Single-channel Speech Separation on WSJ0-2mix v1 (test)
19.1P-SI-SNR (Pref=-30dB)Multi-Decoder DPRNN
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Multi-Decoder DPRNNPer-speaker training=false2020.11 | 19.1 | 19.1 | |
| Model-Select(Mulcat)Per-speaker training=true, Backbone=Mulcat2020.11 | 17.5 | 18.1 | |
| Model-Select(DPRNN)Per-speaker training=true, Backbone=DPRNN2020.11 | 15.2 | 15.9 | |
| Attractor Network2020.11 | 14.7 | 14.9 | |
| OR-PITRecall=95.7%, Aggregation=Average of 2 and 3 speakers2020.11 | 13.1 | 13.4 |