Automatic Speech Recognition on LibriReplay-DOA (test)
10.4WERFastMNMF
Evaluation Results
| Method | Links | |
|---|---|---|
| FastMNMFTraining Strategy=NT, Speaker angle=90°, Overlap ratio=0–25%2026.06 | 10.4 | |
| FastMNMFTraining Strategy=NT, Speaker angle=120°, Overlap ratio=0–25%2026.06 | 12.7 | |
| FastMNMFTraining Strategy=NT, Speaker angle=45°, Overlap ratio=75–100%2026.06 | 12.9 | |
| PATSETraining Strategy=PT+FT, Speaker angle=Overall2026.06 | 14 | |
| FastMNMFTraining Strategy=NT, Speaker angle=120°, Overlap ratio=75–100%2026.06 | 14.1 | |
| FastMNMFTraining Strategy=NT, Speaker angle=45°, Overlap ratio=0–25%2026.06 | 14.3 | |
| FastMNMFTraining Strategy=NT, Speaker angle=120°, Overlap ratio=25–50%2026.06 | 14.7 | |
| FastMNMFTraining Strategy=NT, Speaker angle=45°, Overlap ratio=50–75%2026.06 | 15 | |
| PATSETraining Strategy=PT+FT, Speaker angle=15°, Overlap ratio=0–25%2026.06 | 15.1 | |
| PATSETraining Strategy=PT+FT, Speaker angle=15°, Overlap ratio=25–50%2026.06 | 15.3 | |
| PATSETraining Strategy=Scratch, Speaker angle=15°, Overlap ratio=0–25%2026.06 | 15.6 | |
| FastMNMFTraining Strategy=NT, Speaker angle=120°, Overlap ratio=50–75%2026.06 | 15.8 | |
| FastMNMFTraining Strategy=NT, Speaker angle=45°, Overlap ratio=25–50%2026.06 | 16.5 | |
| PATSETraining Strategy=Scratch, Speaker angle=Overall2026.06 | 18.9 | |
| FastMNMFTraining Strategy=NT, Speaker angle=90°, Overlap ratio=75–100%2026.06 | 19.4 | |
| PATSETraining Strategy=Scratch, Speaker angle=15°, Overlap ratio=25–50%2026.06 | 19.6 | |
| CSS (TIGER)Training Strategy=PT+FT, Speaker angle=15°, Overlap ratio=0–25%2026.06 | 19.9 | |
| FastMNMFTraining Strategy=NT, Speaker angle=90°, Overlap ratio=50–75%2026.06 | 20.1 | |
| FastMNMFTraining Strategy=NT, Speaker angle=Overall2026.06 | 21.1 | |
| PATSETraining Strategy=PT+FT, Speaker angle=15°, Overlap ratio=50–75%2026.06 | 21.4 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=120°, Overlap ratio=0–25%2026.06 | 22 | |
| FastMNMFTraining Strategy=NT, Speaker angle=90°, Overlap ratio=25–50%2026.06 | 22.2 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=90°, Overlap ratio=0–25%2026.06 | 22.3 | |
| PATSETraining Strategy=PT+FT, Speaker angle=15°, Overlap ratio=75–100%2026.06 | 22.8 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=45°, Overlap ratio=25–50%2026.06 | 23.1 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=90°, Overlap ratio=25–50%2026.06 | 24.4 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=120°, Overlap ratio=25–50%2026.06 | 24.7 | |
| PATSETraining Strategy=Scratch, Speaker angle=15°, Overlap ratio=50–75%2026.06 | 24.7 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=45°, Overlap ratio=0–25%2026.06 | 25.1 | |
| CSS (TIGER)Training Strategy=Scratch, Speaker angle=15°, Overlap ratio=0–25%2026.06 | 25.8 | |
| CSS (TIGER)Training Strategy=PT+FT, Speaker angle=15°, Overlap ratio=25–50%2026.06 | 25.9 | |
| DSB + GateTraining Strategy=NT, Speaker angle=120°, Overlap ratio=0–25%2026.06 | 27.3 | |
| PATSETraining Strategy=Scratch, Speaker angle=15°, Overlap ratio=75–100%2026.06 | 28 | |
| DSB + GateTraining Strategy=NT, Speaker angle=90°, Overlap ratio=0–25%2026.06 | 28.3 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=15°, Overlap ratio=0–25%2026.06 | 28.3 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=15°, Overlap ratio=25–50%2026.06 | 28.9 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=120°, Overlap ratio=0–25%2026.06 | 30.2 | |
| FastMNMFTraining Strategy=NT, Speaker angle=15°, Overlap ratio=0–25%2026.06 | 31.3 | |
| DSB + GateTraining Strategy=NT, Speaker angle=Overall2026.06 | 31.5 | |
| CSS (TIGER)Training Strategy=Scratch, Speaker angle=15°, Overlap ratio=25–50%2026.06 | 32.5 | |
| CSS (TIGER)Training Strategy=PT+FT, Speaker angle=Overall2026.06 | 32.8 | |
| DSB + GateTraining Strategy=NT, Speaker angle=45°, Overlap ratio=0–25%2026.06 | 33.3 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=120°, Overlap ratio=25–50%2026.06 | 34.2 | |
| FastMNMFTraining Strategy=NT, Speaker angle=15°, Overlap ratio=75–100%2026.06 | 34.7 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=90°, Overlap ratio=0–25%2026.06 | 34.9 | |
| CSS (TIGER)Training Strategy=Scratch, Speaker angle=Overall2026.06 | 36.3 | |
| DSB + GateTraining Strategy=NT, Speaker angle=15°, Overlap ratio=0–25%2026.06 | 37.1 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=Overall2026.06 | 38.4 | |
| FastMNMFTraining Strategy=NT, Speaker angle=15°, Overlap ratio=25–50%2026.06 | 39.4 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=45°, Overlap ratio=25–50%2026.06 | 39.7 | |
| CSS (TIGER)Training Strategy=PT+FT, Speaker angle=15°, Overlap ratio=50–75%2026.06 | 39.8 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=90°, Overlap ratio=25–50%2026.06 | 40.4 | |
| FastMNMFTraining Strategy=NT, Speaker angle=15°, Overlap ratio=50–75%2026.06 | 40.9 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=120°, Overlap ratio=50–75%2026.06 | 41.5 | |
| CSS (TIGER)Training Strategy=Scratch, Speaker angle=15°, Overlap ratio=50–75%2026.06 | 41.6 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=45°, Overlap ratio=50–75%2026.06 | 42.4 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=45°, Overlap ratio=0–25%2026.06 | 42.6 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=90°, Overlap ratio=50–75%2026.06 | 44.2 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=15°, Overlap ratio=50–75%2026.06 | 45.5 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=15°, Overlap ratio=0–25%2026.06 | 46 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=120°, Overlap ratio=75–100%2026.06 | 48.6 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=120°, Overlap ratio=50–75%2026.06 | 48.6 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=120°, Overlap ratio=75–100%2026.06 | 48.9 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=45°, Overlap ratio=75–100%2026.06 | 49.1 | |
| CSS (TIGER)Training Strategy=PT+FT, Speaker angle=15°, Overlap ratio=75–100%2026.06 | 49.1 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=Overall2026.06 | 49.5 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=15°, Overlap ratio=25–50%2026.06 | 49.9 | |
| CSS (TIGER)Training Strategy=Scratch, Speaker angle=15°, Overlap ratio=75–100%2026.06 | 50.5 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=90°, Overlap ratio=75–100%2026.06 | 51.2 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=90°, Overlap ratio=50–75%2026.06 | 56.1 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=45°, Overlap ratio=50–75%2026.06 | 56.2 | |
| Sortformer + GSSTraining Strategy=NT, Speaker angle=15°, Overlap ratio=75–100%2026.06 | 56.5 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=45°, Overlap ratio=75–100%2026.06 | 57 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=90°, Overlap ratio=75–100%2026.06 | 57.3 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=15°, Overlap ratio=50–75%2026.06 | 60.9 | |
| CSS (FasNet-TAC)Training Strategy=Scratch, Speaker angle=15°, Overlap ratio=75–100%2026.06 | 61.6 |