捕获立体声音频数据

Question

我有一个 MacOS Swift 应用程序，可以处理从麦克风录制的音频数据。麦克风具有立体声功能，但我只能录制单声道数据。

在下面的代码中，如果我 let alwaysMono = true，func setup() 报告活动格式是立体声，但将其覆盖为单声道。一切都适用于单声道输入流。

如果我let alwaysMono = false，setup() 将 nChannels 设置为 2。但 captureOutput 没有得到任何数据。从 UnsafeMutableAudioBufferListPointer 返回的 AudioBuffer 始终具有 nil mData。如果我不检查 nil mData，程序就会崩溃。

如何获得完整的立体声输入？

编辑：在 captureOutput 中，CMSampleBufferGetAudioBufferListWithRetainedBlockBuffer 返回错误代码 -12737，对应于 kCMSampleBufferError_ArrayTooSmall。我检查了传入 captureOutput 的 sampleBuffer arg，我看不出它有任何明显的错误。但我不知道该寻找什么。

另一个编辑：我使用内置单声道麦克风的代码测试了我的代码，令我惊讶的是它认为它也是立体声的，这表明我获取和使用 AVCaptureDevice.activeFormat 的方式显然有问题。我不知道从这里去哪里。

class Recorder: NSObject, AVCaptureAudioDataOutputSampleBufferDelegate {
    let alwaysMono = false
    var nChannels:UInt32 = 1
    let session : AVCaptureSession!
    static let realTimeQueue = DispatchQueue(label: "com.myapp.realtime",
                                             qos: DispatchQoS( qosClass:DispatchQoS.QoSClass.userInitiated, relativePriority: 0 ))
    override init() {
        session = AVCaptureSession()
        super.init()
    }
    static var recorder:Recorder?
    static func record() ->Bool {
        if recorder == nil {
            recorder = Recorder()
            if !recorder!.setup(callback:record) {
                recorder = nil
                return false
            }
        }
        realTimeQueue.async {
            if !recorder!.session.isRunning {
                recorder!.session.startRunning()
            }
        }
        return true
    }
    static func pause() {
        recorder!.session.stopRunning()
    }
    func setup( callback:@escaping (()->Bool)) -> Bool {
        let device = AVCaptureDevice.default( for: AVMediaType.audio )
        if device == nil { return false }
        if let format = Recorder.getActiveFormat() {
            nChannels = format.mChannelLayoutTag == kAudioChannelLayoutTag_Stereo ? 2 : 1
            print("active format is \((nChannels==2) ? "Stereo" : "Mono")")
            if alwaysMono {
                print( "Overriding to mono" )
                nChannels = 1
            }
        }
        if #available(OSX 10.14, *) {
            let status = AVCaptureDevice.authorizationStatus( for:AVMediaType.audio )
            if status == .notDetermined {
                AVCaptureDevice.requestAccess(for: AVMediaType.audio ){ granted in
                    _ = callback()
                }
                return false
            } else if status != .authorized {
                return false
            }
        }
        var input : AVCaptureDeviceInput
        do {
            try device!.lockForConfiguration()
            try input = AVCaptureDeviceInput( device: device! )
            device!.unlockForConfiguration()
        } catch {
            device!.unlockForConfiguration()
            return false
        }
        let output = AVCaptureAudioDataOutput()
        output.setSampleBufferDelegate(self, queue: Intonia.realTimeQueue)
        let settings = [
            AVFormatIDKey: kAudioFormatLinearPCM,
            AVNumberOfChannelsKey : nChannels,
            AVSampleRateKey : 44100,
            AVLinearPCMBitDepthKey : 16,
            AVLinearPCMIsFloatKey : false
            ] as [String : Any]
        output.audioSettings = settings
        session.beginConfiguration()
        if !session.canAddInput( input ) {
            return false
        }
        session.addInput( input )
        if !session.canAddOutput( output ) {
            return false
        }
        session.addOutput( output )
        session.commitConfiguration()
        return true
    }
    func getActiveFormat() -> AudioFormatListItem? {
        if #available(OSX 10.15, *) {
            let device = AVCaptureDevice.default( for: AVMediaType.audio )
            if device == nil { return nil }
            let list = device!.activeFormat.formatDescription.audioFormatList
            if list.count < 1 { return nil }
            return list[0]
        }
        return nil
    }
    func captureOutput(_ captureOutput: AVCaptureOutput,
                       didOutput sampleBuffer: CMSampleBuffer,
                       from connection: AVCaptureConnection){
        var buffer: CMBlockBuffer? = nil
        var audioBufferList = AudioBufferList(
            mNumberBuffers: 1,
            mBuffers: AudioBuffer(mNumberChannels: nChannels, mDataByteSize: 0, mData: nil)
        )
        CMSampleBufferGetAudioBufferListWithRetainedBlockBuffer(
            sampleBuffer,
            bufferListSizeNeededOut: nil,
            bufferListOut: &audioBufferList,
            bufferListSize: MemoryLayout<AudioBufferList>.size,
            blockBufferAllocator: nil,
            blockBufferMemoryAllocator: nil,
            flags: UInt32(kCMSampleBufferFlag_AudioBufferList_Assure16ByteAlignment),
            blockBufferOut: &buffer
        )
        let abl = UnsafeMutableAudioBufferListPointer(&audioBufferList)
        for buff in abl {
            if buff.mData != nil {
                let count = Int(buff.mDataByteSize)/MemoryLayout<Int16>.size
                let samples = UnsafeMutablePointer<Int16>(OpaquePointer(buff.mData))
                process(samples:samples!, count:count)
            } else {
                print("No data!")
            }
        }
    }
    func process( samples: UnsafeMutablePointer<Int16>, count: Int ) {
        let firstValue = samples[0]
        print( "\(count) values received, first is \(firstValue)" )
    }
}

Answer 1

这个问题可能有很多地方出错。首先，这取决于您使用的麦克风。它是你 Mac 上的那个吗？如果是这样，那么它通常是单声道的。但是，您可以让自己成为一种解决方法来测试它是否有效并且不依赖于 always: bool。尝试录制两次，用单声道录制第一个通道，然后用第二个通道录制第二个，使用不同的“加倍”代码。

我看到的第二件事可能是在 setup func 中显式键入立体声通道，以确保它正在获取它的 2 个通道并使用带有显式 == true 或 false 语句的 alwaysMono。看看你是否可以在代码期间打印一些频道信息，这样你就知道哪里出了问题。