notesum.ai

Published at November 27

Multiple Choice Learning for Efficient Speech Separation with Many Speakers

cs.SD

cs.LG

eess.AS

stat.ML

Released Date: November 27, 2024

Authors: David Perera¹, François Derrida¹, Théo Mariotte, Gaël Richard, Slim Essid¹

Aff.: ¹LTCI, Telecom Paris, Institut polytechnique de Paris, France

Arxiv: http://arxiv.org/abs/2411.18497v1

Refer to caption

	WSJ0-mix				LibriMix
Loss	2 spks	3 spks	4 spks	5 spks	10 spks^∗	20 spks
PIT	19.96	16.13	12.00	10.14	3.17	2.98
MCL	19.74	16.01	11.85	9.92	3.85	2.96
SinkPIT	19.63	16.00	12.12	10.22	3.50	3.10