forked from Moplast/moplast.github.io
-
Notifications
You must be signed in to change notification settings - Fork 2
/
index.html
475 lines (454 loc) · 24.9 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
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
<!DOCTYPE HTML>
<!--
Editorial by HTML5 UP
html5up.net | @ajlkn
Free for personal and commercial use under the CCA 3.0 license (html5up.net/license)
-->
<html>
<head>
<title>Ultra Dual-Path Compression For Joint Echo Cancellation and Noise Suppression</title>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1, user-scalable=no" />
<link rel="stylesheet" href="assets/css/main.css" />
</head>
<body class="is-preload">
<!-- Wrapper -->
<div id="wrapper">
<!-- Main -->
<div id="main">
<div class="inner">
<!-- Banner -->
<section id="banner">
<div class="content">
<header>
<h2>Ultra Dual-Path Compression For Joint Echo Cancellation and Noise Suppression</h2>
<p>Accepted by Interspeech 2023</p>
<p>Authors: Hangting Chen, Jianwei Yu, Yi Luo, Rongzhi Gu, Weihua Li, Zhuocheng Lu, Chao Weng</p>
<p>Tencent AI Lab, Audio and Speech Signal Processing Oteam</p>
<p>Email: erichtchen@tencent.com</p>
</header>
<p>Abstract:
Echo cancellation and noise reduction are essential for full-duplex communication, yet most existing neural networks have high computational costs and are inflexible in tuning model complexity.
In this paper, we introduce time-frequency dual-path compression to achieve a wide range of compression ratios on computational cost.
Specifically, for frequency compression, trainable filters are used to replace manually designed filters for dimension reduction.
For time compression, only using frame skipped prediction causes large performance degradation, which can be alleviated by a post-processing network with full sequence modeling.
We have found that under fixed compression ratios, dual-path compression combining both the time and frequency methods will give further performance improvement, covering compression ratios from 4x to 32x with little model size change.
Moreover, the proposed models show competitive performance compared with fast FullSubNet and DeepFilterNet.
</p>
<p><a href=https://github.com/tencent-ailab/UltraDualPathCompression>Core source code is available.</a></p>
<!--
<ul class="actions">
<li><a href="#" class="button big">Learn More</a></li>
</ul>-->
</div>
</section>
<section id="poster">
<h2>Poster</h2>
<tr style="border-top:1px solid black">
<td><img id="poster_fig" alt="" width="100%" src="images/poster.png"></td>
</tr>
</section>
<!-- Section -->
<section id="details">
<header class="major">
<h2>A Quick Review</h2>
</header>
<div class="features">
<article>
<div class="content">
<h3>Background</h3>
<p>Model performance relates to (1) architecture and (2) # params. & computational cost.
When the architecture is determined, the # params as well as the computational cost can be tuned by hyperparameters. </p>
<p>AEC&ANS for full-duplex communication needs low computational cost when deployed on PC/Phones.
Thus, an important research topic is to lower computational cost while lowering performance degradation.</p>
</div>
</article>
<article>
<div class="content">
<h3>Experimental setup</h3>
<p>Models are trained on a simulated dataset with 100K iterations. Though the model performance can be further improved with a larger iteration number, we adopt such an iteration number for fast comparison. </p>
<p>Models are tested on a semi-simulated dataset, where the echos are real-recorded ones from the AEC challenge.</p>
<p>Models have a window size and a hop size of 20ms and 10ms, respectively. Additional look ahead is not allowed due to causal processing.</p>
</div>
</article>
<article>
<div class="content">
<h3>Framework & Method</h3>
<p>The whole framework consists of a linear AEC (Kalman-filter based), an online DPT-FSNet, compression and decompression modules and a post network.</p>
<p>Compression and Decompression are operated on the input spectra and the output features, respectively.
In detail, the compression module compresses the input spectra first along the time axis and then the frequency using linear transform. The time compression uses the past and the current frames while the frequency compression is conducted with the Mel scale.
The decompression module decompresses the output feature first along frequency and then the time axis using linear transform. The time decompression predicts the current feature and copies it for future frames.
A post network (1-layer GRU) is needed to eliminate the degradation caused by the time compression.
</p>
</div>
</article>
<article>
<div class="content">
<h3>Results</h3>
<p>1. Time compression needs an ultra-light post-net.</p>
<p>2. Dual-path compression performs better than single-path compression.
For example, from 1.8G MACs/s to 140M MACs/s, the model achieves a 13x compression ratio but only has 0.22 PESQ drop, 0.16 and 0.14 PESQ higher than single-path compression.</p>
</div>
</article>
</div>
</section>
<section id="results_table">
<h2>Detailed results</h2>
<tr style="border-top:1px solid black">
<td><img id="results_fig" alt="" width="75%" src="images/results.png"></td>
</tr>
</section>
<!-- Section -->
<section id="dt">
<header class="major">
<h2>Double talk</h2>
</header>
<div class="posts">
<article>
<table cellspacing="50" style="width: 50%">
<thead>
<tr>
<th><center>MIC</center> </th>
<th><center>LPB(REF)</center> </th>
<th><center>Uncompressed model</center> </th>
<th><center>Dual-path</center> </th>
<th><center>Dual-path</center> </th>
</tr>
<tr style="font-size: small">
<th> </th>
<th> </th>
<th> </center></th>
<th><center>T2 x F2</center></th>
<th><center>T4 x F4</center></th>
</tr>
<tr style="font-size: small">
<th> </th>
<th> </th>
<th><center>MACs: 1.8G/s</center></th>
<th><center>MACs: 486M/s</center></th>
<th><center>MACs: 140M/s</center></th>
</tr>
</thead>
<tbody>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP15_SNRP99_0000001.wav.mix.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP15_SNRP99_0000001.wav.lpb.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP15_SNRP99_0000001.wav.est.1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP15_SNRP99_0000001.wav.est.4.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP15_SNRP99_0000001.wav.est.16.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP05_SNRP15_0000001.wav.mix.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP05_SNRP15_0000001.wav.lpb.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP05_SNRP15_0000001.wav.est.1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP05_SNRP15_0000001.wav.est.4.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP05_SNRP15_0000001.wav.est.16.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP05_SNRP99_0000001.wav.mix.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP05_SNRP99_0000001.wav.lpb.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP05_SNRP99_0000001.wav.est.1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP05_SNRP99_0000001.wav.est.4.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP05_SNRP99_0000001.wav.est.16.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN05_SNRP15_0000001.wav.mix.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN05_SNRP15_0000001.wav.lpb.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN05_SNRP15_0000001.wav.est.1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN05_SNRP15_0000001.wav.est.4.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN05_SNRP15_0000001.wav.est.16.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN05_SNRP99_0000001.wav.mix.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN05_SNRP99_0000001.wav.lpb.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN05_SNRP99_0000001.wav.est.1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN05_SNRP99_0000001.wav.est.4.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN05_SNRP99_0000001.wav.est.16.wav" ></audio></td>
</tr>
</tbody>
</table>
<!--<ul class="actions">
<li><a href="#" class="button">More</a></li>
</ul>-->
</article>
</div>
</section>
<!-- Section -->
<section id="nest">
<header class="major">
<h2>Near-end single talk</h2>
</header>
<div class="posts">
<article>
<!--a href="#" class="image"><img src="images/pic02.jpg" alt="" /></--a>-->
<table cellspacing="50" style="width: 50%">
<thead>
<tr>
<th><center>MIC</center> </th>
<th><center>Uncompressed model</center> </th>
<th><center>Dual-path</center> </th>
<th><center>Dual-path</center> </th>
</tr>
<tr style="font-size: small">
<th> </th>
<th> </center></th>
<th><center>T2 x F2</center></th>
<th><center>T4 x F4</center></th>
</tr>
<tr style="font-size: small">
<th> </th>
<th><center>MACs: 1.8G/s</center></th>
<th><center>MACs: 486M/s</center></th>
<th><center>MACs: 140M/s</center></th>
</tr>
</thead>
<tbody>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP05_0000000.wav.mix.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP05_0000000.wav.est.1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP05_0000000.wav.est.4.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP05_0000000.wav.est.16.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP05_0000001.wav.mix.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP05_0000001.wav.est.1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP05_0000001.wav.est.4.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP05_0000001.wav.est.16.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP15_0000000.wav.mix.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP15_0000000.wav.est.1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP15_0000000.wav.est.4.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP15_0000000.wav.est.16.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP15_0000001.wav.mix.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP15_0000001.wav.est.1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP15_0000001.wav.est.4.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRP99_SNRP15_0000001.wav.est.16.wav" ></audio></td>
</tr>
</tbody>
</table>
<!--<ul class="actions">
<li><a href="#" class="button">More</a></li>
</ul>-->
</article>
</div>
</section>
<!-- Section -->
<section id="fest">
<header class="major">
<h2>Far-end single talk</h2>
</header>
<div class="posts">
<article>
<table cellspacing="50" style="width: 50%">
<thead>
<tr>
<th><center>MIC</center> </th>
<th><center>LPB(REF)</center> </th>
<th><center>Uncompressed model</center> </th>
<th><center>Dual-path</center> </th>
<th><center>Dual-path</center> </th>
</tr>
<tr style="font-size: small">
<th> </th>
<th> </th>
<th> </center></th>
<th><center>T2 x F2</center></th>
<th><center>T4 x F4</center></th>
</tr>
<tr style="font-size: small">
<th> </th>
<th> </th>
<th><center>MACs: 1.8G/s</center></th>
<th><center>MACs: 486M/s</center></th>
<th><center>MACs: 140M/s</center></th>
</tr>
</thead>
<tbody>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000000.wav.mix.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000000.wav.lpb.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000000.wav.est.1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000000.wav.est.4.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000000.wav.est.16.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000001.wav.mix.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000001.wav.lpb.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000001.wav.est.1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000001.wav.est.4.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000001.wav.est.16.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000002.wav.mix.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000002.wav.lpb.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000002.wav.est.1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000002.wav.est.4.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="demos_audios/test_SIRN99_SNRP99_0000002.wav.est.16.wav" ></audio></td>
</tr>
</tbody>
</table>
<!--<ul class="actions">
<li><a href="#" class="button">More</a></li>
</ul>-->
</article>
</div>
</section>
<!-- Section -->
<section id="rtf_test">
<header class="major">
<h2>Real-time factor (RTF) test</h2>
</header>
<div class="posts">
<article>
<table cellspacing="50" style="width: 100%">
<thead>
<tr>
<th> RTF test on Intel(R) Core(TM) i7-9700 CPU @ 3.00GHz using a model with 13x compression rato (T4 x F4) </th>
</tr>
<tr>
<th> The test is carried with frame-by-frame process to mimic real-world online processing.</th>
</tr>
</thead>
<tbody>
<tr style="border-top:1px solid black">
<td><video id="rtf_test_video" width="150%" controls>
<source src="video/rtf_test.webm" ></video></td>
</tr>
</tbody>
</table>
<!--<ul class="actions">
<li><a href="#" class="button">More</a></li>
</ul>-->
</article>
</div>
</section>
<!-- Section -->
<section>
<header class="major">
<h2>Related works</h2>
</header>
<p>[1] F. Dang, H. Chen, and P. Zhang,
“Dpt-fsnet: Dual-path transformer based full-band and sub-band fusion network for speech enhancement,”
in 2022 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2022. IEEE, 202</p>
<p>[2] S. Zhang, Z. Wang, J. Sun, Y. Fu, B. Tian, Q. Fu, and L. Xie,
“Multi-task deep residual echo suppression with echo-aware loss,”
in IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2022, Virtual and Singapore, 23-27 May 2022. IEEE, 2022, pp. 9127–9
</p>
<p>[3] X. Hao and X. Li,
“Fast fullsubnet: Accelerate full-band and subband fusion model for single-channel speech enhancement,” arXiv preprint arXiv:2212.09019, 2022
</p>
</section>
</div>
</div>
<!-- Sidebar -->
<div id="sidebar">
<div class="inner">
<!-- Search
<section id="search" class="alt">
<form method="post" action="#">
<input type="text" name="query" id="query" placeholder="Search" />
</form>
</section>
-->
<!-- Menu -->
<nav id="menu">
<header class="major">
<h2>Menu</h2>
</header>
<ul>
<li><a href="#banner">Abstract</a></li>
<li><a href="#poster">Poster</a></li>
<li><a href="#details">A Quick Review</a></li>
<li><a href="#results_table">Detailed results</a></li>
<li>
<span class="opener">Audio samples</span>
<ul>
<li><a href="#dt">Double talk</a></li>
<li><a href="#nest">Near-end single talk</a></li>
<li><a href="#fest">Far-end single talk</a></li>
</ul>
</li>
<li><a href="#rtf_test">Real-time factor (RTF) test</a></li>
</ul>
</nav>
<footer id="footer">
<p>
<center>Created by Hangting Chen </center>
</p>
<p>
<center>Augest 13th, 2023</center>
</p>
<p class="copyright" style="display: flex; flex-direction: column; align-items: center; justify-content: center;">
<img src="images/tencent_icon.png" width="50%" alt="">
</p>
</footer>
</div>
</div>
</div>
<!-- Scripts -->
<script src="assets/js/jquery.min.js"></script>
<script src="assets/js/browser.min.js"></script>
<script src="assets/js/breakpoints.min.js"></script>
<script src="assets/js/util.js"></script>
<script src="assets/js/main.js"></script>
</body>
</html>