過去半年參與的一個項(xiàng)目中有遇到并解決了一系列關(guān)于音頻文件的操作方式。相關(guān)資料在網(wǎng)上很難找到，所以先整理下以方便今后查漏補(bǔ)缺。
1.如果你需要分析的音頻文件是在服務(wù)端，那需要先將文件下載在本地，所以我們先實(shí)現(xiàn)根據(jù)URL地址下載音頻文件的功能。
import AVKit
typealias AudioDownloadedCallback = (URL?) -> ()
extension AVAsset {
static func downloadAudioToLocal(audioURL:URL,downloadCallback:@escaping AudioDownloadedCallback) {
// create your destination file url
let destinationUrl = FileManager.getDocumentPathWithFileName(sFileName: audioURL.lastPathComponent)
// to check if it exists before downloading it
if FileManager.default.fileExists(atPath: destinationUrl.path) {
//The file already exists at path
downloadCallback(destinationUrl)
} else {
//download audio file.
URLSession.shared.downloadTask(with: audioURL, completionHandler: { (location, response, error) -> Void in
guard let location = location, error == nil else { return }
do {
//File moved to documents folder
try FileManager.default.moveItem(at: location, to: destinationUrl)
downloadCallback(destinationUrl)
} catch let error as NSError {
print(error.localizedDescription)
downloadCallback(nil)
}
}).resume()
}
}
}
extension FileManager {
static func getDocumentPathWithFileName(sFileName:String) -> URL {
let fileMgr = FileManager.default
let dirPaths = fileMgr.urls(for: .documentDirectory,
in: .userDomainMask).first!
let filePath = dirPaths.appendingPathComponent(sFileName)
return filePath
}
}
文件有了后可以開始讀取文件。這里有兩種方式：
1.通過AVAssetReader進(jìn)行讀取并用NSMutableData記錄下來。
2.生成AVAudioFile音頻實(shí)例，配合AVAudioPCMBuffer進(jìn)行分段讀取。
兩者的區(qū)別在于：第一種方式是整體讀取，當(dāng)AVAssetReader開始執(zhí)行讀取后，讀取的內(nèi)容段并不受控制，每段讀取的長度由CMBlockBufferGetDataLength()計算好了返回。而第二種方式可以自己定義讀取的起始點(diǎn)和步長，你甚至可以選擇性的重復(fù)讀取。具體使用哪個看自己的業(yè)務(wù)場景。
整體讀取AVAssetReader + NSMutableData
import UIKit
import AVFoundation
import AVKit
extension AVURLAsset {
func readAudioBuffer() -> NSMutableData {
var reader:AVAssetReader?
do {
reader = try AVAssetReader(asset: self)
} catch let error{
print("Create asset reader failed.\(error.localizedDescription)")
}
let sampleData = NSMutableData()
if reader != nil {
//we only read one tracks from the audio file. If your file is not mone and you need to read them all. just duplicate operation once again.
let songTrack:AVAssetTrack = self.tracks[0]
let readerOutputSettings: [String: Int] = [ AVFormatIDKey: Int(kAudioFormatLinearPCM),
AVLinearPCMBitDepthKey: 16,
AVLinearPCMIsBigEndianKey: 0,
AVLinearPCMIsFloatKey: 0,
AVLinearPCMIsNonInterleaved: 0]
let output = AVAssetReaderTrackOutput(track: songTrack, outputSettings: readerOutputSettings)
reader?.add(output)
reader?.startReading()
while reader?.status == AVAssetReader.Status.reading {
if let sampleBufferRef = output.copyNextSampleBuffer() {
if let blockBufferRef = CMSampleBufferGetDataBuffer(sampleBufferRef) {
let bufferLength = CMBlockBufferGetDataLength(blockBufferRef)
let data = NSMutableData(length: bufferLength)
CMBlockBufferCopyDataBytes(blockBufferRef, atOffset: 0, dataLength: bufferLength, destination: (data?.mutableBytes)!)
let samples = UnsafeMutablePointer<Int16>(OpaquePointer(UnsafeMutableRawPointer(data!.mutableBytes)))
sampleData.append(samples, length: bufferLength)
CMSampleBufferInvalidate(sampleBufferRef)
}
}
}
}
return sampleData
}
}
PS：代碼中注釋處標(biāo)明這段代碼只是讀第0條音軌。如果你需要分析的文件是立體聲的(也就是存在兩條或以上的音軌)，需要針對不同的音軌進(jìn)行調(diào)整。
效果：
3211552369372_.pic_hd.jpg
拿到結(jié)果后可以根據(jù)自己需求進(jìn)行再加工，比如我自己實(shí)現(xiàn)了一個根據(jù)UI要求的高寬來對音頻數(shù)據(jù)進(jìn)行偏移處理。
extension NSMutableData {
func toInt16Sequence(size:CGSize) -> [Float] {
var filteredSamplesMA:[Float] = [Float]()
let sampleCount = self.length / MemoryLayout<Int16>.size
let binSize = Float(sampleCount) / (Float(size.width) * 0.5)
var i = 0
while i < sampleCount {
let rangeData = self.subdata(with: NSRange(location: i, length: 1))
let item = rangeData.withUnsafeBytes({ (ptr: UnsafePointer<Int>) -> Int in
return ptr.pointee
})
filteredSamplesMA.append(Float(item))
i += Int(binSize)
}
let result = NSMutableData.trackScale(size: size, source: filteredSamplesMA)
return result
}
private static func trackScale(size: CGSize, source: [Float]) -> [Float] {
if let max = source.max() {
let k = Float(size.height) / max
return source.map{ $0 * k }
}
return source
}
}
最終效果：
3221552370149_.pic_hd.jpg
有了這些數(shù)據(jù)后可以去UI層進(jìn)行相應(yīng)的繪制，或者保存下來進(jìn)行其他方式的分析。
接下來介紹第二種方式 AVAudioFile + AVAudioPCMBuffer
這種方式不同于第一種整體讀取，而是可以自己指定每次讀取的起始位置和長度，相比較而言更加靈活一些。
import AVFoundation
extension AVAudioFile {
static func readAmplitude(audioURL:URL) -> NSMutableData {
let asset = AVURLAsset(url: audioURL)
var file:AVAudioFile!
do{
file = try AVAudioFile(forReading: audioURL)
} catch let error{
print("AVAudioFile create failed \(error.localizedDescription)")
}
let amplitudes:NSMutableData = NSMutableData()
//FPS
let frameCountPeiSecond = 30.0
//Calculate totoal frame of the audio file.
let frameCountTotal = asset.duration.seconds * frameCountPeiSecond
//merge samples of each frame
let sampleForEachFrame = Double(file.length.magnitude) / frameCountTotal
var sampleReadIndex:Double = 0.0
while sampleReadIndex < Double(file.length) {
let audioBuffer:AVAudioPCMBuffer = AVAudioPCMBuffer(pcmFormat: file.processingFormat, frameCapacity: AVAudioFrameCount(sampleForEachFrame))!
audioBuffer.frameLength = UInt32(sampleForEachFrame)
//adjust frame position each time
file.framePosition = AVAudioFramePosition(sampleReadIndex)
do{
//read same size buffer of each time
try file.read(into: audioBuffer, frameCount: UInt32(sampleForEachFrame))
amplitudes.append((audioBuffer.floatChannelData?.pointee)!, length: Int(sampleForEachFrame))
} catch let error{
print("read to buffer failed \(error.localizedDescription)")
}
//adding up
sampleReadIndex = Double(sampleReadIndex + sampleForEachFrame)
}
return amplitudes
}
}
結(jié)果打印
image.png
到此數(shù)據(jù)就都分析出來了，接下來可以隨意選用適合自己場景的第三方框架顯示出對應(yīng)的分析圖出來。
