MossFormer: Pushing the Performance Limit of Monaural Speech Separation using Gated Single-Head Transformer with Convolution-Augmented Joint Self-Attentions

Benchmark Model Rank Results
speech-separation-on-wsj0-3mixMossFormer (L) + DM#2SI-SDRi: 21.2
speech-separation-on-wsj0-3mixMossFormer (M) + DM#3SI-SDRi: 20.8
speech-separation-on-whamMossFormer (L) + DM#3SI-SDRi: 17.3
speech-separation-on-whamrMossFormer (L) + DM#5SI-SDRi: 16.3
speech-separation-on-wsj0-2mixMossFormer (L) + DM#7SI-SDRi: 22.8Number of parameters (M): 42.1MACs (G): 86.1
speech-separation-on-wsj0-2mixMossFormer (M) + DM#11SI-SDRi: 22.5