-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathindex.html
479 lines (448 loc) · 24.7 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
476
477
478
479
<!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>Multi-modal Multi-channel Speech Separation</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>Multi-modal Multi-channel Speech Separation</h2>
<p>Authors: Rongzhi Gu, Shi-Xiong Zhang, Lianwu Chen, Yong Xu, Meng Yu, Yuexian Zou, Dong Yu</p>
</header>
<p>Abstract:
Target speech separation refers to extracting a target speaker's voice from overlapped audios of simultaneous talkers.
Previously the use of visual clue for target speech separation has shown great potentials.
This paper extends it with a general multi-modal framework in the far-field scenario.
The framework leverages all the available information of target speaker, including his/her spatial location, voice characteristics and lip movements.
These target-related features are extracted and learned together with the separation task in an end-to-end neural architecture.
An important aspect of the multi-modal joint modeling is the form of fusion.
A factorized attention-based fusion method is proposed to aggregate different modalities.
To validate the robustness of proposed multi-modal system in practice, we evaluate it under different challenging conditions, such as one of the modalities is temporarily missing, invalid or noisy.
Experiments are carried out on a large-scale audio-visual dataset collected from YouTube and simulated into multichannels.
Evaluation results illustrate that our proposed multi-modal framework significantly outperforms single-modal speech separation approaches, even when the target's location or lip information is temporally missing or corrupted.
</p>
<!--
<ul class="actions">
<li><a href="#" class="button big">Learn More</a></li>
</ul>-->
</div>
<span class="image object" style="background-size: contain; background-repeat: no-repeat">
<img src="images/diagram-20191008.png" style="background-repeat:no-repeat; background-size:contain" width="50%" alt=""/>
</span>
</section>
<!-- Section -->
<section id="details">
<header class="major">
<h2>Details</h2>
</header>
<div class="features">
<article>
<div class="content">
<h3>Dataset</h3>
<p>The audio-visual corpus used for experiments is collected from Youtube, in which Mandarin accounts for the vast majority.</p>
<p>The simulated dataset contains 160,000, 15,000 and 1,200 multi-channel noisy and reverberant mixtures for
training, validation and testing. The speakers in the training set and test set are not overlapped.</p>
</div>
</article>
<article>
<div class="content">
<h3>Simulation</h3>
<p>We use a 9-element non-uniform linear array, with spacing 4-3-2-1-1-2-3-4 cm.
The Multi-channel audio signals are generated by convolving single-channel signals with RIRs simulated by image-source method (ISM).</p>
<p>We consider three scenarios for the synthetic examples generation: 1 speaker, 2 speakers and 3 speakers,
respectively accounts for 49%, 30% and 21% in the test dataset.</p>
</div>
</article>
<article>
<div class="content">
<h3>Framework</h3>
<p>As shown in figure above, the proposed system is a multi-stream architecture which takes four inputs:
(i) noisy multi-channel mixture waveforms,
(ii) target speaker's direction calculated by face detection,
(iii) video frames of cropped lip regions,
(iv) enrollment audio(s) of the target speaker.
The system directly outputs estimated monaural target speech,
while all other interfering signals are suppressed. </p>
</div>
</article>
<article>
<div class="content">
<h3>Features</h3>
<p>Audio: 1) Speaker-independent features: spectral feature (logrithm power spectrum) and spatial features (inter-channel phase difference).
2) Speaker-related feature: directional feature computed with the target speaker's direction. </p>
<p>Video: The video stream takes gray image frames of the target speaker's lip region. The structure of lip reading network is similar to the one proposed by [1].
It consists of a spatio-temporal convolution layer and a 18-layer ResNet, followed by several video blocks. The output of the video blocks are target speaker's lip embeddings. </p>
<p>Speaker Embedding: The speaker model was pretrained on speaker verification task. The input to the speaker model is an enrollment utterance of the target speaker.
The speaker model outputs the utterance-level speaker embedding. </p>
</div>
</article>
</div>
</section>
<!-- Section -->
<section id="spk1">
<header class="major">
<h2>1-speaker samples</h2>
</header>
<div class="posts">
<article>
<table cellspacing="50" style="width: 50%">
<thead>
<tr>
<th><center>Mixture (1st channel)</center> </th>
<th><center>reverberant clean</center> </th>
<th><center>audio-only</center> </th>
<th><center>audio-visual</center> </th>
<th><center>multi-modal</center> </th>
</tr>
<tr style="font-size: small">
<th> </th>
<th> </th>
<th><center>(IPD+DF)</center></th>
<th><center>(IPD+DF+lip)</center></th>
<th><center>(IPD+DF+lip+spk. emb.)</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="audio/1spk/mix/51-12_1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/1spk/clean/51-12_11.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/1spk/audioonly/51-12__21.00.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/1spk/audiovisual/51-12__22.59.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/1spk/multimodal/51-12__23.00.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/1spk/mix/45-0_1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/1spk/clean/45-0_11.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/1spk/audioonly/45-0__24.22.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/1spk/audiovisual/45-0__24.76.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/1spk/multimodal/45-0__25.55.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/1spk/mix/16-19_1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/1spk/clean/16-19_11.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/1spk/audioonly/16-19__19.40.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/1spk/audiovisual/16-19__20.71.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/1spk/multimodal/16-19__20.89.wav" ></audio></td>
</tr>
</tbody>
</table>
<!--<ul class="actions">
<li><a href="#" class="button">More</a></li>
</ul>-->
</article>
</div>
</section>
<!-- Section -->
<section id="spk2">
<header class="major">
<h2>2-speaker samples</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>Mixture (1st channel)</center> </th>
<th><center>reverberant clean</center> </th>
<th><center>audio-only</center> </th>
<th><center>audio-visual</center> </th>
<th><center>multi-modal</center> </th>
</tr>
<tr style="font-size: small">
<th> </th>
<th> </th>
<th><center>(IPD+DF)</center></th>
<th><center>(IPD+DF+lip)</center></th>
<th><center>(IPD+DF+lip+spk. emb.)</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="audio/2spk/mix/72-12_1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/clean/72-12_12.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/audioonly/72-12__11.56.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/audiovisual/72-12__12.36.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/multimodal/72-12__13.40.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/mix/96-25_1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/clean/96-25_12.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/audioonly/96-25__-6.29.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/audiovisual/96-25__0.97.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/multimodal/96-25__4.52.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/mix/98-14_1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/clean/98-14_12.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/audioonly/98-14__13.37.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/audiovisual/98-14__13.59.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/multimodal/98-14__15.59.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/mix/101-11_1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/clean/101-11_12.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/audioonly/101-11__6.38.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/audiovisual/101-11__6.38.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/multimodal/101-11__11.29.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/mix/99-4_1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/clean/99-4_12.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/audioonly/99-4__10.79.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/audiovisual/99-4__13.22.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/multimodal/99-4__14.32.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/mix/117-4_1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/clean/117-4_12.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/audioonly/117-4__6.29.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/audiovisual/117-4__8.66.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/2spk/multimodal/117-4__13.32.wav" ></audio></td>
</tr>
</tbody>
</table>
<!--<ul class="actions">
<li><a href="#" class="button">More</a></li>
</ul>-->
</article>
</div>
</section>
<!-- Section -->
<section id="spk3">
<header class="major">
<h2>3-speaker samples</h2>
</header>
<div class="posts">
<article>
<table cellspacing="50" style="width: 50%">
<thead>
<tr>
<th><center>Mixture (1st channel)</center> </th>
<th><center>reverberant clean</center> </th>
<th><center>audio-only</center> </th>
<th><center>audio-visual</center> </th>
<th><center>multi-modal</center> </th>
</tr>
<tr style="font-size: small">
<th> </th>
<th> </th>
<th><center>(IPD+DF)</center></th>
<th><center>(IPD+DF+lip)</center></th>
<th><center>(IPD+DF+lip+spk. emb.)</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="audio/3spk/mix/71-11_1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/clean/71-11_13.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/audioonly/71-11__13.31.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/audiovisual/71-11__14.07.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/multimodal/71-11__14.59.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/mix/76-11_1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/clean/76-11_13.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/audioonly/76-11__9.19.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/audiovisual/76-11__9.50.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/multimodal/76-11__10.12.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/mix/31-1_1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/clean/31-1_13.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/audioonly/31-1__8.27.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/audiovisual/31-1__11.06.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/multimodal/31-1__11.44.wav" ></audio></td>
</tr>
<tr style="border-top:1px solid black">
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/mix/80-25_1.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/clean/80-25_13.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/audioonly/80-25__-5.42.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/audiovisual/80-25__-5.23.wav" ></audio></td>
<td><audio controls class="audio-player" preload="metadata" style="width:180px;">
<source src="audio/3spk/multimodal/80-25__4.98.wav" ></audio></td>
</tr>
</tbody>
</table>
<!--<ul class="actions">
<li><a href="#" class="button">More</a></li>
</ul>-->
</article>
</div>
</section>
<!-- Section -->
<section id="wild">
<header class="major">
<h2>Real recorded samples</h2>
</header>
<div class="posts">
<article>
<table cellspacing="50" style="width: 50%">
<thead>
<tr>
<th> Unprocessed Mixture</th>
<th> Estimated speaker 1</th>
<th> Estimated speaker 2</th>
</tr>
</thead>
<tbody>
<tr style="border-top:1px solid black">
<td><video id="mixture_video" width="380" controls>
<source src="video/mixture.mp4" ></video></td>
<td><video id="spk_left_video" width="380" controls>
<source src="video/spk_left.mp4" ></video></td>
<td><video id="spk_right_video" width="380" controls>
<source src="video/spk_right.mp4" ></video></td>
</tr>
<tr style="border-top:1px solid black">
<td><img id="mixture_spec" alt="" width="380" src="video/mixture_spec.png"></td>
<td><img id="spk_left_spec" alt="" width="380" src="video/spk_left_spec.png"></td>
<td><img id="spk_right_spec" alt="" width="380" src="video/spk_right_spec.png"></td>
</tr>
</tbody>
</table>
<!--<ul class="actions">
<li><a href="#" class="button">More</a></li>
</ul>-->
</article>
</div>
</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="#details">Details</a></li>
<li>
<span class="opener">Audio-visual samples</span>
<ul>
<li><a href="#spk1">1-speaker samples</a></li>
<li><a href="#spk2">2-speaker samples</a></li>
<li><a href="#spk3">3-speaker samples</a></li>
<li><a href="#wild">real recorded samples</a></li>
</ul>
</li>
</ul>
</nav>
<!-- Section -->
<section>
<header class="major">
<h2>Related works</h2>
</header>
<p>[1] Afouras Triantafyllos, Joon Son Chung, and Andrew Zisserman.
"The Conversation: Deep Audio-Visual Speech Enhancement." Proc. Interspeech 2018 (2018): 3244-3248.</p>
<p>[2] Wu Jian, et al. "Time Domain Audio Visual Speech Separation." arXiv preprint arXiv:1904.03760 (2019).
<strong><a href="https://funcwj.github.io/online-demo/page/tavs/">Demos</a></strong></p>
<p>[3] Tan Ke, et al. "Audio-Visual Speech Separation and Dereverberation with a Two-Stage Multimodal Network." arXiv preprint arXiv:1909.07352 (2019).
<strong><a href="https://jupiterethan.github.io/av-enh.github.io/">Demos</a></strong></p>
</section>
<footer id="footer">
<p class="copyright">
<center>© Oct. 2019 </center>
<img src="images/pku_icon2.png" height="50px" alt=""> <img src="images/tencent_icon.png" height="50px" 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>