Conv-TasNet: Surpassing Ideal Time-Frequency Magnitude Masking for Speech Separation

20 Sep 2018Yi LuoNima Mesgarani

Single-channel, speaker-independent speech separation methods have recently seen great progress. However, the accuracy, latency, and computational cost of such methods remain insufficient... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK USES EXTRA
TRAINING DATA
RESULT LEADERBOARD
Music Source Separation MUSDB18 Conv-TasNet (extra) SDR (vocals) 6.74 # 4
SDR (drums) 7.11 # 1
SDR (other) 4.44 # 2
SDR (bass) 7.00 # 1
Music Source Separation MUSDB18 Conv-TasNet SDR (vocals) 6.81 # 3
SDR (drums) 6.08 # 4
SDR (other) 4.37 # 3
SDR (bass) 5.66 # 4
Speech Separation wsj0-2mix Conv-TasNet SI-SDR 15.3 # 5