I'm working on an audio processing app and am creating an AVAudioUnit extension as a part of it. I need to train a small neural network in the app and use it to process audio in real-time in the AudioUnit. The network is mostly convolutions and is ideal for running on the GPU but it should run in real-time on the CPU. The problem that I'm currently facing is that none of the ML frameworks seem to be safe to use for inference within custom AVAudioUnit kernels. My understanding is that only C and C++ should be used in these kernels (in addition to the other rules of real-time computing). Objective-C and Swift are discouraged per the documentation. My background is primarily in ML so I'm newer to Apple development and especially new to real-time development in this ecosystem. I've investigated CoreML, MPS, BNNS/Accelerate, and MLCompute so far but I'm not certain that any of them are safe to use. Any feedback would be greatly appreciated!

I want to process audio from videos using AVAudioEngine, but I'm not sure how to read the audio from videos using AVAudioEngine. I have the following code: let videoURL = /// url pointing to a local video file. let file = try AVAudioFile(forReading: videoURL) It works fine on macOS but on iOS it fails with the error message: [default]          ExtAudioFile.cpp:193   about to throw 'typ?': open audio file [avae]            AVAEInternal.h:109   [AVAudioFile.mm:134:AVAudioFileImpl: (ExtAudioFileOpenURL((CFURLRef)fileURL, &_extAudioFile)): error 1954115647 Error Domain=com.apple.coreaudio.avfaudio Code=1954115647 "(null)" UserInfo={failed call=ExtAudioFileOpenURL((CFURLRef)fileURL, &_extAudioFile)} Reading through the header files it seems that 'typ?' corresponds to kAudioFileUnsupportedFileTypeError, so that tells me that the file type is supported on macOS but not on iOS. So my question is: How can I work with audio from video files in an AVAudioEngine based setup? I already know that I could extract the audio from videos using something like AVAssetExportSession, but that approach requires extra preprocessing time that I rather not spend.

Hi, I have been searching all over for a way to do this on macOS: playing different stereo files on different pairs of audio outputs of the same hardware device. I am currently using AVAudioEngine with two AVAudioPlayerNodes and I can mix them and change the mapping of the entire mix through the use of AudioUnitSetProperty on the engine output, but I cannot have multiple AVAudioPlayerNodes play on different outputs. Obviously, being on macOS I cannot use AVAudioSession... Thank you if anyone has any idea on how to achieve this !

For AVAudioPlayer, there is corresponding AVAudioRecorder. For AVMIDIPLayer, I found nothing for recording from the system's active MIDI input device. Can I record midi events from the system’s active input midi device, without resolving to low level CoreMidi? After configuring AVAudioUnitSampler with just a few lines of code, import AVFoundation var engine = AVAudioEngine() let unit = AVAudioUnitSampler() engine.attach(unit) engine.connect(unit, to: engine.outputNode, format: engine.outputNode.outputFormat (forBus:0)) try! unit.loadInstrument(at:sndurl) //url to .sf2 file try! engine.start() I could send midi events programmatically. // feeding AVAudioUnitMIDIInstrument with midi data let range = (0..<100) let midiStart = range.map { _ in UInt8.random(in: 70...90) } let midiStop = [0] + midiStart let times = range.map { _ in TimeInterval.random(in: 0...100) * 0.3 } for i in range {     DispatchQueue.main.asyncAfter(deadline: .now()+TimeInterval(times[i])){         unit.stopNote(midiStop[i], onChannel: 1)         unit.startNote(midiStart[i], withVelocity: 127, onChannel: 1)     } } But instead, I need to send midi events from a midi instrument, and tap to them for recording.

I have a very simple setup involving AVAudioEngine, and a player node, AVAudioPlayerNode.The player node is attached to the audio engine, then connected to the engine's main mixer node.Then, to start outputing some sound, I start the engine, and finally play the node :NSError* err = nil; BOOL started = [_audioEngine startAndReturnError:&amp;amp;err]; if(started) { [_playerNode play]; } else { // handle error }It happens sometimes, the app crashes (on the call to the play function) because, as the system says : "player started when engine not running".I can't understand how this can happen, at least in the code above.No error is returned in `err`.Does somebody knows what happen or faced such a situation ?

Our UE4 game is stuned when received notification from other apps. It will take 50ms-500ms to call [AVAudioSession sharedInstance]. This cause the main game loop can not to call FAppEntry::Tick() before the func returns.

Hi, my name is Jakob, we are working on sound creation apps with a visual and spatial approach in Unity. For iOS, we are using the "Unity as a library"-approach, so we have a native app part which is written in Swift and access to the AVAudioSession. Everything works great, except for one feature: We want to send the audio data generated with our app as binaural mix to another host app. To do so, we need to feed that data into an AVAudioUnit in our native iOS app. There are different approaches and we keep on making research and working on solutions. Could your new native spatial audio Unity plug-in Apple.PHASE might be a direct bridge feeding an AVAudioNode and combine the needs in between music and spatial audio content creators. Thanks a lot in advance & all the best!

So I have successfully triggered a PTT notification, but when I try to play audio – any audio – it doesn't play. Seems to be an issue with initiating my AVAudioSession. If I do not initiate it, the sound plays (outside of the didActivateAudioSession; such as on view did load), so I know that it's not the audio playing code. For some reason, the AVAudioSession is not allowing me to play sound. Even when I put "PlayandRecord" and when I put "mix" in the options

Im guessing i already know the answer to this but is AVAudioEngine class accessible in c++? No matter what i try i can't seem to get a pointer to it.

I see in Crashlytics few users are getting this exception when connecting the inputNode to mainMixerNode in AVAudioEngine: Fatal Exception: com.apple.coreaudio.avfaudio required condition is false: format.sampleRate == hwFormat.sampleRate Here is my code: self.engine = AVAudioEngine() let format = engine.inputNode.inputFormat(forBus: 0) //main mixer node is connected to output node by default engine.connect(self.engine.inputNode, to: self.engine.mainMixerNode, format: format) Just want to understand how can this error occur and what is the right fix?

I have a working AUv3 AUAudioUnit app extension but I had to work around a strange issue: I found that the internalRenderBlock value is fetched and invoked before allocateRenderResources() is called. I have not found any documentation stating that this would be the case, and intuitively it does not make any sense. Is there something I am doing in my code that would be causing this to be the case? Should I *force* a call to allocateRenderResources() if it has not been called before internalRenderBlock is fetched? Thanks! Brad

Every time the AVAudioSession category is re-activated ( after being inactivated) and the audioengine is restarted (calling stop() and play()), the output from audioplayer node seems to ignore the audio session category, until explicitly connecting the audio player node again using audioEngine.connect(playerNode, to: audioEngine.outputNode, format: audioFile.processingFormat) The documentation regarding this behavior is not clear and would like to clarify the following. Should audioEngine.connect be called everytime the AVAudioSession is activated? Should audioEngine.connect be called after the audio player engine is stopped (audioEngine.stop())?

I am trying to mix the audio from 2 different hardware audio devices together in real-time and record the results. Does anybody have any idea how to do this? This is on macOS. Things I have tried and why it didn't work: Adding 2 audio AVCaptureDevices to an AVCaptureMovieFileOutput or AVAssetWriter. This results in a file that has 2 audio tracks. This doesn't work for me for various reasons. Sure I can mix them together with an AVAssetExportSession, but it needs to be real-time. Programmatically creating an aggregate device and recording that as an AVCaptureDevice. This "sort of" works, but it always results in a recording with strange channel issues. For example, if I combine a 1 channel mic and a 2 channel device, I get a recording with 3 channel audio (L R C). If I make an aggregate out of 2 stereo devices, I get a recording with quadraphonic sound(L R Ls Rs), which won't even play back on some players. If I always force it to stereo, all stereo tracks get turned to mono for some reason. Programmatically creating an aggregate device and trying to use it in an AVAudioEngine. I've had multiple problems with this, but the main one is that when the aggregate device is an input node, it only reports the format of its main device, and no sub-devices. And I can't force it to be 3 or 4 channels without errors. Use an AVCaptureSession to output the sample buffers of both devices, then convert and put those samples into their own AVPlayerNodes. Then mix those AVPlayerNodes into an AVAudioEngine mixer. This actually works, but the resulting audio lags so far behind real-time, that it is unusable. If I record a webcam video along with the audio, the lip-sync is off by like half a second. I really need help with this. If anybody has a way to do this, let me know. Some caveats that have also been tripping me up: The hardware devices that need to be recorded might not be the default input device for the system. The MBP built in mic might be the default device, but I need to record 2 other devices and disclose the built in mic. The devices usually don't have the same audio format. I might be mixing an lpcm mono int16 interleaved with a lpcm stereo float32 non-interleaved. It absolutely has to be real-time and 1 single audio track. It shouldn't be this hard, right?

Hi, I'm trying to send audio data via UDP. I am using Network.framework with networking, so to use send method in NWConnection sending data must be Data type or confirm to DataProtocol. To satisfy those conditions, I have implemented a method to convert from AVAudioPCMBuffer type to Data type. func makeDataFromPCMBuffer(buffer: AVAudioPCMBuffer, time: AVAudioTime) -> Data {         let audioBuffer = buffer.audioBufferList.pointee.mBuffers         let data: Data!         data = .init(bytes: audioBuffer.mData!, count: Int(audioBuffer.mDataByteSize))         return data     } Implementation above is referenced from this post The problem is that the size of converted data is too big to fit in UDP datagram and error below occurs when I try to send data. I have found out that initial size of buffer is too big to fit in maximumDatagramSize. Below is code regarding to buffer.         let tapNode: AVAudioNode = mixerNode         let format = tapNode.outputFormat(forBus: 0)         tapNode.installTap(onBus: 0, bufferSize: 4096, format: format, block: { (buffer, time) in          // size of buffer: AVAudioPCMBuffer is 19200 already.             let bufferData = self.makeDataFromPCMBuffer(buffer: buffer, time: time)             sharedConnection?.sendRecordedBuffer(buffer: bufferData)         }) I need to reduce size of AVAudioPCMBuffer to fit in UDP datagram, But I can't find right way to do it. What would be best way to make data fit in datagram? I thought of dividing data in half, but this is UDP so I'm not sure how to handle those datas when one data has lost. So I'm trying to make AVAudioPCMBuffer fit in datagram. Any help would be very appreciated!

Hi, I'm creating a process to read an existing audio file, add an effect using AVAudioEngine, and then save it as another audio file. However, with the following method using an AVAudioPlayerNode, the save process must wait until the end of playback. import UIKit import AVFoundation class ViewController: UIViewController {          let engine = AVAudioEngine()     let playerNode = AVAudioPlayerNode()     let reverbNode = AVAudioUnitReverb()          override func viewDidLoad() {         super.viewDidLoad()         do {                          let url = URL(fileURLWithPath: Bundle.main.path(forResource: "original", ofType: "mp3")!)             let file = try AVAudioFile(forReading: url)                          // playerNode             engine.attach(playerNode)             // reverbNode             reverbNode.loadFactoryPreset(.largeChamber)             reverbNode.wetDryMix = 5.0             engine.attach(reverbNode)                          engine.connect(playerNode, to: reverbNode, format: file.processingFormat)             engine.connect(reverbNode, to: engine.mainMixerNode, format: file.processingFormat)             playerNode.scheduleFile(file, at: nil, completionCallbackType: .dataPlayedBack){ [self] _ in                 reverbNode.removeTap(onBus: 0)             }             // start             try engine.start()             playerNode.play()                          let url2 = URL(fileURLWithPath: fileInDocumentsDirectory(filename: "changed.wav"))             let outputFile = try! AVAudioFile(forWriting: url2, settings: playerNode.outputFormat(forBus: 0).settings)             reverbNode.installTap(onBus: 0, bufferSize: AVAudioFrameCount(reverbNode.outputFormat(forBus: 0).sampleRate), format: reverbNode.outputFormat(forBus: 0)) { (buffer, when) in                 do {                     try outputFile.write(from: buffer)                 } catch let error {                     print(error)                 }             }         } catch {             print(error.localizedDescription)         }     }     func getDocumentsURL() -> NSURL {         let documentsURL = FileManager.default.urls(for: .documentDirectory, in: .userDomainMask)[0] as NSURL         return documentsURL     }          func fileInDocumentsDirectory(filename: String) -> String {         let fileURL = getDocumentsURL().appendingPathComponent(filename)         return fileURL!.path     } } Is there a way to complete the writing without waiting for the playback to complete? My ideal is to complete the write in the time required by CPU and storage performance. It seems that reverbNode.installTap(...) { (buffer, when) in ...} in the code is processed in parallel with the current playback position, so I would like to dramatically improve the processing speed. Best regards.

I am trying to use AVAudioEngine for listening to mic samples and playing them simultaneously via external speaker or headphones (assuming they are attached to iOS device). I tried the following using AVAudioPlayerNode and it works, but there is too much delay in the audio playback. Is there a way to hear sound realtime without delay? Why the scheduleBuffer API has so much delay I wonder. var engine: AVAudioEngine! var playerNode: AVAudioPlayerNode! var mixer: AVAudioMixerNode! var audioEngineRunning = false public func setupAudioEngine() { self.engine = AVAudioEngine() let input = engine.inputNode let format = input.inputFormat(forBus: 0) playerNode = AVAudioPlayerNode() engine.attach(playerNode) self.mixer = engine.mainMixerNode engine.connect(self.playerNode, to: self.mixer, format: playerNode.outputFormat(forBus: 0)) engine.inputNode.installTap(onBus: 0, bufferSize: 4096, format: format, block: { buffer, time in self.playerNode.scheduleBuffer(buffer, completionHandler: nil) }) do { engine.prepare() try self.engine.start() audioEngineRunning = true self.playerNode.play() } catch { print("error couldn't start engine") audioEngineRunning = false } }

I am using AVAudioSession with playAndRecord category as follows: private func setupAudioSessionForRecording() { let audioSession = AVAudioSession.sharedInstance() do { try audioSession.setActive(false) try audioSession.setPreferredSampleRate(Double(48000)) } catch { NSLog("Unable to deactivate Audio session") } let options:AVAudioSession.CategoryOptions = [.allowAirPlay, .mixWithOthers] do { try AVAudioSession.sharedInstance().setCategory(AVAudioSession.Category.playAndRecord, mode: AVAudioSession.Mode.default, options: options) } catch { NSLog("Could not set audio session category \(error)") } do { try audioSession.setActive(true) } catch { NSLog("Unable to activate AudioSession") } } Next I use AVAudioEngine to repeat what I say in the microphone to external speakers (on the TV connected with iPhone using HDMI Cable). //MARK:- AudioEngine var engine: AVAudioEngine! var playerNode: AVAudioPlayerNode! var mixer: AVAudioMixerNode! var audioEngineRunning = false public func setupAudioEngine() { self.engine = AVAudioEngine() engine.connect(self.engine.inputNode, to: self.engine.outputNode, format: nil) do { engine.prepare() try self.engine.start() } catch { print("error couldn't start engine") } audioEngineRunning = true } public func stopAudioEngine() { engine.stop() audioEngineRunning = false } The issue is I hear some kind of reverb/humming noise after I speak for a few seconds that keeps getting amplified and repeated. If I use a RemoteIO unit instead, no such noise comes out of speakers. I am not sure if my setup of AVAudioEngine is correct. I have tried all kinds of AVAudioSession configuration but nothing changes. The link to sample audio with background speaker noise is posted [here] in the Stackoverflow forum (https://stackoverflow.com/questions/72170548/echo-when-using-avaudioengine-over-hdmi#comment127514327_72170548)

I have a RemoteIO unit that successfully playbacks the microphone samples in realtime via attached headphones. I need to get the same functionality ported using AVAudioEngine, but I can't seem to make a head start. Here is my code, all I do is connect inputNode to playerNode which crashes. var engine: AVAudioEngine! var playerNode: AVAudioPlayerNode! var mixer: AVAudioMixerNode! var engineRunning = false private func setupAudioSession() { var options:AVAudioSession.CategoryOptions = [.allowBluetooth, .allowBluetoothA2DP] do { try AVAudioSession.sharedInstance().setCategory(AVAudioSession.Category.playAndRecord, mode: AVAudioSession.Mode.default, options: options) try AVAudioSession.sharedInstance().setAllowHapticsAndSystemSoundsDuringRecording(true) } catch { MPLog("Could not set audio session category") } let audioSession = AVAudioSession.sharedInstance() do { try audioSession.setActive(false) try audioSession.setPreferredSampleRate(Double(44100)) } catch { print("Unable to deactivate Audio session") } do { try audioSession.setActive(true) } catch { print("Unable to activate AudioSession") } } private func setupAudioEngine() { self.engine = AVAudioEngine() self.playerNode = AVAudioPlayerNode() self.engine.attach(self.playerNode) engine.connect(self.engine.inputNode, to: self.playerNode, format: nil) do { try self.engine.start() } catch { print("error couldn't start engine") } engineRunning = true } But starting AVAudioEngine causes a crash: libc++abi: terminating with uncaught exception of type NSException *** Terminating app due to uncaught exception 'com.apple.coreaudio.avfaudio', reason: 'required condition is false: inDestImpl->NumberInputs() > 0 || graphNodeDest->CanResizeNumberOfInputs()' terminating with uncaught exception of type NSException How do I get realtime record and playback of mic samples via headphones working?

I am working on music application where multiple wav files are scheduling within time frame. Everything is working perfect except one scenario where there is small beef is coming while scheduling player node again. For example - one.wav is playing on PlayerNode1 and now I am rescheduling to second.wav after 2 second then there is small beep is coming. I have tried to stop node by checking isPlaying condition. Still it is not working. Am I doing anything wrong here. if playerNode.isPlaying { playerNode.stop() } playerNode.scheduleFile(audioFile, at: nil, completionHandler: nil) playerNode.play() I am using same player node for performance as there are 24 wav files that needs to be played in 1 minute so there is no point to keeping all player nodes. How would stop beef while rescheduling new audio file for same player node? I have shared link below to check issue. https://drive.google.com/file/d/1FjZtLUj_wUp0LQPyjIwfJNy67HWUlt0I/view?usp=sharing Expected result should be song continuity

I'm developing a game that will use speech recognition to execute various commands. I am using code from Apple's Recognizing Speech in Live Audio documentation page. When I run this in a Swift Playground, it works just fine. However, when I make a SpriteKit game application (basic setup from Xcode's "New Project" menu option), I get the following error: required condition is false: IsFormatSampleRateAndChannelCountValid(hwFormat) Upon further research, it appears that my input node has no channels. The following is the relevant portion of my code, along with debug output: let inputNode = audioEngine.inputNode print("Number of inputs: \(inputNode.numberOfInputs)") // 1 print("Input Format: \(inputNode.inputFormat(forBus: 0))") // <AVAudioFormat 0x600001bcf200: 0 ch, 0 Hz, 'lpcm' (0x00000029) 32-bit little-endian float, deinterleaved> let channelCount = inputNode.inputFormat(forBus: 0).channelCount print("Channel Count: \(channelCount)") // 0 <== Agrees with the inputFormat output listed previously // Configure the microphone input. print("Number of outputs: \(inputNode.numberOfOutputs)") // 1 let recordingFormat = inputNode.outputFormat(forBus: 0) print("Output Format: \(recordingFormat)") // <AVAudioFormat 0x600001bf3160: 2 ch, 44100 Hz, Float32, non-inter> inputNode.installTap(onBus: 0, bufferSize: 256, format: recordingFormat, block: audioTap) // <== This is where the error occurs. // NOTE: 'audioTap' is a function defined in this class. Using this defined function instead of an inline, anonymous function. The code snippet is included in the game's AppDelegate class (which includes import statements for Cocoa, AVFoundation, and Speech), and executes during its applicationDidFinishLaunching function. I'm having trouble understanding why Playground works, but a game app doesn't work. Do I need to do something specific to get the application to recognize the microphone? NOTE: This if for MacOS, NOT iOS. While the "How To" documentation cited earlier indicates iOS, Apple stated at WWDC19 that it is now supported on the MacOS. NOTE: I have included the NSSpeechRecognitionUsageDescription key in the applications plist, and successfully acknowledged the authorization request for the microphone.