TF-Locoformer: Transformer with Local Modeling by Convolution for Speech Separation and Enhancement

Benchmark Model Rank Results
speech-enhancement-on-deep-noise-suppressionTF-Locoformer (M)#1PESQ-WB: 3.72SI-SDR-WB: 23.3STOI: 98.8Number of parameters (M): 15
speech-separation-on-libri2mixTF-Locoformer (M)#2SI-SDRi: 22.1SDRi: 22.2Number of parameters (M): 15
speech-separation-on-whamrTF-Locoformer (M)#1SI-SDRi: 18.5Number of parameters (M): 15SDRi: 16.9
speech-separation-on-whamrTF-Locoformer (S)#2SI-SDRi: 17.4Number of parameters (M): 5SDRi: 15.9
speech-separation-on-wsj0-2mixTF-Locoformer (L) + DM#2SI-SDRi: 25.1SDRi: 25.2Number of parameters (M): 22.5
speech-separation-on-wsj0-2mixTF-Locoformer (M) + DM#3SI-SDRi: 24.6SDRi: 24.7Number of parameters (M): 15.0
speech-separation-on-wsj0-2mixTF-Locoformer (L)#4SI-SDRi: 24.2SDRi: 24.3Number of parameters (M): 22.5
speech-separation-on-wsj0-2mixTF-Locoformer (M)#6SI-SDRi: 23.6SDRi: 23.8Number of parameters (M): 15.0
speech-separation-on-wsj0-2mixTF-Locoformer (S) + DM#8SI-SDRi: 22.8SDRi: 23Number of parameters (M): 5.0
speech-separation-on-wsj0-2mixTF-Locoformer (S)#14SI-SDRi: 22SDRi: 22.1Number of parameters (M): 5.0