使用 AVAssetWriter 录制无缝音频

我正在尝试录制音频片段并重新组合它们，而不产生音频间隙。

最终的目标是也有视频，但我发现音频本身在与ffmpeg -f concat -i list.txt -c copy out.mp4

如果我将音频放入 HLS 播放列表中，也会有间隙，所以我认为这不是 ffmpeg 所独有的。

这个想法是样本连续进来，我的控制器将样本路由到正确的位置AVAssetWriter。如何消除音频中的间隙？

import Foundation
import UIKit
import AVFoundation

class StreamController: UIViewController, AVCaptureAudioDataOutputSampleBufferDelegate, AVCaptureVideoDataOutputSampleBufferDelegate {
    var closingAudioInput: AVAssetWriterInput?
    var closingAssetWriter: AVAssetWriter?

    var currentAudioInput: AVAssetWriterInput?
    var currentAssetWriter: AVAssetWriter?

    var nextAudioInput: AVAssetWriterInput?
    var nextAssetWriter: AVAssetWriter?

    var videoHelper: VideoHelper?

    var startTime: NSTimeInterval = 0
    let closeAssetQueue: dispatch_queue_t = dispatch_queue_create("closeAssetQueue", nil);

    override func viewDidLoad() {
        super.viewDidLoad()
        startTime = NSDate().timeIntervalSince1970
        createSegmentWriter()
        videoHelper = VideoHelper()
        videoHelper!.delegate = self
        videoHelper!.startSession()
        NSTimer.scheduledTimerWithTimeInterval(1, target: self, selector: "createSegmentWriter", userInfo: nil, repeats: true)
    }

    func createSegmentWriter() {
        print("Creating segment writer at t=\(NSDate().timeIntervalSince1970 - self.startTime)")
        let outputPath = OutputFileNameHelper.instance.pathForOutput()
        OutputFileNameHelper.instance.incrementSegmentIndex()
        try? NSFileManager.defaultManager().removeItemAtPath(outputPath)
        nextAssetWriter = try! AVAssetWriter(URL: NSURL(fileURLWithPath: outputPath), fileType: AVFileTypeMPEG4)
        nextAssetWriter!.shouldOptimizeForNetworkUse = true

        let audioSettings: [String:AnyObject] = EncodingSettings.AUDIO
        nextAudioInput = AVAssetWriterInput(mediaType: AVMediaTypeAudio, outputSettings: audioSettings)
        nextAudioInput!.expectsMediaDataInRealTime = true
        nextAssetWriter?.addInput(nextAudioInput!)

        nextAssetWriter!.startWriting()
    }

    func closeWriterIfNecessary() {
        if closing && audioFinished {
            closing = false
            audioFinished = false
            let outputFile = closingAssetWriter?.outputURL.pathComponents?.last
            closingAssetWriter?.finishWritingWithCompletionHandler() {
                let delta = NSDate().timeIntervalSince1970 - self.startTime
                print("segment \(outputFile!) finished at t=\(delta)")
            }
            self.closingAudioInput = nil
            self.closingAssetWriter = nil
        }
    }

    var audioFinished = false
    var closing = false

    func captureOutput(captureOutput: AVCaptureOutput!, didOutputSampleBuffer sampleBuffer: CMSampleBufferRef, fromConnection connection: AVCaptureConnection!) {
        if let nextWriter = nextAssetWriter {
            if nextWriter.status.rawValue != 0 {
                if (currentAssetWriter != nil) {
                    closing = true
                }

                var sampleTiming: CMSampleTimingInfo = kCMTimingInfoInvalid
                CMSampleBufferGetSampleTimingInfo(sampleBuffer, 0, &sampleTiming)

                print("Switching asset writers at t=\(NSDate().timeIntervalSince1970 - self.startTime)")
                closingAssetWriter = currentAssetWriter
                closingAudioInput = currentAudioInput

                currentAssetWriter = nextAssetWriter
                currentAudioInput = nextAudioInput

                nextAssetWriter = nil
                nextAudioInput = nil

                currentAssetWriter?.startSessionAtSourceTime(sampleTiming.presentationTimeStamp)
            }
        }

        if let _ = captureOutput as? AVCaptureVideoDataOutput {
        } else if let _ = captureOutput as? AVCaptureAudioDataOutput {
            captureAudioSample(sampleBuffer)
        }

        dispatch_async(closeAssetQueue) {
            self.closeWriterIfNecessary()
        }
    }

    func printTimingInfo(sampleBuffer: CMSampleBufferRef, prefix: String) {
        var sampleTiming: CMSampleTimingInfo = kCMTimingInfoInvalid
        CMSampleBufferGetSampleTimingInfo(sampleBuffer, 0, &sampleTiming)
        let presentationTime = Double(sampleTiming.presentationTimeStamp.value) / Double(sampleTiming.presentationTimeStamp.timescale)
        print("\(prefix):\(presentationTime)")
    }

    func captureAudioSample(sampleBuffer: CMSampleBufferRef) {
        printTimingInfo(sampleBuffer, prefix: "A")
        if (closing && !audioFinished) {
            if closingAudioInput?.readyForMoreMediaData == true {
                closingAudioInput?.appendSampleBuffer(sampleBuffer)
            }
            closingAudioInput?.markAsFinished()
            audioFinished = true
        } else {
            if currentAudioInput?.readyForMoreMediaData == true {
                currentAudioInput?.appendSampleBuffer(sampleBuffer)
            }
        }
    }
}

对于像 AAC 这样的数据包格式，您在开始处有静默启动帧（也称为编码器延迟），在结尾处有剩余帧（当您的音频长度不是数据包大小的倍数时）。在你的例子中，每个文件的开头有 2112 个。启动帧和剩余帧破坏了在不转码的情况下连接文件的可能性，因此您不能真正责怪ffmpeg -c copy因为不产生无缝输出。

我不确定这会给您带来视频的什么结果 - 显然，即使存在启动帧，音频也会与视频同步。

这完全取决于您打算如何连接最终的音频（以及最终的视频）。如果您自己使用AVFoundation，那么您可以使用以下方法检测并解释启动/剩余帧

CMGetAttachment(buffer, kCMSampleBufferAttachmentKey_TrimDurationAtStart, NULL) 
CMGetAttachment(audioBuffer, kCMSampleBufferAttachmentKey_TrimDurationAtEnd, NULL) 

作为短期解决方案，您可以切换到非“打包”以获得无缝、可连接（使用 ffmpeg）文件。

e.g.

AVFormatIDKey: kAudioFormatAppleIMA4, fileType: AVFileTypeAIFC、后缀“.aifc”或AVFormatIDKey: kAudioFormatLinearPCM, fileType: AVFileTypeWAVE，后缀“.wav”

附注您可以使用无处不在的命令查看启动和剩余帧以及数据包大小afinfo tool.

afinfo chunk.mp4

数据格式：2通道，44100 Hz，'aac'（0x00000000）0位/通道，0字节/包，1024帧/包，0字节/帧
...
音频 39596 个有效帧 + 2112 个启动帧 + 276 个剩余帧 = 41984
...