Speech Reconstruction on Gigaspeech noisy speech 2021 (OOD)
9.7WERGround Truth
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Ground Truth2026.01 | 9.7 | 5.1 | 2.84 | — | — | — | |
| PAST 1:82026.01 | 10.9 | 6 | 3.09 | 0.98 | 0.89 | 0.07 | |
| DualCodec 1:8trained on relevant data=true2026.01 | 11 | 6 | 3.11 | 0.98 | 0.84 | 0.08 | |
| Kanade 25Hz2026.01 | 11.3 | 6.2 | 3.27 | 0.96 | 0.81 | 0.09 | |
| X-Codec 2trained on relevant data=true2026.01 | 11.5 | 6.3 | 2.99 | 0.97 | 0.87 | 0.08 | |
| BiCodectrained on relevant data=true2026.01 | 11.9 | 6.6 | 3.07 | 0.96 | 0.87 | 0.08 | |
| Kanade 12.5Hz2026.01 | 16.2 | 9.3 | 3.25 | 0.95 | 0.74 | 0.13 | |
| StableCodectrained on relevant data=true2026.01 | 27.1 | 16.3 | 3.51 | 0.9 | 0.84 | 0.09 | |
| WavTokenizer2026.01 | 33.9 | 21.9 | 2.64 | 0.88 | 0.82 | 0.1 |