獲取 PCM 數(shù)據(jù)
處理 PCM 數(shù)據(jù)
Float32 轉(zhuǎn) Int16
ArrayBuffer 轉(zhuǎn) Base64
PCM 文件播放
重采樣
PCM 轉(zhuǎn) MP3
PCM 轉(zhuǎn) WAV
短時能量計算
Web Worker優(yōu)化性能
音頻存儲(IndexedDB)
WebView 開啟 WebRTC
獲取 PCM 數(shù)據(jù)
查看 DEMO
https://github.com/deepkolos/pc-pcm-wave
樣例代碼:
const mediaStream = await window.navigator.mediaDevices.getUserMedia({
audio: {
// sampleRate: 44100, // 采樣率 不生效需要手動重采樣
channelCount: 1, // 聲道
// echoCancellation: true,
// noiseSuppression: true, // 降噪 實測效果不錯
},
})
const audioContext = new window.AudioContext()
const inputSampleRate = audioContext.sampleRate
const mediaNode = audioContext.createMediaStreamSource(mediaStream)
if (!audioContext.createScriptProcessor) {
audioContext.createScriptProcessor = audioContext.createJavaScriptNode
}
// 創(chuàng)建一個jsNode
const jsNode = audioContext.createScriptProcessor(4096, 1, 1)
jsNode.connect(audioContext.destination)
jsNode.onaudioprocess = (e) => {
// e.inputBuffer.getChannelData(0) (left)
// 雙通道通過e.inputBuffer.getChannelData(1)獲取 (right)
}
mediaNode.connect(jsNode)
簡要流程如下:
start=>start: 開始
getUserMedia=>operation: 獲取MediaStream
audioContext=>operation: 創(chuàng)建AudioContext
scriptNode=>operation: 創(chuàng)建scriptNode并關(guān)聯(lián)AudioContext
onaudioprocess=>operation: 設(shè)置onaudioprocess并處理數(shù)據(jù)
end=>end: 結(jié)束
start->getUserMedia->audioContext->scriptNode->onaudioprocess->end
停止錄制只需要把 audioContext 掛在的 node 卸載即可,然后把存儲的每一幀數(shù)據(jù)合并即可產(chǎn)出 PCM 數(shù)據(jù)
jsNode.disconnect()
mediaNode.disconnect()
jsNode.onaudioprocess = null
PCM 數(shù)據(jù)處理
通過 WebRTC 獲取的 PCM 數(shù)據(jù)格式是 Float32 的, 如果是雙通道錄音的話, 還需要增加合并通道
const leftDataList = [];
const rightDataList = [];
function onAudioProcess(event) {
// 一幀的音頻PCM數(shù)據(jù)
let audioBuffer = event.inputBuffer;
leftDataList.push(audioBuffer.getChannelData(0).slice(0));
rightDataList.push(audioBuffer.getChannelData(1).slice(0));
}
// 交叉合并左右聲道的數(shù)據(jù)
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 轉(zhuǎn) Int16
const float32 = new Float32Array(1)
const int16 = Int16Array.from(
float32.map(x => (x > 0 ? x * 0x7fff : x * 0x8000)),
)
arrayBuffer 轉(zhuǎn) Base64
注意: 在瀏覽器上有個 btoa() 函數(shù)也是可以轉(zhuǎn)換為 Base64 但是輸入?yún)?shù)必須為字符串, 如果傳遞 buffer 參數(shù)會先被 toString() 然后再 Base64 , 使用 ffplay 播放反序列化的 Base64 , 會比較刺耳
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))
驗證 Base64 是否正確, 可以在 node 下把產(chǎn)出的 Base64 轉(zhuǎn)換為 Int16 的 PCM 文件, 然后使用 FFPlay 播放, 看看音頻是否正常播放
PCM 文件播放
# 單通道 采樣率:16000 Int16
ffplay -f s16le -ar 16k -ac 1 test.pcm
# 雙通道 采樣率:48000 Float32
ffplay -f f32le -ar 48000 -ac 2 test.pcm
重采樣/調(diào)整采樣率
雖然 getUserMedia 參數(shù)可設(shè)置采樣率, 但是在最新Chrome也不生效, 所以需要手動做個重采樣
const mediaStream = await window.navigator.mediaDevices.getUserMedia({
audio: {
// sampleRate: 44100, // 采樣率 設(shè)置不生效
channelCount: 1, // 聲道
// echoCancellation: true, // 減低回音
// noiseSuppression: true, // 降噪, 實測效果不錯
},
})
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每一幀的buffer合并后的數(shù)組
mergeArray(audioBuffers),
inputSampleRate,
outputSampleRate,
)
PCM 轉(zhuǎn) MP3
import { Mp3Encoder } from 'lamejs'
let mp3buf
const mp3Data = []
const sampleBlockSize = 576 * 10 // 工作緩存區(qū), 576的倍數(shù)
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)
// 工具函數(shù)
function float32ToInt16(audioBuffers, inputSampleRate, outputSampleRate) {
const float32 = resample(
// 需要onAudioProcess每一幀的buffer合并后的數(shù)組
mergeArray(audioBuffers),
inputSampleRate,
outputSampleRate,
)
const int16 = Int16Array.from(
float32.map(x => (x > 0 ? x * 0x7fff : x * 0x8000)),
)
return int16
}
使用 lamejs 即可, 但是體積較大(160+KB), 如果沒有存儲需求可使用 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 轉(zhuǎn) 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)
// 需要用一個view來操控buffer
const view = new DataView(buffer)
// 寫入wav頭部信息
// 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)
// 寫入PCM數(shù)據(jù)
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每一幀的buffer合并后的數(shù)組
createWavBuffer(mergeArray(audioBuffers))
WAV 基本上是 PCM 加上一些音頻信息
簡單的短時能量計算
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
}
由于計算結(jié)果有會因設(shè)備的錄音增益差異較大, 計算出數(shù)據(jù)也較大, 所以使用比值簡單區(qū)分人聲和噪音
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 優(yōu)化性能
音頻數(shù)據(jù)數(shù)據(jù)量較大, 所以可以使用 Web Worker 進行優(yōu)化, 不卡 UI 線程
在 Webpack 項目里 Web Worker 比較簡單, 安裝 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)
// 轉(zhuǎn)MP3/轉(zhuǎn)Base64/轉(zhuǎn)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()
// 簡單使用, 項目可以在recorder實例化的時候創(chuàng)建worker實例, 有并法需求可多個實例
return new Promise(resolve => {
worker.postMessage({
audioBuffers: audioBuffers,
inputSampleRate: inputSampleRate,
outputSampleRate: outputSampleRate,
type: 'mp3',
})
worker.onmessage = event => resolve(event.data)
})
}
音頻的存儲
瀏覽器持久化儲存的地方有 LocalStorage 和 IndexedDB , 其中 LocalStorage 較為常用, 但是只能儲存字符串, 而 IndexedDB 可直接儲存 Blob , 所以優(yōu)先選擇 IndexedDB ,使用 LocalStorage 則需要轉(zhuǎn) Base64 體積將會更大
所以為了避免占用用戶太多空間, 所以選擇MP3格式進行存儲
> 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 簡單封裝如下, 熟悉后臺的同學(xué)可以找個 ORM 庫方便數(shù)據(jù)讀寫
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 開啟 WebRTC
見 WebView WebRTC not working
webView.setWebChromeClient(new WebChromeClient(){
@TargetApi(Build.VERSION_CODES.LOLLIPOP)
@Override
public void onPermissionRequest(final PermissionRequest request) {
request.grant(request.getResources());
}
});
到此這篇關(guān)于HTML5錄音實踐總結(jié)(Preact)的文章就介紹到這了,更多相關(guān)html5錄音內(nèi)容請搜索腳本之家以前的文章或繼續(xù)瀏覽下面的相關(guān)文章，希望大家以后多多支持腳本之家！
