Sound Demo of PuLSS with oracle speaker diarization on CHiME-6 Corpus


We provide a sound demo based on the test set of the CHiME-6 corpus. For technical details, please refer to the following paper:

Z.-Q. Wang and S. Cornell, "Cross-Talk Speech Reduction, by Separation, for Separation", in submission, 2026.
This demo presents the results of PuLSS with oracle speaker diarization.

Note:
• Each spectrogram is masked using oracle speaker-activity timestamps, resulting in pure black regions in the spectrograms.
• There are four speakers, so there are four masked mixtures (the same input mixture masked by four oracle speaker-activity timestamps), and four separated signals by each algorithm.
• PuLSS clearly outperforms GSS, supervised approach, and mixtures. It better separates the mixed speekers..
• Firefox/Chrome is recommended.

We provide the following examples (click to jump to the example):
○ S01 at 00:02:06
○ S01 at 00:22:57
○ S01 at 01:04:32
○ S01 at 01:34:23
○ S01 at 02:04:53
○ S01 at 02:14:47
○ S01 at 02:21:02
○ S21 at 00:02:19
○ S21 at 00:20:51
○ S21 at 01:17:05
○ S21 at 01:53:04
○ S21 at 02:14:40
○ S21 at 02:18:49
○ S21 at 02:25:53

Signal segment:
S01 at 00:02:06
Systems Speaker Signal Log-compreesed Power Spectrogram
Mixture
(0 of Table III)		 P01
Mixture
(0 of Table III)		 P02
Mixture
(0 of Table III)		 P03
Mixture
(0 of Table III)		 P04
Supervised
(2 of Table III)		 P01
Supervised
(2 of Table III)		 P02
Supervised
(2 of Table III)		 P03
Supervised
(2 of Table III)		 P04
GSS (8-channel)
(1 of Table III)		 P01
GSS (8-channel)
(1 of Table III)		 P02
GSS (8-channel)
(1 of Table III)		 P03
GSS (8-channel)
(1 of Table III)		 P04
PuLSS
(7a of Table III)		 P01
PuLSS
(7a of Table III)		 P02
PuLSS
(7a of Table III)		 P03
PuLSS
(7a of Table III)		 P04

Signal segment:
S01 at 00:22:57
Systems Speaker Signal Log-compreesed Power Spectrogram
Mixture
(0 of Table III)		 P01
Mixture
(0 of Table III)		 P02
Mixture
(0 of Table III)		 P03
Mixture
(0 of Table III)		 P04
Supervised
(2 of Table III)		 P01
Supervised
(2 of Table III)		 P02
Supervised
(2 of Table III)		 P03
Supervised
(2 of Table III)		 P04
GSS (8-channel)
(1 of Table III)		 P01
GSS (8-channel)
(1 of Table III)		 P02
GSS (8-channel)
(1 of Table III)		 P03
GSS (8-channel)
(1 of Table III)		 P04
PuLSS
(7a of Table III)		 P01
PuLSS
(7a of Table III)		 P02
PuLSS
(7a of Table III)		 P03
PuLSS
(7a of Table III)		 P04

Signal segment:
S01 at 01:04:32
Systems Speaker Signal Log-compreesed Power Spectrogram
Mixture
(0 of Table III)		 P01
Mixture
(0 of Table III)		 P02
Mixture
(0 of Table III)		 P03
Mixture
(0 of Table III)		 P04
Supervised
(2 of Table III)		 P01
Supervised
(2 of Table III)		 P02
Supervised
(2 of Table III)		 P03
Supervised
(2 of Table III)		 P04
GSS (8-channel)
(1 of Table III)		 P01
GSS (8-channel)
(1 of Table III)		 P02
GSS (8-channel)
(1 of Table III)		 P03
GSS (8-channel)
(1 of Table III)		 P04
PuLSS
(7a of Table III)		 P01
PuLSS
(7a of Table III)		 P02
PuLSS
(7a of Table III)		 P03
PuLSS
(7a of Table III)		 P04

Signal segment:
S01 at 01:34:23
Systems Speaker Signal Log-compreesed Power Spectrogram
Mixture
(0 of Table III)		 P01
Mixture
(0 of Table III)		 P02
Mixture
(0 of Table III)		 P03
Mixture
(0 of Table III)		 P04
Supervised
(2 of Table III)		 P01
Supervised
(2 of Table III)		 P02
Supervised
(2 of Table III)		 P03
Supervised
(2 of Table III)		 P04
GSS (8-channel)
(1 of Table III)		 P01
GSS (8-channel)
(1 of Table III)		 P02
GSS (8-channel)
(1 of Table III)		 P03
GSS (8-channel)
(1 of Table III)		 P04
PuLSS
(7a of Table III)		 P01
PuLSS
(7a of Table III)		 P02
PuLSS
(7a of Table III)		 P03
PuLSS
(7a of Table III)		 P04

Signal segment:
S01 at 02:04:53
Systems Speaker Signal Log-compreesed Power Spectrogram
Mixture
(0 of Table III)		 P01
Mixture
(0 of Table III)		 P02
Mixture
(0 of Table III)		 P03
Mixture
(0 of Table III)		 P04
Supervised
(2 of Table III)		 P01
Supervised
(2 of Table III)		 P02
Supervised
(2 of Table III)		 P03
Supervised
(2 of Table III)		 P04
GSS (8-channel)
(1 of Table III)		 P01
GSS (8-channel)
(1 of Table III)		 P02
GSS (8-channel)
(1 of Table III)		 P03
GSS (8-channel)
(1 of Table III)		 P04
PuLSS
(7a of Table III)		 P01
PuLSS
(7a of Table III)		 P02
PuLSS
(7a of Table III)		 P03
PuLSS
(7a of Table III)		 P04

Signal segment:
S01 at 02:14:47
Systems Speaker Signal Log-compreesed Power Spectrogram
Mixture
(0 of Table III)		 P01
Mixture
(0 of Table III)		 P02
Mixture
(0 of Table III)		 P03
Mixture
(0 of Table III)		 P04
Supervised
(2 of Table III)		 P01
Supervised
(2 of Table III)		 P02
Supervised
(2 of Table III)		 P03
Supervised
(2 of Table III)		 P04
GSS (8-channel)
(1 of Table III)		 P01
GSS (8-channel)
(1 of Table III)		 P02
GSS (8-channel)
(1 of Table III)		 P03
GSS (8-channel)
(1 of Table III)		 P04
PuLSS
(7a of Table III)		 P01
PuLSS
(7a of Table III)		 P02
PuLSS
(7a of Table III)		 P03
PuLSS
(7a of Table III)		 P04

Signal segment:
S01 at 02:21:02
Systems Speaker Signal Log-compreesed Power Spectrogram
Mixture
(0 of Table III)		 P01
Mixture
(0 of Table III)		 P02
Mixture
(0 of Table III)		 P03
Mixture
(0 of Table III)		 P04
Supervised
(2 of Table III)		 P01
Supervised
(2 of Table III)		 P02
Supervised
(2 of Table III)		 P03
Supervised
(2 of Table III)		 P04
GSS (8-channel)
(1 of Table III)		 P01
GSS (8-channel)
(1 of Table III)		 P02
GSS (8-channel)
(1 of Table III)		 P03
GSS (8-channel)
(1 of Table III)		 P04
PuLSS
(7a of Table III)		 P01
PuLSS
(7a of Table III)		 P02
PuLSS
(7a of Table III)		 P03
PuLSS
(7a of Table III)		 P04

Signal segment:
S21 at 00:02:19
Systems Speaker Signal Log-compreesed Power Spectrogram
Mixture
(0 of Table III)		 P45
Mixture
(0 of Table III)		 P46
Mixture
(0 of Table III)		 P47
Mixture
(0 of Table III)		 P48
Supervised
(2 of Table III)		 P45
Supervised
(2 of Table III)		 P46
Supervised
(2 of Table III)		 P47
Supervised
(2 of Table III)		 P48
GSS (8-channel)
(1 of Table III)		 P45
GSS (8-channel)
(1 of Table III)		 P46
GSS (8-channel)
(1 of Table III)		 P47
GSS (8-channel)
(1 of Table III)		 P48
PuLSS
(7a of Table III)		 P45
PuLSS
(7a of Table III)		 P46
PuLSS
(7a of Table III)		 P47
PuLSS
(7a of Table III)		 P48

Signal segment:
S21 at 00:20:51
Systems Speaker Signal Log-compreesed Power Spectrogram
Mixture
(0 of Table III)		 P45
Mixture
(0 of Table III)		 P46
Mixture
(0 of Table III)		 P47
Mixture
(0 of Table III)		 P48
Supervised
(2 of Table III)		 P45
Supervised
(2 of Table III)		 P46
Supervised
(2 of Table III)		 P47
Supervised
(2 of Table III)		 P48
GSS (8-channel)
(1 of Table III)		 P45
GSS (8-channel)
(1 of Table III)		 P46
GSS (8-channel)
(1 of Table III)		 P47
GSS (8-channel)
(1 of Table III)		 P48
PuLSS
(7a of Table III)		 P45
PuLSS
(7a of Table III)		 P46
PuLSS
(7a of Table III)		 P47
PuLSS
(7a of Table III)		 P48

Signal segment:
S21 at 01:17:05
Systems Speaker Signal Log-compreesed Power Spectrogram
Mixture
(0 of Table III)		 P45
Mixture
(0 of Table III)		 P46
Mixture
(0 of Table III)		 P47
Mixture
(0 of Table III)		 P48
Supervised
(2 of Table III)		 P45
Supervised
(2 of Table III)		 P46
Supervised
(2 of Table III)		 P47
Supervised
(2 of Table III)		 P48
GSS (8-channel)
(1 of Table III)		 P45
GSS (8-channel)
(1 of Table III)		 P46
GSS (8-channel)
(1 of Table III)		 P47
GSS (8-channel)
(1 of Table III)		 P48
PuLSS
(7a of Table III)		 P45
PuLSS
(7a of Table III)		 P46
PuLSS
(7a of Table III)		 P47
PuLSS
(7a of Table III)		 P48

Signal segment:
S21 at 01:53:04
Systems Speaker Signal Log-compreesed Power Spectrogram
Mixture
(0 of Table III)		 P45
Mixture
(0 of Table III)		 P46
Mixture
(0 of Table III)		 P47
Mixture
(0 of Table III)		 P48
Supervised
(2 of Table III)		 P45
Supervised
(2 of Table III)		 P46
Supervised
(2 of Table III)		 P47
Supervised
(2 of Table III)		 P48
GSS (8-channel)
(1 of Table III)		 P45
GSS (8-channel)
(1 of Table III)		 P46
GSS (8-channel)
(1 of Table III)		 P47
GSS (8-channel)
(1 of Table III)		 P48
PuLSS
(7a of Table III)		 P45
PuLSS
(7a of Table III)		 P46
PuLSS
(7a of Table III)		 P47
PuLSS
(7a of Table III)		 P48

Signal segment:
S21 at 02:14:40
Systems Speaker Signal Log-compreesed Power Spectrogram
Mixture
(0 of Table III)		 P45
Mixture
(0 of Table III)		 P46
Mixture
(0 of Table III)		 P47
Mixture
(0 of Table III)		 P48
Supervised
(2 of Table III)		 P45
Supervised
(2 of Table III)		 P46
Supervised
(2 of Table III)		 P47
Supervised
(2 of Table III)		 P48
GSS (8-channel)
(1 of Table III)		 P45
GSS (8-channel)
(1 of Table III)		 P46
GSS (8-channel)
(1 of Table III)		 P47
GSS (8-channel)
(1 of Table III)		 P48
PuLSS
(7a of Table III)		 P45
PuLSS
(7a of Table III)		 P46
PuLSS
(7a of Table III)		 P47
PuLSS
(7a of Table III)		 P48

Signal segment:
S21 at 02:18:49
Systems Speaker Signal Log-compreesed Power Spectrogram
Mixture
(0 of Table III)		 P45
Mixture
(0 of Table III)		 P46
Mixture
(0 of Table III)		 P47
Mixture
(0 of Table III)		 P48
Supervised
(2 of Table III)		 P45
Supervised
(2 of Table III)		 P46
Supervised
(2 of Table III)		 P47
Supervised
(2 of Table III)		 P48
GSS (8-channel)
(1 of Table III)		 P45
GSS (8-channel)
(1 of Table III)		 P46
GSS (8-channel)
(1 of Table III)		 P47
GSS (8-channel)
(1 of Table III)		 P48
PuLSS
(7a of Table III)		 P45
PuLSS
(7a of Table III)		 P46
PuLSS
(7a of Table III)		 P47
PuLSS
(7a of Table III)		 P48

Signal segment:
S21 at 02:25:53
Systems Speaker Signal Log-compreesed Power Spectrogram
Mixture
(0 of Table III)		 P45
Mixture
(0 of Table III)		 P46
Mixture
(0 of Table III)		 P47
Mixture
(0 of Table III)		 P48
Supervised
(2 of Table III)		 P45
Supervised
(2 of Table III)		 P46
Supervised
(2 of Table III)		 P47
Supervised
(2 of Table III)		 P48
GSS (8-channel)
(1 of Table III)		 P45
GSS (8-channel)
(1 of Table III)		 P46
GSS (8-channel)
(1 of Table III)		 P47
GSS (8-channel)
(1 of Table III)		 P48
PuLSS
(7a of Table III)		 P45
PuLSS
(7a of Table III)		 P46
PuLSS
(7a of Table III)		 P47
PuLSS
(7a of Table III)		 P48