Speech Reconstruction on Japanese Versatile Speech unseen language speech 2019 (OOD)
4.6WERGround Truth
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Ground Truth2026.01 | 4.6 | 2.5 | 3.63 | — | — | — | |
| DualCodec 1:8trained on relevant data=true2026.01 | 5 | 2.8 | 3.67 | 0.99 | 0.81 | 0.09 | |
| PAST 1:82026.01 | 5.2 | 2.8 | 3.62 | 0.98 | 0.88 | 0.09 | |
| X-Codec 2trained on relevant data=true2026.01 | 5.4 | 2.9 | 3.59 | 0.98 | 0.89 | 0.1 | |
| Kanade 25Hz2026.01 | 5.6 | 3 | 3.72 | 0.97 | 0.84 | 0.17 | |
| BiCodec2026.01 | 5.7 | 3.1 | 3.73 | 0.98 | 0.86 | 0.1 | |
| Kanade 12.5Hz2026.01 | 12.2 | 7.2 | 3.77 | 0.94 | 0.7 | 0.21 | |
| WavTokenizer2026.01 | 18.2 | 11.3 | 2.92 | 0.88 | 0.82 | 0.14 | |
| StableCodec2026.01 | 25 | 16.5 | 3.83 | 0.91 | 0.9 | 0.1 |