Kalman-EM Algorithm for Dereverberation

B. Schwartz, S. Gannot, and E. A. P. Habets, “Online speech dereverberation using Kalman filter and EM algorithm,” Accepted for publication, IEEE Trans. on Audio, Speech, and Language Processing, Nov. 2014.

Static, 480ms, Male

CLEAN 0:07 REVERB 0:07 RKEMD 0:07

Static, 630ms, Female

CLEAN 0:06 REVERB 0:07 RKEMD 0:07

Static, 940ms, Female

CLEAN 0:12 REVERB 0:12 RKEMD 0:12

Static, 940ms, Male

CLEAN 0:10 REVERB 0:10 RKEMD 0:10

Static, Noisy, RSNR 5dB, 940ms, Male

CLEAN 0:10 REVERB 0:10 RKEMD 0:10

Dynamic, 750ms, David. The speaker is standing (0-8 sec), walking to a different point (8-16.5 sec), and finally sitting down (16.5-22.6 sec).

CLEAN 0:23 REVERB 0:23 RKEMD 0:23

Dynamic, 750ms, Rafael. The speaker is standing (0-9 sec), then turning his head and speaking to the opposite direction (9-19.3 sec).

CLEAN 0:19 REVERB 0:19 RKEMD 0:19

Dynamic, 750ms, Merav. Dynamic, 750ms, Merav. The speaker is standing (0-7 sec), walking to a different point (7-15 sec), and finally sitting down(15-26 sec).

CLEAN 0:26 REVERB 0:26 RKEMD 0:26

Dynamic, 750ms, Michal. The speaker is sitting (0-9 sec), then turning her head and speaking to the opposite direction (9-20 sec).

CLEAN 0:20 REVERB 0:20 RKEMD 0:20

Dynamic, 750ms, Rafael. The speaker is standing (0-11 sec), walking to a different point (11-19 sec), then standing in place (19-29 sec).

CLEAN 0:29 REVERB 0:29 RKEMD 0:29