Performance issues with `AVAssetWriter`

Question

I'm trying to build a Camera app that records Video and Audio buffers (AVCaptureVideoDataOutput and AVCaptureAudioDataOutput) to an mp4/mov file using AVAssetWriter.

When creating the Recording Session, I noticed that it blocks for around 5-7 seconds before starting the recording, so I dug deeper to find out why.

This is how I create my AVAssetWriter:

let assetWriter = try AVAssetWriter(outputURL: tempURL, fileType: .mov)
let videoWriter = self.createVideoWriter(...)
assetWriter.add(videoWriter)
let audioWriter = self.createAudioWriter(...)
assetWriter.add(audioWriter)
assetWriter.startWriting()

There's two slow parts here in that code:

1. The createAudioWriter(...) function takes ages!

   This is how I create the audio AVAssetWriterInput:

   // audioOutput is my AVCaptureAudioDataOutput, audioInput is the microphone
   let settings = audioOutput.recommendedAudioSettingsForAssetWriter(writingTo: .mov)
   let format = audioInput.device.activeFormat.formatDescription
   
   let audioWriter = AVAssetWriterInput(mediaType: .audio,
                                        outputSettings: settings,
                                        sourceFormatHint: format)
   audioWriter.expectsMediaDataInRealTime = true
   

   The above code takes up to 3000ms on an iPhone 11 Pro!

   When I remove the recommended settings and just pass nil as outputSettings:

   audioWriter = AVAssetWriterInput(mediaType: .audio,
                                    outputSettings: nil)
   audioWriter.expectsMediaDataInRealTime = true
   

   ...It initializes almost instantly - something like 30 to 50ms.

2. Starting the AVAssetWriter takes ages!

   Calling this method:

   assetWriter.startWriting()
   

   ...takes takes 3000 to 5000ms on my iPhone 11 Pro!

Does anyone have any ideas why this is so slow? Am I doing something wrong?

It feels like passing nil as the outputSettings is not a good idea, and recommendedAudioSettingsForAssetWriter should be the way to go, but 3 seconds initialization time is not acceptable.

Here's the full code: RecordingSession.swift from react-native-vision-camera. This gets called from here.

I'd appreciate any help, thanks!

EDIT:

These are the recommendedVideoSettingsForAssetWriter settings I get:

["AVVideoCompressionPropertiesKey": {
    AllowFrameReordering = 1;
    AllowOpenGOP = 1;
    AverageBitRate = 47848572;
    BaseLayerFrameRate = 15;
    ExpectedFrameRate = 60;
    MaxAllowedFrameQP = 41;
    MaxKeyFrameIntervalDuration = 1;
    MinAllowedFrameQP = 15;
    MinimizeMemoryUsage = 1;
    Priority = 80;
    ProfileLevel = "HEVC_Main_AutoLevel";
    RealTime = 1;
    RelaxAverageBitRateTarget = 1;
},
 "AVVideoWidthKey": 2160,
 "AVVideoCodecKey": hvc1,
 "AVVideoHeightKey": 3840]

(video initializes in ~10ms)

And these are the recommendedAudioSettingsForAssetWriter(..) I get:

["AVEncoderQualityForVBRKey": 91,
 "AVEncoderBitRatePerChannelKey": 96000,
 "AVEncoderBitRateStrategyKey": AVAudioBitRateStrategy_Variable,
 "AVFormatIDKey": 1633772320,
 "AVNumberOfChannelsKey": 1,
 "AVSampleRateKey": 44100]

(audio initializes in ~1500-3000ms)

Note that subsequent calls are faster. I am also changing the AVAudioSessionCategory while configuring the audio AVAssetWriter, not sure if that has anything to do with it.

EDIT 2:

I just found out that it takes so long to initialize the audio AVAssetWriter because I start the audio AVCaptureSession while I initialize the audio AVAssetWriter:

audioQueue.async { // <-- should be async/non-blocking
  AVAudioSession.sharedInstance().setCategory(.playAndRecord,
                                              options: [.mixWithOthers,
                                                        .allowBluetoothA2DP,
                                                        .defaultToSpeaker,
                                                        .allowAirPlay])
  audioCaptureSession.startRunning() // <-- this takes 1.5-3s
}
let audioWriter = AVAssetWriterInput(...) <-- w/ recommended settings

Apparently the AVAssetWriter also waits for the AVAudioSession? Maybe there's some internal Mutex going on, because I'm 100% sure that this should run in parallel...

Either way - I can't really start the audio session in advance because I don't want to introduce that mini audio stutter if the user starts the app and has music playing, and I know that Snapchat is also achieving that on the fly - any thoughts?