Repository navigation
Expand file tree
/
Copy pathdemos.html
More file actions
586 lines (568 loc) · 28.3 KB
/
Copy pathdemos.html
File metadata and controls
586 lines (568 loc) · 28.3 KB
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
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
---
layout: page
---
<!--<h6 class="mx-auto text-center" style="color: orangered">PAGE UPDATE IN PROGRESS</h6>-->
<p class="lead mb-0" align="justify">
Dear reader, the YouTube videos will start in the segment of interest. The only way to control the video is through
the "Play" and "Stop" buttons that we have provided. The recommended way of visualizing the demos is
to pay attention to the target singer's lip movements and then to listen to the estimated samples.
</br>
<b>Note: </b>Please be aware that, sometimes, the perceivable differences in the singing voice estimates
across different models are subtle.
We <b>strongly recommend the use of headphones</b> while listening to them. We also would like to draw attention to the
spectrograms for visual cues that might better help perceive these differences.
</br>
</p></br>
<h1>Examples for Singing Voice Separation Application</h1>
<p class="lead mb-0" align="justify">
Here, we show some demos of how our models perform in estimating the desired target voice
in real world videos (taken from YouTube).
Note that none of these singers were a part of the training set.
</p>
</br>
<div class="col-md-8" style="padding-left:0;padding-bottom: 1em; display: flex">
<div class="column" style="text-align:center; padding-right: 1em">
<h4>Love of My Life (Queen)</h4>
</div>
<div class="column" style="text-align:center;">
<div class="spl_videoplay2" videoid="#q4">
<button type="button" class="btn btn-outline-primary">Play</button>
</div>
</div>
<div class="column">
<div class="col-sm-3 spl_videostop" videoid="#q4">
<button type="button" class="btn btn-outline-primary">Stop</button>
</div>
</div>
</div>
<div id="spl"
class="spl_videoplay" videoid="#q4"
onclick="this.nextElementSibling.style.display='block'; this.style.display='none';">
<div class="img-container" style="position: relative">
<img src="../custom_thumbnail.png" style=" position:relative;width:100%; height:345px;padding-bottom:0.4em;cursor:pointer"/>
<button class="ytp-large-play-button ytp-button" aria-label="Play" style="position: absolute; top: 50%; left: 50%;transform: translate(-50%, -50%);-ms-transform: translate(-50%, -50%);">
</button>
<svg height="100%" version="1.1" viewBox="0 0 68 48" width="8%" style="position: absolute; top: 50%; left: 50%;transform: translate(-50%, -50%);-ms-transform: translate(-50%, -50%);cursor: pointer"><path class="ytp-large-play-button-bg" d="M66.52,7.74c-0.78-2.93-2.49-5.41-5.42-6.19C55.79,.13,34,0,34,0S12.21,.13,6.9,1.55 C3.97,2.33,2.27,4.81,1.48,7.74C0.06,13.05,0,24,0,24s0.06,10.95,1.48,16.26c0.78,2.93,2.49,5.41,5.42,6.19 C12.21,47.87,34,48,34,48s21.79-0.13,27.1-1.55c2.93-0.78,4.64-3.26,5.42-6.19C67.94,34.95,68,24,68,24S67.94,13.05,66.52,7.74z" fill="#f00"></path><path d="M 45,24 27,14 27,34" fill="#fff"></path></svg>
</div>
</div>
<div style="display:none">
<iframe id="q4" class="q4_video"
width="100%" height="345"
src="https://www.youtube.com/embed/sEnTMgzw8ow?start=164&end=170&autoplay=1&showinfo=0&controls=0&enablejsapi=1&disablekb=1"
title="Queen Medley 4" frameborder="0"
allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture"
allowfullscreen></iframe>
</br>
</div>
<div class="row">
<div class="border">
<div class="col">
<h5>Target Voice: Lead Singer (top-right)</h5>
</div>
<div style="display: flex">
<div class="column" style="text-align:center;display: flex">
<div style="padding-top: 0.5em;display: flex">
<div class="col-xs-1 excol3">
<div class="src4_model">Mixture</div>
<img class="spec"
src="../estimated_demo/sEnTMgzw8ow_2_38_to_2_53/mix_mod.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/sEnTMgzw8ow_2_38_to_2_53/mix_mod.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;px">Y-Net-mr [1]</div>
<img class="spec"
src="../estimated_demo/sEnTMgzw8ow_2_38_to_2_53/y-net-mr7_mod.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/sEnTMgzw8ow_2_38_to_2_53/y-net-mr7_mod.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;px">Y-Net-mr-V (Ours)</div>
<img class="spec"
src="../estimated_demo/sEnTMgzw8ow_2_38_to_2_53/vocalist_mod.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/sEnTMgzw8ow_2_38_to_2_53/vocalist_mod.mp3"
controls=""></audio>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
<br>
<p class="lead mb-0" align="justify">
<h6 style="font-size: 1.25rem">Observation:</h6>
<span style="font-size: 1.15rem">In the last 3 seconds where it goes "And Now You Leave",
there are leakages of other voices in the estimate of Y-Net-mr to a point that
the lyrics are not very intelligible in this estimate. Our model isolates the
target voice better than Y-Net-mr and also improves the intelligibility of the lyrics.</span>
</p>
<p></p>
<hr>
<br>
<div class="col-md-8" style="padding-left:0;padding-bottom: 1em; display: flex">
<div class="column" style="text-align:center; padding-right: 1em">
<h4>Respect (Aretha Franklin)</h4>
</div>
<div class="column" style="text-align:center;">
<div class="videoplay" videoid="#afr1">
<button type="button" class="btn btn-outline-primary">Play</button>
</div>
</div>
<div class="column">
<div class="col-sm-3 videostop" videoid="#afr1">
<button type="button" class="btn btn-outline-primary">Stop</button>
</div>
</div>
</div>
<iframe id="afr1" class="afr1_video"
width="100%" height="345"
src="https://www.youtube.com/embed/kce_zDH-OVA?start=46&end=50&showinfo=0&controls=0&enablejsapi=1&disablekb=1"
title="Aretha Franklin - Respect #1" frameborder="0"
allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture"
allowfullscreen></iframe>
</br>
<div class="row">
<div class="border">
<div class="col">
<h5>Target Voice: Lead voice</h5>
</div>
<div style="display: flex">
<div class="column" style="text-align:center;display: flex">
<div style="padding-top: 0.5em;display: flex">
<div class="col-xs-1 excol3">
<div class="src4_model">Mixture</div>
<img class="spec"
src="../estimated_demo/kce_zDH-OVA_0_43_to_0_5/mix.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/kce_zDH-OVA_0_43_to_0_5/mix.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;">Y-Net-mr [1]</div>
<img class="spec"
src="../estimated_demo/kce_zDH-OVA_0_43_to_0_5/y-net-mr7.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/kce_zDH-OVA_0_43_to_0_5/y-net-mr7.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;">Y-Net-mr-V (Ours)</div>
<img class="spec"
src="../estimated_demo/kce_zDH-OVA_0_43_to_0_5/vocalist.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/kce_zDH-OVA_0_43_to_0_5/vocalist.mp3"
controls=""></audio>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
<br>
<p class="lead mb-0" align="justify">
<h6 style="font-size: 1.25rem">Observation:</h6>
<span style="font-size: 1.15rem">Note that the Y-Net-mr estimate in the above example contains leakage of background vocals around the last second of audio,
while our estimate doesn't have this leakage.</span>
</p>
<p></p>
<hr>
</br>
<div class="col-md-8" style="padding-left:0;padding-bottom: 1em; display: flex">
<div class="column" style="text-align:center; padding-right: 1em">
<h4>The Show Must Go On (Queen)</h4>
</div>
<div class="column" style="text-align:center;">
<div class="videoplay" videoid="#q2">
<button type="button" class="btn btn-outline-primary">Play</button>
</div>
</div>
<div class="column">
<div class="col-sm-3 videostop" videoid="#q2">
<button type="button" class="btn btn-outline-primary">Stop</button>
</div>
</div>
</div>
<iframe id="q2" class="q2_video"
width="100%" height="345"
src="https://www.youtube.com/embed/sEnTMgzw8ow?start=96&end=100&showinfo=0&controls=0&enablejsapi=1&disablekb=1"
title="Queen Medley 2" frameborder="0"
allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture"
allowfullscreen></iframe>
</br>
<div class="row">
<div class="border">
<div class="col">
<h5>Target Voice: Lead Singer (center)</h5>
</div>
<div style="display: flex">
<div class="column" style="text-align:center;display: flex">
<div style="padding-top: 0.5em;display: flex">
<div class="col-xs-1 excol3">
<div class="src4_model">Mixture</div>
<img class="spec"
src="../estimated_demo/sEnTMgzw8ow_1_29_to_1_47/mix_mod.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/sEnTMgzw8ow_1_29_to_1_47/mix_mod.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;">Y-Net-mr [1]</div>
<img class="spec"
src="../estimated_demo/sEnTMgzw8ow_1_29_to_1_47/y-net-mr7_mod.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/sEnTMgzw8ow_1_29_to_1_47/y-net-mr7_mod.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;">Y-Net-mr-V (Ours)</div>
<img class="spec"
src="../estimated_demo/sEnTMgzw8ow_1_29_to_1_47/vocalist_mod.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/sEnTMgzw8ow_1_29_to_1_47/vocalist_mod.mp3"
controls=""></audio>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
<br>
<p class="lead mb-0" align="justify">
<h6 style="font-size: 1.25rem">Observation:</h6>
<span style="font-size: 1.15rem">Note that the voice estimated by Y-Net-mr estimate in the above example is distorted more than the estimate of our model
in the last second where the word "scar". The word "scar" is more intelligible in our estimate.</span>
</p>
<p></p>
<hr>
<br>
<div class="col-md-8" style="padding-left:0;padding-bottom: 1em; display: flex">
<div class="column" style="text-align:center;; padding-right: 1em">
<h4>Aicha (Cheb Khaled)</h4>
</div>
<div class="column" style="text-align:center;">
<div class="videoplay" videoid="#fs">
<button type="button" class="btn btn-outline-primary">Play</button>
</div>
</div>
<div class="column">
<div class="col-sm-3 videostop" videoid="#fs">
<button type="button" class="btn btn-outline-primary">Stop</button>
</div>
</div>
</div>
<iframe id="fs" class="fs_video"
width="100%" height="345"
src="https://www.youtube.com/embed/it6Ud6PDPes?start=142&end=146&showinfo=0&controls=0&enablejsapi=1&disablekb=1"
title="French Song" frameborder="0"
allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture"
allowfullscreen></iframe>
</br>
<div class="row">
<div class="border">
<div class="col">
<h5>Target Voice: Lead Singer (center)</h5>
</div>
<div style="display: flex">
<div class="column" style="text-align:center;display: flex">
<div style="padding-top: 0.5em;display: flex">
<div class="col-xs-1 excol3">
<div class="src4_model">Mixture</div>
<img class="spec"
src="../estimated_demo/it6Ud6PDPes_2_22_to_2_27/mix.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/it6Ud6PDPes_2_22_to_2_27/mix.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div class="src4_model">Y-Net-mr [1]</div>
<img class="spec"
src="../estimated_demo/it6Ud6PDPes_2_22_to_2_27/y-net-mr7.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/it6Ud6PDPes_2_22_to_2_27/y-net-mr7.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;">Y-Net-mr-V (Ours)</div>
<img class="spec"
src="../estimated_demo/it6Ud6PDPes_2_22_to_2_27/vocalist.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/it6Ud6PDPes_2_22_to_2_27/vocalist.mp3"
controls=""></audio>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
<br><p class="lead mb-0" align="justify">
<h6 style="font-size: 1.25rem">Observation:</h6>
<span style="font-size: 1.15rem"> Note that the sound clarity of the voice in our estimate (especially in the beginning) is clearly
better than the Y-Net-mr estimate in the above example. Also, our model better suppresses the
interferences of backing vocals around the interval between 1s to 2s (better seen
in the spectrograms around this time interval).</span>
</p>
<p></p>
<hr>
<br>
<div class="row">
<h3 class="col-sm-4" style="display: inline-block">References</h3>
</div>
<p class="lead mb-0" align="justify">
[1] J. F. Montesinos, V. S. Kadandale, and G. Haro, “A cappella: Audio-visual Singing Voice Separation,” in BMVC, 2021.
</br>
</p>
<!--<div class="row" style="padding-bottom: 1em">
<div class="column" style="text-align:center;m; padding-right: 1em">
<h4>Isn't She Lovely (Sample 1)</h4>
</div>
<div class="column" style="text-align:center;">
<div class="videoplay" videoid="#isl1">
<button type="button" class="btn btn-outline-primary">Play</button>
</div>
</div>
<div class="column">
<div class="col-sm-3 videostop" videoid="#isl1">
<button type="button" class="btn btn-outline-primary">Stop</button>
</div>
</div>
</div>
<iframe id="isl1" class="isl1_video"
width="100%" height="345"
src="https://www.youtube.com/embed/cttFanV0o7c?start=49&end=57&showinfo=0&controls=0&enablejsapi=1&disablekb=1"
title="Isn't She Lovely Sample1" frameborder="0"
allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture"
allowfullscreen></iframe>
</br>
<div class="row">
<div class="border">
<div class="col">
<h5>Target Voice: Top-right Face</h5>
</div>
<div style="display: flex">
<div class="column" style="text-align:center;0.9em;display: flex">
<div style="padding-top: 0.5em;display: flex">
<div class="col-xs-1 excol3">
<div class="src4_model">Mixture</div>
<img class="spec"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/mix_top-right.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/mix_top-right.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div class="src4_model">U-Net</div>
<img class="spec"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/unet-7_top-right.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/unet-7_top-right.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;">LLCP</div>
<img class="spec"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/llcp_top-right.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/llcp_top-right.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;">Y-net-m (Ours)</div>
<img class="spec"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/y-net-m7_top-right.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/y-net-m7_top-right.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;">Y-Net-g7</div>
<img class="spec"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/y-net-g7_top-right.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/y-net-g7_top-right.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;">Y-Net-gr (Ours)</div>
<img class="spec"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/y-net-gr7_top-right.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/y-net-gr7_top-right.mp3"
controls=""></audio>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
<p class="lead mb-0" align="justify">
This is an example with multiple voices and no instrumental accompaniment.
Since our audio-visual models (Y-Net) rely on the lip motion to estimate a voice,
we choose a segment in this video where the lip motions differ for various voices at some point.
In this example, the desired voice to separate is that of the <b>top-right</b> face.
We show that our Y-Net outperforms the
baseline models - the audio-only model U-Net and LLCP (Looking to listen at the cocktail party model [1]).
Listen carefully to the part where it goes "Isn't she pretty".
In LLCP estimate, multiple voices are leaking into its estimate. Our Y-net models perform better.
It is a particularly challenging mixture since all the voices are
performed by the same singer.</br>
</p>
<p></p>
</br></br>
<div class="row" style="padding-bottom: 1em">
<div class="column" style="text-align:center;m; padding-right: 1em">
<h4>Isn't She Lovely (Sample 2)</h4>
</div>
<div class="column" style="text-align:center;">
<div class="videoplay" style="display: inline-block" videoid="#isl2">
<button type="button" class="btn btn-outline-primary">Play</button>
</div>
</div>
<div class="column">
<div class="col-sm-3 videostop" videoid="#isl2">
<button type="button" class="btn btn-outline-primary">Stop</button>
</div>
</div>
</div>
<iframe id="isl2" class="isl2_video"
width="100%" height="345"
src="https://www.youtube.com/embed/cttFanV0o7c?start=39&end=43&showinfo=0&controls=0&enablejsapi=1&disablekb=1"
title="Isn't She Lovely Sample2" frameborder="0"
allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture"
allowfullscreen></iframe>
</br>
<div class="row">
<div class="border">
<div class="col">
<h5>Target Voice: Bottom-left Face</h5>
</div>
<div style="display: flex">
<div class="column" style="text-align:center;0.9em;display: flex">
<div style="padding-top: 0.5em;display: flex">
<div class="col-xs-1 excol3">
<div class="src4_model">Mixture</div>
<img class="spec"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/mix_bottom-left.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/mix_bottom-left.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div class="src4_model">U-Net</div>
<img class="spec"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/unet-7_bottom-left.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/unet-7_bottom-left.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;">LLCP</div>
<img class="spec"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/llcp_bottom-left.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/llcp_bottom-left.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;">Y-net-m (Ours)</div>
<img class="spec"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/y-net-m7_bottom-left.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/y-net-m7_bottom-left.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;">Y-Net-mr7</div>
<img class="spec"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/y-net-mr7_bottom-left.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/y-net-mr7_bottom-left.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;">Y-Net-g7</div>
<img class="spec"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/y-net-g7_bottom-left.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/y-net-g7_bottom-left.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="text-align:center;">Y-Net-gr (Ours)</div>
<img class="spec"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/y-net-gr7_bottom-left.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/cttFanV0o7c_0_07_to_2_44/y-net-gr7_bottom-left.mp3"
controls=""></audio>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
<p class="lead mb-0" align="justify">
Here, we show another segment from the same video as in the previous example.
But, here, we estimate the voice sung by the <b>bottom-left</b> face which is not a lead voice.
The <b>bottom-left</b> face is singing something that is not lyrical. We chose this segment for this demo
because in the previous one, the accompaniment voices were making the same lip movements as the main voice.
None of the models were trained to separate a non-lyrical voice like this, but our Y-Net outperforms the other
models by suppressing the other voices.
This song shows that our Y-Net can also separate accompaniment voices.
In this case, Y-Net-r performs better than Y-Net-m, for example, in the part "making one as lovely", the Y-Net-m leaks
more lead voice into the estimate than Y-Net-r.
</br>
</p>
<p></p>
</br></br>-->