-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
444 lines (405 loc) · 15.6 KB
/
index.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>Document</title>
<script src="tf.min.js"></script>
<script src="tf-backend-webgpu.js"></script>
</head>
<body>
<h1>本DEMO 可以将人声移除, 上传文件后请耐心等待，直到进度条完成</h1>
<input type="file" id="file" />
<progress id="progress" value="0" max="100"></progress
><span id="tips"></span>
<script>
// Function to convert AudioBuffer to WAV Blob
function bufferToWavBlob(buffer) {
const numberOfChannels = 2;
const sampleRate = 44100 / 2;
const length = buffer[0].length;
const arrayBuffer = new ArrayBuffer(44 + length * 2 * numberOfChannels);
const view = new DataView(arrayBuffer);
// RIFF identifier
writeString(view, 0, "RIFF");
// RIFF chunk length
view.setUint32(4, 36 + length * 2, true);
// RIFF type (WAVE)
writeString(view, 8, "WAVE");
// format chunk identifier
writeString(view, 12, "fmt ");
// format chunk length
view.setUint32(16, 16, true);
// sample format (PCM)
view.setUint16(20, 1, true);
// channel count
view.setUint16(22, numberOfChannels, true);
// sample rate
view.setUint32(24, sampleRate, true);
// byte rate (sample rate * block align)
view.setUint32(28, sampleRate * 2 * numberOfChannels, true);
// block align (channel count * bytes per sample)
view.setUint16(32, numberOfChannels * 2, true);
// bits per sample
view.setUint16(34, 16, true);
// data chunk identifier
writeString(view, 36, "data");
// data chunk length
view.setUint32(40, length * 2, true);
// console.log("view", buffer, buffer?.getChannelData(0));
// 处理每个通道的数据
for (let channel = 0; channel < numberOfChannels; channel++) {
floatTo16BitPCM(view, 44 + channel * length * 2, buffer[channel]);
}
return new Blob([view], { type: "audio/wav" });
}
function writeString(view, offset, string) {
for (let i = 0; i < string.length; i++) {
view.setUint8(offset + i, string.charCodeAt(i));
}
}
function floatTo16BitPCM(output, offset, input) {
for (let i = 0; i < input.length; i++, offset += 2) {
const s = Math.max(-1, Math.min(1, input[i]));
output.setInt16(offset, s < 0 ? s * 0x8000 : s * 0x7fff, true);
}
}
async function istft(x) {
const n = 6144 / 2 + 1;
const f_pad = tf.zeros([1, 4, 1, 32]);
let x1 = tf.concat([x, f_pad], x.shape.length - 2);
f_pad.dispose();
let x2 = x1.reshape([1, 2, 2, 3073, 32]);
x1.dispose();
let x3 = x2.reshape([-1, 2, n, 32]);
x2.dispose();
let x4 = x3.transpose([0, 2, 3, 1]);
x3.dispose();
const splitTensors = tf.split(x4, 2, 0);
x4.dispose();
let l = splitTensors[0];
let l2 = tf.squeeze(l, [0]);
let r = splitTensors[1];
let r2 = tf.squeeze(r, [0]);
let result = await getIstftLR(l2, r2, 6144, 1024, createHannWindow);
l.dispose();
l2.dispose();
r.dispose();
r2.dispose();
return result;
}
function getStft(input) {
return tf.tidy(() => {
let res = tf.signal.stft(input, 6144, 1024, 6144, createHannWindow);
let real = tf.real(res);
let imag = tf.imag(res);
real = tf.slice(real, [0, 0], [32, 3072]);
imag = tf.slice(imag, [0, 0], [32, 3072]);
real = real.transpose();
imag = imag.transpose();
let output = tf.stack([real, imag], 0);
let zeroLike = tf.zerosLike(output);
let isNaN = tf.isNaN(output);
return tf.where(isNaN, zeroLike, output);
});
}
async function getIstftLR(
spectrogramL,
spectrogramR,
fftSize,
hopSize,
windowFn
) {
let tasks = [];
let needsToDispose = [];
const numFrames = spectrogramL.shape[1];
let size = fftSize + hopSize * (numFrames - 1);
let outputArrayL = new Float32Array(size);
let outputArrayR = new Float32Array(size);
for (let g = 0; g < 2; g++) {
let spectrogram;
if (g == 0) {
spectrogram = spectrogramL;
} else {
spectrogram = spectrogramR;
}
const numFrames = spectrogram.shape[1];
const numFrequencyBins = spectrogram.shape[0];
let window = windowFn ? windowFn(fftSize) : tf.ones([fftSize]);
for (let i = 0; i < numFrames; i++) {
let frame = spectrogram.slice([0, i], [numFrequencyBins, 1]);
let frame2 = tf.squeeze(frame);
let realPart = frame2.slice([0, 0], [-1, 1]);
let imagPart = frame2.slice([0, 1], [-1, 1]);
let complexTensor = tf.complex(realPart, imagPart);
let complexTensor2 = tf.squeeze(complexTensor);
let timeDomainFrame = tf.spectral.irfft(complexTensor2);
let timeDomainFrame2 = tf.mul(timeDomainFrame, window);
frame.dispose();
frame2.dispose();
realPart.dispose();
imagPart.dispose();
complexTensor.dispose();
complexTensor2.dispose();
timeDomainFrame.dispose();
needsToDispose.push(timeDomainFrame2);
tasks.push(timeDomainFrame2.data());
}
window.dispose();
}
let completedTasks = await Promise.all(tasks);
for (const iterator of needsToDispose) {
iterator.dispose();
}
let framesL = completedTasks.splice(0, numFrames);
let framesR = completedTasks;
for (let i = 0; i < numFrames; i++) {
const timeDomainArray = framesL[i];
const startIdx = i * hopSize;
for (let j = 0; j < fftSize; j++) {
if (startIdx + j < outputArrayL.length) {
outputArrayL[startIdx + j] += timeDomainArray[j];
}
}
}
for (let i = 0; i < numFrames; i++) {
const timeDomainArray = framesR[i];
const startIdx = i * hopSize;
for (let j = 0; j < fftSize; j++) {
if (startIdx + j < outputArrayR.length) {
outputArrayR[startIdx + j] += timeDomainArray[j];
}
}
}
return [outputArrayL, outputArrayR];
}
function createHannWindow(length) {
let ts = tf.tensor(tf.signal.hannWindow(length).dataSync());
return ts;
}
// Create an AudioContext instance
const audioContext = new (window.AudioContext ||
window.webkitAudioContext)({
sampleRate: 44100,
channelCount: 1,
echoCancellation: false,
autoGainControl: false,
noiseSuppression: false,
});
let writtenPointer = 0;
document.getElementById("file").addEventListener("change", function (e) {
const file = e.target.files[0];
const reader = new FileReader();
reader.onload = function (e) {
const audioData = e.target.result;
handler(audioData);
};
reader.readAsArrayBuffer(file);
});
// Load the audio file using an XMLHttpRequest
// const xhr = new XMLHttpRequest();
// xhr.open("GET", "./test.mp3", true);
// xhr.responseType = "arraybuffer";
const handler = function a(audioData) {
// // Decode the audio data
audioContext.decodeAudioData(audioData, async function (buffer) {
await tf.setBackend("webgpu");
const { numberOfChannels, sampleRate } = buffer;
// offlineAudioContext相对AudioContext更加节省资源
const offlineAudioContext = new OfflineAudioContext(
numberOfChannels,
buffer.length,
sampleRate
);
var source = offlineAudioContext.createBufferSource();
source.buffer = buffer;
source.connect(offlineAudioContext.destination);
source.start();
// 处理不同声道的音频信息 1.如果是单声道， 需要克隆一份声道数据 2. 如果是双声道，则直接使用
const channelData =
numberOfChannels === 1
? [
Array.from(buffer.getChannelData(0)),
Array.from(buffer.getChannelData(0)),
]
: [
Array.from(buffer.getChannelData(0)),
Array.from(buffer.getChannelData(1)),
];
const model = await tf.loadGraphModel("./models/model.json");
const chunkSize = 31744;
// 根据chunkSize进行分块处理 channelData
const chunkedChannelData = [];
for (let i = 0; i < channelData[0].length; i += chunkSize) {
const chunk = [
channelData[0].slice(i, i + chunkSize),
channelData[1].slice(i, i + chunkSize),
];
chunkedChannelData.push(chunk);
}
console.log("buffer0", chunkedChannelData);
const writeBuffers0 = [];
const writeBuffers1 = [];
const writeVocalBuffers0 = [];
const writeVocalBuffers1 = [];
let progress = 0;
for await (let chunk of chunkedChannelData) {
const $channelData = chunk;
const runner = async () => {
tf.engine().startScope();
const [buffer0, buffer1] = $channelData;
if (buffer0.length < 31744) {
return;
}
let totalBufferPointer = buffer0.length;
let currentLast = totalBufferPointer;
let originalLeft = buffer0.slice(
buffer0.length - 31744,
buffer0.length
);
let originalRight = buffer1.slice(
buffer1.length - 31744,
buffer1.length
);
let left = new Array(3072)
.fill(0)
.concat(originalLeft)
.concat(new Array(3072).fill(0));
let right;
if (buffer1.length >= 31744) {
right = new Array(3072)
.fill(0)
.concat(originalRight)
.concat(new Array(3072).fill(0));
} else {
right = left.concat();
}
buffer0.splice(0, buffer0.length - 31744 - 1);
buffer1.splice(0, buffer1.length - 31744 - 1);
let inputL = tf.tensor1d(left);
let inputR = tf.tensor1d(right);
let l = getStft(inputL);
let r = getStft(inputR);
inputL.dispose();
inputR.dispose();
let t0 = tf.stack([l, r], 3);
l.dispose();
r.dispose();
let t1 = t0.transpose([0, 3, 1, 2]);
t0.dispose();
let reshaped = t1.reshape([
1,
2,
2,
-1,
t1.shape[t1.shape.length - 1],
]);
let reshaped2 = reshaped.reshape([
1,
4,
-1,
t1.shape[t1.shape.length - 1],
]);
t1.dispose();
reshaped.dispose();
let data = tf.tidy(() => {
return model.predict(reshaped2);
});
let vocaldata = tf.tidy(() => {
return tf.sub(reshaped2, data);
});
reshaped2.dispose();
let signals = await istft(data);
let signals2 = await istft(vocaldata);
tf.engine().endScope();
let elapse = currentLast - writtenPointer;
let outdata0 = Array.from(signals[0]);
outdata0 = outdata0.splice(3072, 31744);
let outdata1 = Array.from(signals[1]);
outdata1 = outdata1.splice(3072, 31744);
let writeBuffer0 = [];
let writeBuffer1 = [];
if (elapse < outdata0.length) {
writeBuffer0 = writeBuffer0.concat(
outdata0.slice(outdata0.length - elapse, outdata0.length)
);
} else {
writeBuffer0 = writeBuffer0.concat(outdata0);
}
if (elapse < outdata0.length) {
writeBuffer1 = writeBuffer1.concat(
outdata1.slice(outdata1.length - elapse, outdata1.length)
);
} else {
writeBuffer1 = writeBuffer1.concat(outdata1);
}
writeBuffers0.push(...writeBuffer0);
writeBuffers1.push(...writeBuffer1);
outdata0 = Array.from(signals2[0]);
outdata0 = outdata0.splice(3072, 31744);
outdata1 = Array.from(signals2[1]);
outdata1 = outdata1.splice(3072, 31744);
writeBuffer0 = [];
writeBuffer1 = [];
if (elapse < outdata0.length) {
writeBuffer0 = writeBuffer0.concat(
outdata0.slice(outdata0.length - elapse, outdata0.length)
);
} else {
writeBuffer0 = writeBuffer0.concat(outdata0);
}
if (elapse < outdata0.length) {
writeBuffer1 = writeBuffer1.concat(
outdata1.slice(outdata1.length - elapse, outdata1.length)
);
} else {
writeBuffer1 = writeBuffer1.concat(outdata1);
}
writeVocalBuffers0.push(...writeBuffer0);
writeVocalBuffers1.push(...writeBuffer1);
tf.dispose(vocaldata);
tf.dispose(data);
};
await runner();
progress += 1;
document.getElementById("progress").value = Math.floor(
(progress * 100) / chunkedChannelData.length
);
document.getElementById("tips").innerText = `进度${Math.floor(
(progress * 100) / chunkedChannelData.length
)}%`;
// console.log("progress", (progress += 1), chunkedChannelData.length);
}
// 测试时解开
const blob = bufferToWavBlob([writeBuffers0, writeBuffers1]);
// const blob = bufferToWavBlob(buffer);
// Create a download link
const a = document.createElement("a");
document.body.appendChild(a);
a.style = "display: none";
// Create a URL for the Blob and set it as the href attribute
const url = URL.createObjectURL(blob);
a.href = url;
a.download = "background-music.wav";
const vblob = bufferToWavBlob([writeVocalBuffers0, writeVocalBuffers1]);
// const blob = bufferToWavBlob(buffer);
// Create a download link
const va = document.createElement("a");
document.body.appendChild(va);
va.style = "display: none";
// Create a URL for the Blob and set it as the href attribute
const vurl = URL.createObjectURL(vblob);
va.href = vurl;
va.download = "vocal-music.wav";
// Trigger a click on the link to start the download
a.click();
va.click();
// Clean up the URL object
window.URL.revokeObjectURL(url);
window.URL.revokeObjectURL(vurl);
});
};
xhr.send();
</script>
</body>
</html>