLoopGetMain() or make sure AudioQueueNewOutput() and CFRunLoopGetCurrent() is executed on the main thread or a thread that you have control over and has a proper run loop.

Try changing self for (void*)self. Like this:
status = AudioQueueNewOutput(&_streamDescription,
AQBufferCallback,
(void*)self,
CFRunLoopGetCurrent(),
kCFRunLoopCommonModes,
0,
&_audioQueue);

Related

AudioKit tap skips over time intervals

I am building an app that uses microphone input to detect sounds and trigger events. I based my code on AKAmplitudeTap, but I when I ran it, I found that I was only obtaining sample data for intervals with missing sections.
The tap code looks like this (with the guts ripped out and simply keeping track of how many samples would have been processed):
open class MyTap {
// internal let bufferSize: UInt32 = 1_024 // 8-9 kSamples/sec
internal let bufferSize: UInt32 = 4096 // 39.6 kSamples/sec
// internal let bufferSize: UInt32 = 16536 // 43.3 kSamples/sec
public init(_ input: AKNode?) {
input?.avAudioNode.installTap(onBus: 0, bufferSize: bufferSize, format: nil ) { buffer, _ in
sampleCount += self.bufferSize
}
}
I initialize the tap with:
func afterLoad() {
assert(!loaded)
AKSettings.audioInputEnabled = true
do {
try AKSettings.setSession(category: .playAndRecord, with: .allowBluetoothA2DP)
} catch {
print("Could not set session category.")
}
mic = AKMicrophone()
myTap = MyTap(mic) // seriously, can it be that easy?
loaded = true
}
The original tap code was capturing samples to a buffer, but I saw that big chunks of time were missing with a buffer size of 1024. I suspected that the processing time for the sample buffer might be excessive, so...
I simplified the code to simply keep track of how many samples were being passed to the tap. In another part of the code, I simply print out sampleCount/elapsedTime and, as noted in the comments after 'bufferSize' I get different amounts of samples per second.
The sample rate converges on 43.1 KSamples/sec with a 16K buffer, and only collects about 20% of the samples with a 1K buffer. I would prefer to use the small buffer size to obtain near real-time response to detected sounds. As I've been writing this, the 4K buffer version has been running and has stabilized at 39678 samples/sec.
Am I missing something? Can a tap with a small buffer size actually capture 44.1 Khz sample data?
Problem resolved... the tap requires this line of code
buffer.frameLength = self.bufferSize
... and suddenly all the samples appear. I obviously stripped out a bit too much code from the code I obviously didn't understand.

Using Novocaine in an audio app

I'm building an iPhone app that generates random guitar music by playing back individual recorded guitar notes in "caf" format. These notes vary in duration from 3 to 11 seconds, depending on the amount of sustain.
I originally used the AVAudioPlayer for playback, and in the simulator at 120 bpm, playing 16th notes it sung beautifully, but on my handset, as soon as I
upped the tempo a little over 60 bpm playing just 1/4 notes, it ran like a dog and wouldn't keep in time. My elation was very short lived.
To reduce latency, I tried to implement playback via Audio Units using the Apple MixerHost project as a template for an audio engine, but kept getting a bad access error after I bolted it on and connected everything up.
After many hours of it doing my head in, I gave up on that avenue of thought and I bolted on the Novocaine audio engine instead.
I have now run into a brick wall trying to connect it up to my model.
On the most basic level, my model is a Neck object containing an NSDictionary of Note objects.
Each Note object knows what string and fret of the guitar neck it's on and contains its own AVAudioPlayer.
I build a chromatic guitar neck containing either 122 notes (6 strings by 22 frets) or 144 notes (6 strings by 24 frets) depending on the neck size selected in the user preferences.
I use these Notes as my single point of truth so all scalar Notes generated by the music engine are pointers to this chromatic note bucket.
#interface Note : NSObject <NSCopying>
{
NSString *name;
AVAudioPlayer *soundFilePlayer;
int stringNumber;
int fretNumber;
}
I always start off playback with the root Note or Chord of the selected scale and then generate the note to play next so I am always playing one note behind the generated note. This way, the next Note to play is always queued up ready to go.
Playback control of these Notes is a achieved with the following code:
- (void)runMusicGenerator:(NSNumber *)counter
{
if (self.isRunning) {
Note *NoteToPlay;
// pulseRate is the time interval between beats
// staticNoteLength = 1/4 notes, 1/8th notes, 16th notes, etc.
float delay = self.pulseRate / [self grabStaticNoteLength];
// user setting to play single, double or triplet notes.
if (self.beatCounter == CONST_BEAT_COUNTER_INIT_VAL) {
NoteToPlay = [self.GuitarNeck generateNoteToPlayNext];
} else {
NoteToPlay = [self.GuitarNeck cloneNote:self.GuitarNeck.NoteToPlayNow];
}
self.GuitarNeck.NoteToPlayNow = NoteToPlay;
[self callOutNoteToPlay];
[self performSelector:#selector(runDrill:) withObject:NoteToPlay afterDelay:delay];
}
- (Note *)generateNoteToPlayNext
{
if ((self.musicPaused) || (self.musicStopped)) {
// grab the root note on the string to resume
self.NoteToPlayNow = [self grabRootNoteForString];
//reset the flags
self.musicPaused = NO;
self.musicStopped = NO;
} else {
// Set NoteRingingOut to NoteToPlayNow
self.NoteRingingOut = self.NoteToPlayNow;
// Set NoteToPlaNowy to NoteToPlayNext
self.NoteToPlayNow = self.NoteToPlayNext;
if (!self.NoteToPlayNow) {
self.NoteToPlayNow = [self grabRootNoteForString];
// now prep the note's audio player for playback
[self.NoteToPlayNow.soundFilePlayer prepareToPlay];
}
}
// Load NoteToPlayNext
self.NoteToPlayNext = [self generateRandomNote];
}
- (void)callOutNoteToPlay
{
self.GuitarNeck.NoteToPlayNow.soundFilePlayer.delegate = (id)self;
[self.GuitarNeck.NoteToPlayNow.soundFilePlayer setVolume:1.0];
[self.GuitarNeck.NoteToPlayNow.soundFilePlayer setCurrentTime:0];
[self.GuitarNeck.NoteToPlayNow.soundFilePlayer play];
}
Each Note's AVAudioPlayer is loaded as follows:
- (AVAudioPlayer *)buildStringNotePlayer:(NSString *)nameOfNote
{
NSString *soundFileName = #"S";
soundFileName = [soundFileName stringByAppendingString:[NSString stringWithFormat:#"%d", stringNumber]];
soundFileName = [soundFileName stringByAppendingString:#"F"];
if (fretNumber < 10) {
soundFileName = [soundFileName stringByAppendingString:#"0"];
}
soundFileName = [soundFileName stringByAppendingString:[NSString stringWithFormat:#"%d", fretNumber]];
NSString *soundPath = [[NSBundle mainBundle] pathForResource:soundFileName ofType:#"caf"];
NSURL *fileURL = [NSURL fileURLWithPath:soundPath];
AVAudioPlayer *audioPlayer = [[AVAudioPlayer alloc] initWithContentsOfURL:fileURL error:nil];
return notePlayer;
}
Here is where I come a cropper.
According to the Novocaine Github page ...
Playing Audio
Novocaine *audioManager = [Novocaine audioManager];
[audioManager setOutputBlock:^(float *audioToPlay, UInt32 numSamples, UInt32 numChannels) {
// All you have to do is put your audio into "audioToPlay".
}];
But in the downloaded project, you use the following code to load the audio ...
// AUDIO FILE READING OHHH YEAHHHH
// ========================================
NSURL *inputFileURL = [[NSBundle mainBundle] URLForResource:#"TLC" withExtension:#"mp3"];
fileReader = [[AudioFileReader alloc]
initWithAudioFileURL:inputFileURL
samplingRate:audioManager.samplingRate
numChannels:audioManager.numOutputChannels];
[fileReader play];
fileReader.currentTime = 30.0;
[audioManager setOutputBlock:^(float *data, UInt32 numFrames, UInt32 numChannels)
{
[fileReader retrieveFreshAudio:data numFrames:numFrames numChannels:numChannels];
NSLog(#"Time: %f", fileReader.currentTime);
}];
Here is where I really start to get confused because the first method uses a float and the second one uses a URL.
How do you pass a "caf" file to a float? I am not sure how to implement Novocaine - it is still fuzzy in my head.
My questions that I hope someone can help me with are as follows ...
Are Novocaine objects similar to AVAudioPlayer objects, just more versatile and tweaked to the max for minimum latency? i.e. self contained audio playing (/recording/generating) units?
Can I use Novocaine in my model as it is? i.e. 1 Novocaine object per chromatic note or should I have 1 novocain object that contains all the Chromatic Notes? Or do I just store the URL in the note instead and pass that to a Novocaine player?
How can I put my audio into "audioToPlay" when my audio is a "caf" file and "audioToPlay" take a float?
If I include and declare a Novocaine property in Note.m do I then have to rename the class to Note.mm in order to use the Novocaine object?
How do I play multiple Novocaine objects concurrently in order to reproduce chords and intervals?
Can I loop a Novocaine object's playback?
Can I set the playback length of a note? i.e. play a 10 sec note for only 1 sec?
Can I modify the above code to use Novocaine?
Is the method I am using for runMusicGenerator the correct one to use in order to maintain a tempo that is up to professional standards?
Novocaine makes your life easier by eliminating the need for you to setup the RemoteIO AudioUnit manually. This includes having to painfully fill a bunch of CoreAudio structs and providing a bunch of callbacks such as this audio process callback.
static OSStatus PerformThru(void *inRefCon, AudioUnitRenderActionFlags *ioActionFlags, const AudioTimeStamp *inTimeStamp, UInt32 inBusNumber, UInt32 inNumberFrames, AudioBufferList *ioData);
Instead Novocaine handles that in its implementation and then calls your block, which you set by doing this.
[audioManager setOutputBlock: ^(float *audioToPlay, UInt32 numSamples, UInt32 numChannels){} ];
Whatever you write to audioToPlay gets played.
Novocaine sets up the RemoteIO AudioUnit for you. This is a low-level CoreAudio API, different from the high-level AVFoundation, and very low-latency as expected. You are right in that Novocaine is self-contained. You can record, generate, and process audio in realtime.
Novocaine is a singleton, you cannot have multiple Novocaine instances. One way to do it is to store your guitar sound/sounds in a separate class or array, and then write a bunch of methods, using Novocaine to play them.
You have a bunch of options. You can use Novocaine's AudioFileReader to play your .caf file for you. You do this by allocating an AudioFileReader and then passing the URL of the .caf file you want to play, as per example code. You then stick [fileReader retrieveFreshAudio:data numFrames:numFrames numChannels:numChannels] in your block, as per example code. Each time your block is called, AudioFileReader grabs and buffers a chunk of audio from disk and puts it in audioToPlay which subsequently gets played. There are some disadvantages with this. For short sounds (such as your guitar sound I'm assuming) repeatedly calling retrieveFreshAudio is a performance hit. It is generally a better idea (for short sounds) to perform a synchronous, sequential read of the entire file into memory. Novocaine does not provide a way to do this (yet). You will have to use ExtAudioFileServices to do this. The Apple example project MixerHost details how to do this.
If you are using AudioFileReader yes. You only rename to .mm when you are #import ing from Obj-C++ headers or #include ing C++ headers.
As mentioned earlier, only 1 Novocaine instance is allowed. You can achieve polyphony by mixing multiple audio sources. This is simply just adding buffers together. If you have made multiple versions of the same guitar sound at different pitches, just read them all in to memory, and mix away. If you only want to have one guitar sound, then you have to, in realtime, change the playback rate of however many notes you are playing and then mixdown.
Novocaine is agnostic to what you are actually playing and does not care how long you are playing a sample for. In order to loop a sound, you have to maintain a count of how many samples have elapsed, check if you are at the end of your sound, and then set that count back to 0.
Yes. Assuming a 44.1k sample rate, 1 sec of audio = 44100 samples. You would then reset your count when it reaches 44100.
Yes. It looks something like this. Assuming you have 4 guitar sounds which are mono and longer than 1 second long, and you have read them into memory float *guitarC, *guitarE, *guitarG, *guitarB; (jazzy CMaj7 chord w00t), and want to mix them down for 1 second and loop that back in mono:
[audioManager setOutputBlock:^(float *data, UInt32 numFrames, UInt32 numChannels){
static int count = 0;
for(int i=0; i<numFrames; ++i){
//Mono mix each sample of each sound together. Since result can be 4x louder, divide the total amp by 4.
//You should be using `vDSP_vadd` from the accelerate framework for added performance.
data[count] = (guitarC[count] + guitarE[count] + guitarG[count] + guitarB[count]) * 0.25;
if(++count >= 44100) count = 0; //Plays the mix for 1 sec
}
}];
Not exactly. Using performSelector or any mechanism scheduled on a runloop or thread is not guaranteed to be precise. You might experience timing irregularities when the CPU load fluctuates, for example. Use the audio block if you want sample accurate timing.

OpenSL ES can not play audio on Android emulator

I decode amrnb to PCM, then put right pcm buffer to Enqueue buffer (I'm sure PCM data is right), but no sound is heard. And when feeding buffer, log outputs:
/AudioTrack(14857): obtainBuffer timed out (is the CPU pegged?)
My code is below, and my questions are:
Is there something wrong when I use the OpenSL ES?
Is it true that OpenSL ES only works on the real device?
Sample code:
void AudioTest()
{
StartAudioPlay();
while(1)
{
//decode AMR to PCM
/* Convert to little endian and write to wav */
//write buffer to buffer queue
AudioBufferWrite(littleendian, 320);
}
}
void bqPlayerCallback(SLAndroidSimpleBufferQueueItf bq, void *context)
{
//do nothing
}
void AudioBufferWrite(const void* buffer, int size)
{
(*gBQBufferQueue)->Enqueue(gBQBufferQueue, buffer, size );
}
// create buffer queue audio player
void SlesCreateBQPlayer(/*AudioCallBackSL funCallback, void *soundMix,*/ int rate, int nChannel, int bitsPerSample )
{
SLresult result;
// configure audio source
SLDataLocator_AndroidSimpleBufferQueue loc_bufq = {SL_DATALOCATOR_ANDROIDSIMPLEBUFFERQUEUE, 2};
SLDataFormat_PCM format_pcm = {SL_DATAFORMAT_PCM, 1, SL_SAMPLINGRATE_8,
SL_PCMSAMPLEFORMAT_FIXED_16, SL_PCMSAMPLEFORMAT_FIXED_16,
SL_SPEAKER_FRONT_CENTER, SL_BYTEORDER_LITTLEENDIAN};
SLDataSource audioSrc = {&loc_bufq, &format_pcm};
// configure audio sink
SLDataLocator_OutputMix loc_outmix = {SL_DATALOCATOR_OUTPUTMIX, gOutputMixObject};
SLDataSink audioSnk = {&loc_outmix, NULL};
// create audio player
const SLInterfaceID ids[3] = {SL_IID_BUFFERQUEUE, SL_IID_EFFECTSEND, SL_IID_VOLUME};
const SLboolean req[3] = {SL_BOOLEAN_TRUE, SL_BOOLEAN_TRUE, SL_BOOLEAN_TRUE};
result = (*gEngineEngine)->CreateAudioPlayer(gEngineEngine, &gBQObject, &audioSrc, &audioSnk,
3, ids, req);
// realize the player
result = (*gBQObject)->Realize(gBQObject, SL_BOOLEAN_FALSE);
// get the play interface
result = (*gBQObject)->GetInterface(gBQObject, SL_IID_PLAY, &gBQPlay);
// get the buffer queue interface
result = (*gBQObject)->GetInterface(gBQObject, SL_IID_BUFFERQUEUE,
&gBQBufferQueue);
// register callback on the buffer queue
result = (*gBQBufferQueue)->RegisterCallback(gBQBufferQueue, bqPlayerCallback, NULL/*soundMix*/);
// get the effect send interface
result = (*gBQObject)->GetInterface(gBQObject, SL_IID_EFFECTSEND,
&gBQEffectSend);
// set the player's state to playing
result = (*gBQPlay)->SetPlayState(gBQPlay, SL_PLAYSTATE_PLAYING );
}
I'm not entirely sure, but I think you're correct in that the emulator's OpenSL ES support doesn't actually work. I've never gotten it to work in practice, while it works on any device I've tried.
In my application I have to support Android 2.2 as well, so I have a fallback to use JNI to access the Java AudioTrack APIs. I added a special case to my app to always use the AudioTrack interface when the emulator is detected.

AudioQueue screws up output after modification

I am currently working on an audio DSP App development. The project requires direct access and modification of audio data. Right now I can successfully access and modify the raw audio data using AudioQueue but encounters error during playback. The output audio after any modification turns out be noise.
In short, the code is something like this:
(Modified from Speakhere sample code. The rest remains unchanged.)
void AQPlayer::AQBufferCallback(void * inUserData,
AudioQueueRef inAQ,
AudioQueueBufferRef inCompleteAQBuffer)
{
AQPlayer *THIS = (AQPlayer *)inUserData;
if (THIS->mIsDone) return;
UInt32 numBytes;
UInt32 nPackets = THIS->GetNumPacketsToRead();
OSStatus result = AudioFileReadPackets(THIS->GetAudioFileID(),
false,
&numBytes,
inCompleteAQBuffer->mPacketDescriptions,
THIS->GetCurrentPacket(),
&nPackets,
inCompleteAQBuffer->mAudioData);
if (result)
printf("AudioFileReadPackets failed: %d", (int)result);
if (nPackets > 0) {
inCompleteAQBuffer->mAudioDataByteSize = numBytes;
inCompleteAQBuffer->mPacketDescriptionCount = nPackets;
//My modification starts from here
//Modifying audio data
SInt16 *testBuffer = (SInt16*)inCompleteAQBuffer->mAudioData;
for (int i = 0; i < (inCompleteAQBuffer->mAudioDataByteSize)/sizeof(SInt16); i++)
{
//printf("before modification %d", (int)*testBuffer);
*testBuffer = (SInt16) *testBuffer/2; //Say some simple modification
//printf("after modification %d", (int)*testBuffer);
testBuffer++;
}
AudioQueueEnqueueBuffer(inAQ, inCompleteAQBuffer, 0, NULL);
}
During debugging, the data in buffer is displayed as expected, but the actual output is nothing but noise.
Here are some other strange behaviors of the code that makes both the whole team crazy:
If there is no change to the data (add/sub by 0, multiply by 1) or the whole buffer is assigned to a constant (say 0, then the audio will be muted), the playback behaves normally (Of course!) But if I perform anything more than it, it still turns out to be noise.
In the case I hardcode a single tone as test audio, the output noise spreads into another channel also.
So where is the bug in this code? Or if I am on the wrong track, what is the correct approach to modify the audio data and perform playback CORRECTLY? Any insight will be sincerely appreciated.
Thank you very much :-)
Cheers,
Manca
are you SURE the sample format is SInt16? And how many channels are there? You seem to treat the audio as a single channel short stream, but suppose the format is actually dual channel Float32 or so, and you do the modifications there, than the effect would be exactly as you describe, including the noise on other channels.

Audio data streaming having latency issue in iPhone

I have written a voice streaming application in iPhone using AudioQue. At the audio recording starts I initiated the network connection and pass the instance of NSAudioOutStream to
AudioInputCallback using inUserData reference.
void AudioInputCallback(
void *inUserData,
AudioQueueRef inAQ,
AudioQueueBufferRef inBuffer,
const AudioTimeStamp *inStartTime,
UInt32 inNumberPacketDescriptions,
const AudioStreamPacketDescription *inPacketDescs) {
RecordState* recordState = (RecordState*)inUserData;
if(!recordState->recording) {
NSLog(#"Record ending...");
}
else{
[recordState->soStream write:inBuffer->mAudioData maxLength:inBuffer->mAudioDataByteSize];
NSLog([NSString stringWithFormat:#"Count:%d Size:%d¥n", sentCnt++, inBuffer->mAudioDataByteSize]);
}
recordState->currentPacket += inNumberPacketDescriptions;
AudioQueueEnqueueBuffer(recordState->queue, inBuffer, 0, NULL);
}
According to the init parameters of the AudioQueue the length of the inBuffer is 16000 bytes. However, in WIFi application works without any doubt. But in 3G network client-server commutation is not stable.
Anybody has got the same experience or someone can suggest a tip to solve this.
One way for fixing this matter is use Queue to insert audio buffer (size of 16000 bytes) and initial some other thread to enqueue the buffers time by time and send to server.
But anybody can tell me how to synchronize one queue among two thread.