ViewController.swift

import UIKit
import AVFoundation

class ViewController: UIViewController {

	var audioEngine = AVAudioEngine()
	var audioData = Data()

	override func viewDidLoad() {
		super.viewDidLoad()

		func toData(PCMBuffer: AVAudioPCMBuffer) -> Data {
			let channelCount = 1  // given PCMBuffer channel count is 1
			let channels = UnsafeBufferPointer(start: PCMBuffer.int16ChannelData, count: channelCount)
			let ch0Data = NSData(bytes: channels[0], length:Int(PCMBuffer.frameCapacity * PCMBuffer.format.streamDescription.pointee.mBytesPerFrame)) as Data
			return ch0Data
		}

		guard let outputFormat = AVAudioFormat(commonFormat: .pcmFormatInt16, sampleRate: 16_000, channels: AVAudioChannelCount(1), interleaved: false) else {
			return
		}
		let inputFormat = self.audioEngine.inputNode.outputFormat(forBus: 0)
		let converter = AVAudioConverter(from: inputFormat, to: outputFormat)
		self.audioEngine.inputNode.installTap(onBus: AVAudioNodeBus(0), bufferSize: AVAudioFrameCount(32000), format: self.audioEngine.inputNode.outputFormat(forBus: 0)) { (pcmBuffer, time) in
			guard let outputBuffer = AVAudioPCMBuffer(pcmFormat: outputFormat, frameCapacity:pcmBuffer.frameCapacity) else {
				return
			}
			outputBuffer.frameLength = pcmBuffer.frameLength
			let inputBlock: AVAudioConverterInputBlock = { inNumPackets, outStatus in
				outStatus.pointee = AVAudioConverterInputStatus.haveData
				return pcmBuffer
			}
			var  error:NSError? = nil
			converter?.convert(to: outputBuffer, error: &error, withInputFrom: inputBlock)
			self.audioData.append(toData(PCMBuffer: outputBuffer))
		}
		self.audioEngine.prepare()
		do {
			try self.audioEngine.start()
		} catch {
			print("\(error)")
		}
	}

}

因為 resample 後 buffer 有意義的長度也會改變，converter 好像會把 frameCapacity 的長度塞滿（不確定），outputBuffer 若給一樣長，在down-sampling 時就會出現重複的音，比如說「今天」，聽到結果會變成「今天今天今天」(48_000 -> 16_000 時），所以下面這段 buffer declaration

guard let outputBuffer = AVAudioPCMBuffer(pcmFormat: outputFormat, frameCapacity:pcmBuffer.frameCapacity) else {
    return
}

可以改成如下以避免此問題

let sampleRateRatio = 16_000/inputFormat.sampleRate

// ...

let outputFrameCapacity = AVAudioFrameCount(Double(pcmBuffer.frameCapacity) * sampleRateRatio)
guard let outputBuffer = AVAudioPCMBuffer(pcmFormat: outputFormat, frameCapacity: outputFrameCapacity) else {
    return
}