AudioWorklet - 將輸出設置為 Float32Array 以流式傳輸實時音頻？

Question

我有從服務器到客戶端的音頻數據流。 它從一個 Node.js 緩沖區（一個 Uint8Array）開始，然后通過 port.postMessage() 發送到 AudiWorkletProcessor，在那里它被轉換成一個 Float32Array 並存儲在 this.data 中。 我花了幾個小時試圖將輸出設置為 Float32Array 中包含的音頻數據。 記錄 Float32Array 預處理顯示准確的數據，但在處理過程中記錄它表明在發布新消息時它沒有改變。 這可能是我的低級音頻編程知識的一個空白。

當數據到達客戶端時，調用以下函數：

  process = (data) => {
        this.node.port.postMessage(data)
  }

順便說一句，（你可以讓我知道）也許我應該使用參數描述符而不是 postMessage？ 無論如何，這是我的 AudioWorkletProcessor：

class BypassProcessor extends AudioWorkletProcessor {

  constructor() {
    super();
    this.isPlaying = true;
    this.port.onmessage = this.onmessage.bind(this)
  }

  static get parameterDescriptors() {
    return [{ // Maybe we should use parameters. This is not utilized at present.
      name: 'stream',
      defaultValue: 0.707
    }];
  }

  convertBlock = (incomingData) => { // incoming data is a UInt8Array
    let i, l = incomingData.length;
    let outputData = new Float32Array(incomingData.length);
    for (i = 0; i < l; i++) {
      outputData[i] = (incomingData[i] - 128) / 128.0;
    }
    return outputData;
  }

  onmessage(event) {
    const { data } = event;
    let ui8 = new Uint8Array(data);
    this.data = this.convertBlock(ui8)
  }

  process(inputs, outputs) {
    const input = inputs[0];
    const output = outputs[0];
    if (this.data) {
      for (let channel = 0; channel < output.length; ++channel) {
        const inputChannel = input[channel]
        const outputChannel = output[channel]
        for (let i = 0; i < inputChannel.length; ++i) {
          outputChannel[i] = this.data[i]
        }
      }
    }

    return true;

  }
}

registerProcessor('bypass-processor', BypassProcessor);

如何簡單地將 AudioWorkletProcessor 的輸出設置為通過的數據？

Answer 1

AudioWorkletProcessor 僅處理每 128 個字節，因此您需要管理自己的緩沖區以確保AudioWorklet是這種情況，可能是通過添加 FIFO。 我使用在 WebAssembly 中實現的 RingBuffer(FIFO) 解決了類似的問題，在我的情況下，我收到了一個 160 字節的緩沖區。

看看我的 AudioWorkletProcessor 實現

import Module from './buffer-kernel.wasmodule.js';
import { HeapAudioBuffer, RingBuffer, ALAW_TO_LINEAR } from './audio-helper.js';

class SpeakerWorkletProcessor extends AudioWorkletProcessor {
  constructor(options) {
    super();
    this.payload = null;
    this.bufferSize = options.processorOptions.bufferSize; // Getting buffer size from options
    this.channelCount = options.processorOptions.channelCount;
    this.inputRingBuffer = new RingBuffer(this.bufferSize, this.channelCount);
    this.outputRingBuffer = new RingBuffer(this.bufferSize, this.channelCount);
    this.heapInputBuffer = new HeapAudioBuffer(Module, this.bufferSize, this.channelCount);
    this.heapOutputBuffer = new HeapAudioBuffer(Module, this.bufferSize, this.channelCount);
    this.kernel = new Module.VariableBufferKernel(this.bufferSize);
    this.port.onmessage = this.onmessage.bind(this);
  }

  alawToLinear(incomingData) {
    const outputData = new Float32Array(incomingData.length);
    for (let i = 0; i < incomingData.length; i++) {
      outputData[i] = (ALAW_TO_LINEAR[incomingData[i]] * 1.0) / 32768;
    }
    return outputData;
  }

  onmessage(event) {
    const { data } = event;
    if (data) {
      this.payload = this.alawToLinear(new Uint8Array(data)); //Receiving data from my Socket listener and in my case converting PCM alaw to linear
    } else {
      this.payload = null;
    }
  }

  process(inputs, outputs) {
    const output = outputs[0];
    if (this.payload) {
      this.inputRingBuffer.push([this.payload]); // Pushing data from my Socket

      if (this.inputRingBuffer.framesAvailable >= this.bufferSize) { // if the input data size hits the buffer size, so I can "outputted"  
        this.inputRingBuffer.pull(this.heapInputBuffer.getChannelData());
        this.kernel.process(
          this.heapInputBuffer.getHeapAddress(),
          this.heapOutputBuffer.getHeapAddress(),
          this.channelCount,
        );
        this.outputRingBuffer.push(this.heapOutputBuffer.getChannelData());
      }
      this.outputRingBuffer.pull(output); // Retriving data from FIFO and putting our output
    }
    return true;
  }
}

registerProcessor(`speaker-worklet-processor`, SpeakerWorkletProcessor);

查看 AudioContext 和 AudioWorklet 實例

 this.audioContext = new AudioContext({
      latencyHint: 'interactive',
      sampleRate: this.sampleRate,
      sinkId: audioinput || "default"
    });

    this.audioBuffer = this.audioContext.createBuffer(1, this.audioSize, this.sampleRate);
    this.audioSource = this.audioContext.createBufferSource();
    this.audioSource.buffer = this.audioBuffer;
    this.audioSource.loop = true;
    this.audioContext.audioWorklet
    .addModule('workers/speaker-worklet-processor.js')
    .then(() => {
      this.speakerWorklet = new AudioWorkletNode(
        this.audioContext,
        'speaker-worklet-processor',
        {
          channelCount: 1,
          processorOptions: {
            bufferSize: 160, //Here I'm passing the size of my output, I'm just saying to RingBuffer what size I need 
            channelCount: 1,
          },
        },
      );
      this.audioSource.connect(this.speakerWorklet).connect(this.audioContext.destination);
    }).catch((err)=>{
      console.log("Receiver ", err);
    })

看看我是如何從 Socket 接收和發送數據到 audioWorklet

  protected onMessage(e: any): void { //My Socket message listener
    const { data:serverData } = e;
    const socketId = e.socketId;
    if (this.audioWalking && this.ws && !this.ws.isPaused() && this.ws.info.socketId === socketId) {
      const buffer = arrayBufferToBuffer(serverData);
      const rtp = RTPParser.parseRtpPacket(buffer);
      const sharedPayload = new Uint8Array(new SharedArrayBuffer(rtp.payload.length)); //sharing javascript buffer memory between main thread and worklet thread
      sharedPayload.set(rtp.payload, 0);
      this.speakerWorklet.port.postMessage(sharedPayload); //Sending data to worklet
    }
  } 

為了幫助人們，我把這個解決方案的重要部分放在 Github 上

• audio-worklet-processor-wasm-example

我遵循了這個例子，它解釋了 RingBuffer 是如何工作的

• wasm-環形緩沖區