-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdemos_all.html
3500 lines (3456 loc) · 171 KB
/
demos_all.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
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
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
---
layout: page
---
PRESS CTRL+SHIFT+R TO VIEW THE UPDATED PAGE.
<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 recommend the use of headphones when 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>Real World Examples</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="row" style="padding-bottom: 1em">
<div class="column" style="padding-left: 1em; padding-right: 1em">
<h4>Isn't She Lovely (Sample 1)</h4>
</div>
<div class="column" style="padding-left: 1em">
<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="padding-left: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="padding-left: 35px">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="padding-left: 65px">Y-net-m</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="padding-left: 80px">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="padding-left: 80px">Y-Net-gr7</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="padding-left: 1em; padding-right: 1em">
<h4>Isn't She Lovely (Sample 2)</h4>
</div>
<div class="column" style="padding-left: 1em">
<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="padding-left: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="padding-left: 35px">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="padding-left: 65px">Y-net-m</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="padding-left: 80px">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="padding-left: 80px">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="padding-left: 80px">Y-Net-gr7</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>
<div class="row" style="padding-bottom: 1em">
<div class="column" style="padding-left: 1em; padding-right: 1em">
<h4>Paranoid Android</h4>
</div>
<div class="column" style="padding-left: 1em">
<div class="videoplay" videoid="#pa">
<button type="button" class="btn btn-outline-primary">Play</button>
</div>
</div>
<div class="column">
<div class="col-sm-3 videostop" videoid="#pa">
<button type="button" class="btn btn-outline-primary">Stop</button>
</div>
</div>
</div>
<iframe id="pa" class="pa_video"
width="100%" height="345"
src="https://www.youtube.com/embed/Gayh_GrCKgU?start=313&end=317&showinfo=0&controls=0&enablejsapi=1&disablekb=1"
title="Paranoid Android" 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-center Face</h5>
</div>
<div style="display: flex">
<div class="column" style="padding-left: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/Gayh_GrCKgU_5_11_to_5_35/mix_center.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/Gayh_GrCKgU_5_11_to_5_35/mix_center.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div class="src4_model">U-Net</div>
<img class="spec"
src="../estimated_demo/Gayh_GrCKgU_5_11_to_5_35/unet-7_center.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/Gayh_GrCKgU_5_11_to_5_35/unet-7_center.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 35px">LLCP</div>
<img class="spec"
src="../estimated_demo/Gayh_GrCKgU_5_11_to_5_35/llcp_center.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/Gayh_GrCKgU_5_11_to_5_35/llcp_center.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 65px">Y-net-m</div>
<img class="spec"
src="../estimated_demo/Gayh_GrCKgU_5_11_to_5_35/y-net-m7_center.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/Gayh_GrCKgU_5_11_to_5_35/y-net-m7_center.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-mr7</div>
<img class="spec"
src="../estimated_demo/Gayh_GrCKgU_5_11_to_5_35/y-net-mr7_center.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/Gayh_GrCKgU_5_11_to_5_35/y-net-mr7_center.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-g7</div>
<img class="spec"
src="../estimated_demo/Gayh_GrCKgU_5_11_to_5_35/y-net-g7_center.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/Gayh_GrCKgU_5_11_to_5_35/y-net-g7_center.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-gr7</div>
<img class="spec"
src="../estimated_demo/Gayh_GrCKgU_5_11_to_5_35/y-net-gr7_center.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/Gayh_GrCKgU_5_11_to_5_35/y-net-gr7_center.mp3"
controls=""></audio>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
<p class="lead mb-0" align="justify">
This is an example with multiple singing voices of the same singer along with the background music.
It is more challenging than the previous examples because of the presence of background music in addition to
the voices. We are trying to estimate the <b>center</b> singer's voice.
Both in LLCP estimate and U-Net estimate, there is undersirable abrupt
variation in the intensity levels of the target voice.
Such abnormal effects are not found in the Y-Net estimates.
Also, the Y-Net estimates this voice better than the baseline models in the sense that the target singer
voice is enhanced the most in the Y-Net estimates.
</p>
<p></p>
</br></br>
<div class="row" style="padding-bottom: 1em">
<div class="column" style="padding-left: 1em; padding-right: 1em">
<h4>Disney Medley</h4>
</div>
<div class="column" style="padding-left: 1em">
<div class="videoplay" videoid="#dm">
<button type="button" class="btn btn-outline-primary">Play</button>
</div>
</div>
<div class="column">
<div class="col-sm-3 videostop" videoid="#dm">
<button type="button" class="btn btn-outline-primary">Stop</button>
</div>
</div>
</div>
<iframe id="dm" class="dm_video"
width="100%" height="345"
src="https://www.youtube.com/embed/Dyo7jzaCUhk?start=44&end=52&showinfo=0&controls=0&enablejsapi=1&disablekb=1"
title="Disney Medley" 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: Center Face</h5>
</div>
<div style="display: flex">
<div class="column" style="padding-left: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/Dyo7jzaCUhk_0_02_to_5_2/mix_center.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/Dyo7jzaCUhk_0_02_to_5_2/mix_center.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div class="src4_model">U-Net</div>
<img class="spec"
src="../estimated_demo/Dyo7jzaCUhk_0_02_to_5_2/unet-7_center.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/Dyo7jzaCUhk_0_02_to_5_2/unet-7_center.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 35px">LLCP</div>
<img class="spec"
src="../estimated_demo/Dyo7jzaCUhk_0_02_to_5_2/llcp_center.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/Dyo7jzaCUhk_0_02_to_5_2/llcp_center.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 65px">Y-net-m</div>
<img class="spec"
src="../estimated_demo/Dyo7jzaCUhk_0_02_to_5_2/y-net-m7_center.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/Dyo7jzaCUhk_0_02_to_5_2/y-net-m7_center.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-mr7</div>
<img class="spec"
src="../estimated_demo/Dyo7jzaCUhk_0_02_to_5_2/y-net-mr7_center.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/Dyo7jzaCUhk_0_02_to_5_2/y-net-mr7_center.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-g7</div>
<img class="spec"
src="../estimated_demo/Dyo7jzaCUhk_0_02_to_5_2/y-net-g7_center.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/Dyo7jzaCUhk_0_02_to_5_2/y-net-g7_center.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-gr7</div>
<img class="spec"
src="../estimated_demo/Dyo7jzaCUhk_0_02_to_5_2/y-net-gr7_center.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/Dyo7jzaCUhk_0_02_to_5_2/y-net-gr7_center.mp3"
controls=""></audio>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
<p class="lead mb-0" align="justify">
This example seems to contain only the voices, without any instrumental accompaniment. However, there are voices
which are trying to mimic instruments. For example, there is a beatboxing voice which mimics drum kit sounds
like kicks and snares. We are interested in estimating the voice of the singer in the <b>center</b>.
The estimate of the audio-only model U-Net sounds attenuated in the region around the words "in perfect".
The other audio-visual models do not have this problem. The beatboxing voice is suppressed better in our Y-Net
estimates than in the LLCP estimate: for example, the snare drum sound that appears around the word "for"
is leaking more in the LLCP estimate and the U-Net estimate than in the Y-Net estimates. Y-Net-r gets the word
"once" more clearly than the Y-Net-m. Y-Net-r seems to be the best model here.
</p>
<p></p>
</br></br>
<div class="row" style="padding-bottom: 1em">
<div class="column" style="padding-left: 1em; padding-right: 1em">
<h4>Queen Medley</h4>
</div>
<div class="column" style="padding-left: 1em">
<div class="videoplay" videoid="#qm">
<button type="button" class="btn btn-outline-primary">Play</button>
</div>
</div>
<div class="column">
<div class="col-sm-3 videostop" videoid="#qm">
<button type="button" class="btn btn-outline-primary">Stop</button>
</div>
</div>
</div>
<iframe id="qm" class="qm_video"
width="100%" height="345"
src="https://www.youtube.com/embed/sEnTMgzw8ow?start=131&end=143&showinfo=0&controls=0&enablejsapi=1&disablekb=1"
title="Queen Medley" 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: Center Face</h5>
</div>
<div style="display: flex">
<div class="column" style="padding-left: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/sEnTMgzw8ow_2_11_to_2_33/mix.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/sEnTMgzw8ow_2_11_to_2_33/mix.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div class="src4_model">U-Net</div>
<img class="spec"
src="../estimated_demo/sEnTMgzw8ow_2_11_to_2_33/unet-7.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/sEnTMgzw8ow_2_11_to_2_33/unet-7.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 35px">LLCP</div>
<img class="spec"
src="../estimated_demo/sEnTMgzw8ow_2_11_to_2_33/llcp.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/sEnTMgzw8ow_2_11_to_2_33/llcp.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 65px">Y-net-m</div>
<img class="spec"
src="../estimated_demo/sEnTMgzw8ow_2_11_to_2_33/y-net-m7.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/sEnTMgzw8ow_2_11_to_2_33/y-net-m7.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-mr7</div>
<img class="spec"
src="../estimated_demo/sEnTMgzw8ow_2_11_to_2_33/y-net-mr7.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/sEnTMgzw8ow_2_11_to_2_33/y-net-mr7.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-g7</div>
<img class="spec"
src="../estimated_demo/sEnTMgzw8ow_2_11_to_2_33/y-net-g7.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/sEnTMgzw8ow_2_11_to_2_33/y-net-g7.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-gr7</div>
<img class="spec"
src="../estimated_demo/sEnTMgzw8ow_2_11_to_2_33/y-net-gr7.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/sEnTMgzw8ow_2_11_to_2_33/y-net-gr7.mp3"
controls=""></audio>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
<p class="lead mb-0" align="justify">
In this example, we are trying to estimate the voice of the singer in the <b>center</b>.
It seems to be an example with only voices. However, like in the Disney Medley, this example contains
voices that mimic drum kit sounds along with the other singing voices.
As expected, our models do not single out a voice in the part which goes "the show must go on" since it is sung simultaeneously
by the center, left-middle and right-middle faces with almost the same lip motions.
Note that the audio-visual models enhance "the show must go on" part much better than the audio-only model (U-Net).
LLCP model attenuates the word "must" more than Y-Net. In Y-Net-m, all the voices (including the background vocals)
seem to be boosted just before the words "the show" (also visible in the spectrogram around t=3s).
Such an effect is not seen in Y-Net-r. Y-Net-r seems to be the best performing model here.
</br>
</p>
<p></p>
</br></br>
<div class="row" style="padding-bottom: 1em">
<div class="column" style="padding-left: 1em; padding-right: 1em">
<h4>Bad Guy</h4>
</div>
<div class="column" style="padding-left: 1em">
<div class="videoplay" videoid="#bg">
<button type="button" class="btn btn-outline-primary">Play</button>
</div>
</div>
<div class="column">
<div class="col-sm-3 videostop" videoid="#bg">
<button type="button" class="btn btn-outline-primary">Stop</button>
</div>
</div>
</div>
<iframe id="bg" class="bg_video"
width="100%" height="345"
src="https://www.youtube.com/embed/pRh9rKd2j64?start=30&end=42&showinfo=0&controls=0&enablejsapi=1&disablekb=1"
title="Bad Guy" 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: Female Singer</h5>
</div>
<div style="display: flex">
<div class="column" style="padding-left: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/pRh9rKd2j64_0_15_to_0_55/mix.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/pRh9rKd2j64_0_15_to_0_55/mix.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div class="src4_model">U-Net</div>
<img class="spec"
src="../estimated_demo/pRh9rKd2j64_0_15_to_0_55/unet-7.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/pRh9rKd2j64_0_15_to_0_55/unet-7.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 35px">LLCP</div>
<img class="spec"
src="../estimated_demo/pRh9rKd2j64_0_15_to_0_55/llcp.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/pRh9rKd2j64_0_15_to_0_55/llcp.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 65px">Y-net-m</div>
<img class="spec"
src="../estimated_demo/pRh9rKd2j64_0_15_to_0_55/y-net-m7.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/pRh9rKd2j64_0_15_to_0_55/y-net-m7.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-mr7</div>
<img class="spec"
src="../estimated_demo/pRh9rKd2j64_0_15_to_0_55/y-net-mr7.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/pRh9rKd2j64_0_15_to_0_55/y-net-mr7.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-g7</div>
<img class="spec"
src="../estimated_demo/pRh9rKd2j64_0_15_to_0_55/y-net-g7.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/pRh9rKd2j64_0_15_to_0_55/y-net-g7.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-gr7</div>
<img class="spec"
src="../estimated_demo/pRh9rKd2j64_0_15_to_0_55/y-net-gr7.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/pRh9rKd2j64_0_15_to_0_55/y-net-gr7.mp3"
controls=""></audio>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
<p class="lead mb-0" align="justify">
This is an example of a mix containing single voice and background music. We notice that the audio-visual models
do not particularly improve upon the performance of the audio-only model U-Net. In general, unlike in mixes
containing multiple voices, we notice that the audio-visual models did not provide additional advantage over the
audio-only models in the case of examples with single voice and accompaniment.
</p>
</br></br>
<h1>Our Multi-voice Montage</h1>
<p class="lead mb-0" align="justify">
The videos from YouTube that we saw earlier do not provide the original isolated voices.
For this reason, we could only qualitatively evaluate the performance
of target voice separation models on such videos. To be able to evaluate quantitatively,
we recorded a multi-voice montage with an amateur singer so that we can use the
individual voices as ground truth during evaluation. It is a montage with six voices and an
accompaniment track.
It is a very challenging example because the same singer is shown singing multiple voices at the
same time.
The song being performed is "The Circle of Life", a popular soundtrack
from the movie "The Lion King". Apart from English, the song lyrics also contains Zulu:
a language that was not a part of the training set.
Also, this singer was not a part of the training set.
Below are the estimates for the voices: <b>Rafiki (bottom-center)</b>,
<b>Leones (bottom-left)</b> and <b>Backing Vocals (top-left)</b>.
See the right column in Table 1 of the paper for the quantitative metrics.
</p>
</br>
<div class="row" style="padding-bottom: 1em">
<div class="column" style="padding-left: 1em; padding-right: 1em">
<h4>The Circle Of Life</h4>
</div>
<div class="column" style="padding-left: 1em">
<div class="videoplay" videoid="#tcol">
<button type="button" class="btn btn-outline-primary">Play</button>
</div>
</div>
<div class="column">
<div class="col-sm-3 videostop" videoid="#tcol">
<button type="button" class="btn btn-outline-primary">Stop</button>
</div>
</div>
</div>
<iframe id="tcol" class="tcol_video"
width="100%" height="345"
src="https://www.youtube.com/embed/XvBOH8FGias?start=175&end=187&showinfo=0&controls=0&enablejsapi=1&disablekb=1"
title="The circle of life" frameborder="0"
allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture"
allowfullscreen></iframe>
</br>
<div class="row">
<div class="border">
<div style="display: flex">
<div class="column" style="padding-left: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/tcol/mix.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/mix.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div class="src4_model">Ground truth Rafiki</div>
<img class="spec"
src="../estimated_demo/tcol/gt_rafiki.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/gt_rafiki.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div class="src4_model">Ground truth Leones</div>
<img class="spec"
src="../estimated_demo/tcol/gt_leones.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/gt_leones.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 35px">Ground truth Backing Vocals</div>
<img class="spec"
src="../estimated_demo/tcol/gt_backtrack.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/gt_backtrack.mp3"
controls=""></audio>
</div>
</div>
</div>
</div>
</div>
</div>
<div class="col">
</br></br>
<p class="lead mb-0" align="justify">
Below are the estimates for the target voice <b>Rafiki</b> (sung by the <b>bottom-center</b> face).
After 6s (see the ground truth spectrogram),
this voice is reduced to silence. From the spectrograms, it is clearly evident that
our Y-Net captures the silence in this region better than the other models. Y-Net-r
seems to be the best performing model here.
</p>
</br></br>
</div>
<div class="border">
<div class="col">
<h5>Target Voice: Rafiki (Bottom-center Face)</h5>
</div>
<div style="display: flex">
<div class="column" style="padding-left:0.9em;display: flex">
<div style="padding-top: 0.5em;display: flex">
<div class="col-xs-1 excol3">
<div class="src4_model">Ground truth</div>
<img class="spec"
src="../estimated_demo/tcol/gt_rafiki.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/gt_rafiki.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div class="src4_model">U-Net</div>
<img class="spec"
src="../estimated_demo/tcol/unet-7_rafiki.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/unet-7_rafiki.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 35px">LLCP</div>
<img class="spec"
src="../estimated_demo/tcol/llcp_rafiki.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/llcp_rafiki.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 65px">Y-net-m</div>
<img class="spec"
src="../estimated_demo/tcol/y-net-m7_rafiki.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/y-net-m7_rafiki.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-mr7</div>
<img class="spec"
src="../estimated_demo/tcol/y-net-mr7_rafiki.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/y-net-mr7_rafiki.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-g7</div>
<img class="spec"
src="../estimated_demo/tcol/y-net-g7_rafiki.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/y-net-g7_rafiki.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-gr7</div>
<img class="spec"
src="../estimated_demo/tcol/y-net-gr7_rafiki.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/y-net-gr7_rafiki.mp3"
controls=""></audio>
</div>
</div>
</div>
</div>
</div>
</div>
<div class="col">
</br></br>
<p class="lead mb-0" align="justify">
Below are the estimates for the target voice <b>Leones</b> (sung by the <b>bottom-left</b> face).
The <b>Rafiki</b> voice leaks the most in the U-Net estimate and the LLCP estimate.
Y-Net suppresses <b>Rafiki</b> voice better. However, there are leakages from other voices
in the estimate of all the models. Note that the leakages in Y-Net estimates mainly constitute
the other voices (not separately shown here for brevity)
which are singing the Zulu phrase "<i>Ingonyama nengw’ enamabala</i>" with slight delay than
compared to the ground truth <b>Leones</b> voice, producing effectively
the same lip motion sequence. This is understandable because Y-Net relies on the lip motion to
estimate a voice and it can not perform well when there are similar background voices produced
by similar lip motion sequence as in this example.
Around 8s, there is more leakage of <b>Backing Vocals</b> (sung by <b>top-left</b> face) in
the Y-Net-r estimate than in Y-Net-m estimate.
Y-Net-m seems to be the best performing model here.
</p>
</br></br>
</div>
<div class="border">
<div class="col">
<h5>Target Voice: Leones (Bottom-left Face)</h5>
</div>
<div style="display: flex">
<div class="column" style="padding-left:0.9em;display: flex">
<div style="padding-top: 0.5em;display: flex">
<div class="col-xs-1 excol3">
<div class="src4_model">Ground truth</div>
<img class="spec"
src="../estimated_demo/tcol/gt_leones.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/gt_leones.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div class="src4_model">U-Net</div>
<img class="spec"
src="../estimated_demo/tcol/unet-7_leones.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/unet-7_leones.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 35px">LLCP</div>
<img class="spec"
src="../estimated_demo/tcol/llcp_leones.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/llcp_leones.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 65px">Y-net-m</div>
<img class="spec"
src="../estimated_demo/tcol/y-net-m7_leones.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/y-net-m7_leones.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-mr7</div>
<img class="spec"
src="../estimated_demo/tcol/y-net-mr7_leones.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/y-net-mr7_leones.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-g7</div>
<img class="spec"
src="../estimated_demo/tcol/y-net-g7_leones.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"
src="../estimated_demo/tcol/y-net-g7_leones.mp3"
controls=""></audio>
</div>
</div>
<div class="col-xs-1 excol3">
<div style="padding-left: 80px">Y-Net-gr7</div>
<img class="spec"
src="../estimated_demo/tcol/y-net-gr7_leones.png">
<div class="col-xs-1 excol3">
<audio class="src4audio"