HTML5录音实践总结(Preact)

获取 PCM 数据

处理 PCM 数据

Float32 转 Int16

ArrayBuffer 转 Base64

PCM 文件播放

重采样

PCM 转 MP3

PCM 转 WAV

短时能量计算

Web Worker优化性能

音频存储(IndexedDB)

WebView 开启 WebRTC

获取 PCM 数据

查看 DEMO

https://github.com/deepkolos/pc-pcm-wave

样例代码:

const mediaStream = await window.navigator.mediaDevices.getUserMedia({
    audio: {
		// sampleRate: 44100, // 采样率 不生效需要手动重采样
        channelCount: 1, // 声道
        // echoCancellation: true,
        // noiseSuppression: true, // 降噪 实测效果不错
    },
})
const audioContext = new window.AudioContext()
const inputSampleRate = audioContext.sampleRate
const mediaNode = audioContext.createMediaStreamSource(mediaStream)

if (!audioContext.createScriptProcessor) {
	audioContext.createScriptProcessor = audioContext.createJavaScriptNode
}
// 创建一个jsNode
const jsNode = audioContext.createScriptProcessor(4096, 1, 1)
jsNode.connect(audioContext.destination)
jsNode.onaudioprocess = (e) => {
    // e.inputBuffer.getChannelData(0) (left)
    // 双通道通过e.inputBuffer.getChannelData(1)获取 (right)
}
mediaNode.connect(jsNode)

简要流程如下:

start=>start: 开始
getUserMedia=>operation: 获取MediaStream
audioContext=>operation: 创建AudioContext
scriptNode=>operation: 创建scriptNode并关联AudioContext
onaudioprocess=>operation: 设置onaudioprocess并处理数据
end=>end: 结束

start->getUserMedia->audioContext->scriptNode->onaudioprocess->end

停止录制只需要把 audioContext 挂在的 node 卸载即可,然后把存储的每一帧数据合并即可产出 PCM 数据

jsNode.disconnect()
mediaNode.disconnect()
jsNode.onaudioprocess = null

PCM 数据处理

通过 WebRTC 获取的 PCM 数据格式是 Float32 的, 如果是双通道录音的话, 还需要增加合并通道

const leftDataList = [];
const rightDataList = [];
function onAudioProcess(event) {
  // 一帧的音频PCM数据
  let audioBuffer = event.inputBuffer;
  leftDataList.push(audioBuffer.getChannelData(0).slice(0));
  rightDataList.push(audioBuffer.getChannelData(1).slice(0));
}

// 交叉合并左右声道的数据
function interleaveLeftAndRight(left, right) {
  let totalLength = left.length + right.length;
  let data = new Float32Array(totalLength);
  for (let i = 0; i < left.length; i++) {
    let k = i * 2;
    data[k] = left[i];
    data[k + 1] = right[i];
  }
  return data;
}

Float32 转 Int16

const float32 = new Float32Array(1)
const int16 = Int16Array.from(
	float32.map(x => (x > 0 ? x * 0x7fff : x * 0x8000)),
)

arrayBuffer 转 Base64

注意: 在浏览器上有个 btoa() 函数也是可以转换为 Base64 但是输入参数必须为字符串, 如果传递 buffer 参数会先被 toString() 然后再 Base64 , 使用 ffplay 播放反序列化的 Base64 , 会比较刺耳

使用 base64-arraybuffer 即可完成

import { encode } from 'base64-arraybuffer'

const float32 = new Float32Array(1)
const int16 = Int16Array.from(
	float32.map(x => (x > 0 ? x * 0x7fff : x * 0x8000)),
)
console.log(encode(int16.buffer))

验证 Base64 是否正确, 可以在 node 下把产出的 Base64 转换为 Int16 的 PCM 文件, 然后使用 FFPlay 播放, 看看音频是否正常播放

PCM 文件播放

# 单通道 采样率:16000 Int16
ffplay -f s16le -ar 16k -ac 1 test.pcm

# 双通道 采样率:48000 Float32
ffplay -f f32le -ar 48000 -ac 2 test.pcm

重采样/调整采样率

虽然 getUserMedia 参数可设置采样率, 但是在最新Chrome也不生效, 所以需要手动做个重采样

const mediaStream = await window.navigator.mediaDevices.getUserMedia({
    audio: {
    	// sampleRate: 44100, // 采样率 设置不生效
        channelCount: 1, // 声道
        // echoCancellation: true, // 减低回音
        // noiseSuppression: true, // 降噪, 实测效果不错
    },
})

使用 wave-resampler 即可完成

import { resample } from 'wave-resampler'

const inputSampleRate =  44100
const outputSampleRate = 16000
const resampledBuffers = resample(
    // 需要onAudioProcess每一帧的buffer合并后的数组
	mergeArray(audioBuffers),
	inputSampleRate,
	outputSampleRate,
)

PCM 转 MP3

import { Mp3Encoder } from 'lamejs'

let mp3buf
const mp3Data = []
const sampleBlockSize = 576 * 10 // 工作缓存区, 576的倍数
const mp3Encoder = new Mp3Encoder(1, outputSampleRate, kbps)
const samples = float32ToInt16(
  audioBuffers,
  inputSampleRate,
  outputSampleRate,
)

let remaining = samples.length
for (let i = 0; remaining >= 0; i += sampleBlockSize) {
  const left = samples.subarray(i, i + sampleBlockSize)
  mp3buf = mp3Encoder.encodeBuffer(left)
  mp3Data.push(new Int8Array(mp3buf))
  remaining -= sampleBlockSize
}

mp3Data.push(new Int8Array(mp3Encoder.flush()))
console.log(mp3Data)

// 工具函数
function float32ToInt16(audioBuffers, inputSampleRate, outputSampleRate) {
  const float32 = resample(
    // 需要onAudioProcess每一帧的buffer合并后的数组
    mergeArray(audioBuffers),
    inputSampleRate,
    outputSampleRate,
  )
  const int16 = Int16Array.from(
    float32.map(x => (x > 0 ? x * 0x7fff : x * 0x8000)),
  )
  return int16
}

使用 lamejs 即可, 但是体积较大(160+KB), 如果没有存储需求可使用 WAV 格式

> ls -alh
-rwxrwxrwx 1 root root  95K  4月 22 12:45 12s.mp3*
-rwxrwxrwx 1 root root 1.1M  4月 22 12:44 12s.wav*
-rwxrwxrwx 1 root root 235K  4月 22 12:41 30s.mp3*
-rwxrwxrwx 1 root root 2.6M  4月 22 12:40 30s.wav*
-rwxrwxrwx 1 root root  63K  4月 22 12:49 8s.mp3*
-rwxrwxrwx 1 root root 689K  4月 22 12:48 8s.wav*

PCM 转 WAV

function mergeArray(list) {
  const length = list.length * list[0].length
  const data = new Float32Array(length)
  let offset = 0
  for (let i = 0; i < list.length; i++) {
    data.set(list[i], offset)
    offset += list[i].length
  }
  return data
}

function writeUTFBytes(view, offset, string) {
  var lng = string.length
  for (let i = 0; i < lng; i++) {
    view.setUint8(offset + i, string.charCodeAt(i))
  }
}

function createWavBuffer(audioData, sampleRate = 44100, channels = 1) {
  const WAV_HEAD_SIZE = 44
  const buffer = new ArrayBuffer(audioData.length * 2 + WAV_HEAD_SIZE)
  // 需要用一个view来操控buffer
  const view = new DataView(buffer)
  // 写入wav头部信息
  // RIFF chunk descriptor/identifier
  writeUTFBytes(view, 0, 'RIFF')
  // RIFF chunk length
  view.setUint32(4, 44 + audioData.length * 2, true)
  // RIFF type
  writeUTFBytes(view, 8, 'WAVE')
  // format chunk identifier
  // FMT sub-chunk
  writeUTFBytes(view, 12, 'fmt')
  // format chunk length
  view.setUint32(16, 16, true)
  // sample format (raw)
  view.setUint16(20, 1, true)
  // stereo (2 channels)
  view.setUint16(22, channels, true)
  // sample rate
  view.setUint32(24, sampleRate, true)
  // byte rate (sample rate * block align)
  view.setUint32(28, sampleRate * 2, true)
  // block align (channel count * bytes per sample)
  view.setUint16(32, channels * 2, true)
  // bits per sample
  view.setUint16(34, 16, true)
  // data sub-chunk
  // data chunk identifier
  writeUTFBytes(view, 36, 'data')
  // data chunk length
  view.setUint32(40, audioData.length * 2, true)

  // 写入PCM数据
  let index = 44
  const volume = 1
  const { length } = audioData
  for (let i = 0; i < length; i++) {
    view.setInt16(index, audioData[i] * (0x7fff * volume), true)
    index += 2
  }
  return buffer
}

// 需要onAudioProcess每一帧的buffer合并后的数组
createWavBuffer(mergeArray(audioBuffers))

WAV 基本上是 PCM 加上一些音频信息

简单的短时能量计算

function shortTimeEnergy(audioData) {
  let sum = 0
  const energy = []
  const { length } = audioData
  for (let i = 0; i < length; i++) {
    sum += audioData[i] ** 2

    if ((i + 1) % 256 === 0) {
      energy.push(sum)
      sum = 0
    } else if (i === length - 1) {
      energy.push(sum)
    }
  }
  return energy
}

由于计算结果有会因设备的录音增益差异较大, 计算出数据也较大, 所以使用比值简单区分人声和噪音

查看 DEMO

const NoiseVoiceWatershedWave = 2.3
const energy = shortTimeEnergy(e.inputBuffer.getChannelData(0).slice(0))
const avg = energy.reduce((a, b) => a + b) / energy.length

const nextState = Math.max(...energy) / avg > NoiseVoiceWatershedWave ? 'voice' : 'noise'

Web Worker 优化性能

音频数据数据量较大, 所以可以使用 Web Worker 进行优化, 不卡 UI 线程

在 Webpack 项目里 Web Worker 比较简单, 安装 worker-loader 即可

preact.config.js

export default (config, env, helpers) => {
    config.module.rules.push({
        test: /\.worker\.js$/,
        use: { loader: 'worker-loader', options: { inline: true } },
      })
}

recorder.worker.js

self.addEventListener('message', event => {
  console.log(event.data)
  // 转MP3/转Base64/转WAV等等
  const output = ''
  self.postMessage(output)
}

使用 Worker

async function toMP3(audioBuffers, inputSampleRate, outputSampleRate = 16000) {
  const { default: Worker } = await import('./recorder.worker')
  const worker = new Worker()
  // 简单使用, 项目可以在recorder实例化的时候创建worker实例, 有并法需求可多个实例

  return new Promise(resolve => {
    worker.postMessage({
      audioBuffers: audioBuffers,
      inputSampleRate: inputSampleRate,
      outputSampleRate: outputSampleRate,
      type: 'mp3',
    })
    worker.onmessage = event => resolve(event.data)
  })
}

音频的存储

浏览器持久化储存的地方有 LocalStorage 和 IndexedDB , 其中 LocalStorage 较为常用, 但是只能储存字符串, 而 IndexedDB 可直接储存 Blob , 所以优先选择 IndexedDB ,使用 LocalStorage 则需要转 Base64 体积将会更大

所以为了避免占用用户太多空间, 所以选择MP3格式进行存储

> ls -alh
-rwxrwxrwx 1 root root  95K  4月 22 12:45 12s.mp3*
-rwxrwxrwx 1 root root 1.1M  4月 22 12:44 12s.wav*
-rwxrwxrwx 1 root root 235K  4月 22 12:41 30s.mp3*
-rwxrwxrwx 1 root root 2.6M  4月 22 12:40 30s.wav*
-rwxrwxrwx 1 root root  63K  4月 22 12:49 8s.mp3*
-rwxrwxrwx 1 root root 689K  4月 22 12:48 8s.wav*

IndexedDB 简单封装如下, 熟悉后台的同学可以找个 ORM 库方便数据读写

const indexedDB =
  window.indexedDB ||
  window.webkitIndexedDB ||
  window.mozIndexedDB ||
  window.OIndexedDB ||
  window.msIndexedDB

const IDBTransaction =
  window.IDBTransaction ||
  window.webkitIDBTransaction ||
  window.OIDBTransaction ||
  window.msIDBTransaction

const readWriteMode =
  typeof IDBTransaction.READ_WRITE === 'undefined'
    ? 'readwrite'
    : IDBTransaction.READ_WRITE

const dbVersion = 1
const storeDefault = 'mp3'

let dbLink

function initDB(store) {
  return new Promise((resolve, reject) => {
    if (dbLink) resolve(dbLink)

    // Create/open database
    const request = indexedDB.open('audio', dbVersion)

    request.onsuccess = event => {
      const db = request.result

      db.onerror = event => {
        reject(event)
      }

      if (db.version === dbVersion) resolve(db)
    }

    request.onerror = event => {
      reject(event)
    }

    // For future use. Currently only in latest Firefox versions
    request.onupgradeneeded = event => {
      dbLink = event.target.result
      const { transaction } = event.target

      if (!dbLink.objectStoreNames.contains(store)) {
        dbLink.createObjectStore(store)
      }

      transaction.oncomplete = event => {
        // Now store is available to be populated
        resolve(dbLink)
      }
    }
  })
}

export const writeIDB = async (name, blob, store = storeDefault) => {
  const db = await initDB(store)

  const transaction = db.transaction([store], readWriteMode)
  const objStore = transaction.objectStore(store)

  return new Promise((resolve, reject) => {
    const request = objStore.put(blob, name)
    request.onsuccess = event => resolve(event)
    request.onerror = event => reject(event)
    transaction.commit && transaction.commit()
  })
}

export const readIDB = async (name, store = storeDefault) => {
  const db = await initDB(store)

  const transaction = db.transaction([store], readWriteMode)
  const objStore = transaction.objectStore(store)

  return new Promise((resolve, reject) => {
    const request = objStore.get(name)
    request.onsuccess = event => resolve(event.target.result)
    request.onerror = event => reject(event)
    transaction.commit && transaction.commit()
  })
}

export const clearIDB = async (store = storeDefault) => {
  const db = await initDB(store)

  const transaction = db.transaction([store], readWriteMode)
  const objStore = transaction.objectStore(store)
  return new Promise((resolve, reject) => {
    const request = objStore.clear()
    request.onsuccess = event => resolve(event)
    request.onerror = event => reject(event)
    transaction.commit && transaction.commit()
  })
}

WebView 开启 WebRTC

见 WebView WebRTC not working

webView.setWebChromeClient(new WebChromeClient(){
	@TargetApi(Build.VERSION_CODES.LOLLIPOP)
	@Override
	public void onPermissionRequest(final PermissionRequest request) {
		request.grant(request.getResources());
	}
});

到此这篇关于HTML5录音实践总结(Preact)的文章就介绍到这了,更多相关html5录音内容请搜索易盾网络以前的文章或继续浏览下面的相关文章，希望大家以后多多支持易盾网络！