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

Benchmark Model Rank Results
speech-enhancement-on-ears-whamConv-TasNet#4PESQ-WB: 2.31SI-SDR: 16.93ESTOI: 0.70SIGMOS: 2.69DNSMOS: 3.47…
speech-separation-on-wsj0-3mixConv-TasNet-gLN#7SI-SDRi: 12.7
music-source-separation-on-musdb18Conv-TasNet (extra)#12SDR (avg): 6.32SDR (vocals): 6.74SDR (drums): 7.11SDR (bass): 7.00
music-source-separation-on-musdb18Conv-TasNet#20SDR (avg): 5.73SDR (vocals): 6.81SDR (drums): 6.08SDR (bass): 5.66…
speech-separation-on-wsj0-2mixConv-TasNet#31SI-SDRi: 15.3SDRi: 15.6Number of parameters (M): 5.1