Speech Reconstruction on Salmon Sentiment Consistency emotional 2025b (OOD)
2.9WERGround Truth
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Ground Truthvariation=standard2026.01 | 2.9 | 1 | 3.79 | — | — | — | |
| PAST 1:8variation=standard2026.01 | 3 | 1 | 3.91 | 0.99 | 0.85 | 0.09 | |
| DualCodec 1:8trained on relevant data=true, variation=standard2026.01 | 3.6 | 1.1 | 3.91 | 0.98 | 0.88 | 0.08 | |
| X-Codec 2trained on relevant data=true, variation=standard2026.01 | 3.8 | 1.2 | 3.77 | 0.97 | 0.85 | 0.09 | |
| PAST 1:8variation=w/ change2026.01 | 4.2 | 1.7 | 3.77 | 0.98 | 0.9 | 0.08 | |
| DualCodec 1:8trained on relevant data=true, variation=w/ change2026.01 | 4.4 | 1.8 | 3.76 | 0.98 | 0.9 | 0.1 | |
| Kanade 25Hzvariation=standard2026.01 | 4.4 | 1.5 | 3.85 | 0.96 | 0.73 | 0.16 | |
| Kanade 25Hzvariation=w/ change2026.01 | 4.7 | 1.9 | 3.88 | 0.96 | 0.75 | 0.18 | |
| Ground Truthvariation=w/ change2026.01 | 4.9 | 1.6 | 3.62 | — | — | — | |
| BiCodectrained on relevant data=true, variation=standard2026.01 | 5.4 | 1.7 | 3.84 | 0.98 | 0.81 | 0.1 | |
| X-Codec 2trained on relevant data=true, variation=w/ change2026.01 | 5.7 | 2.2 | 3.67 | 0.97 | 0.89 | 0.11 | |
| BiCodectrained on relevant data=true, variation=w/ change2026.01 | 6 | 2.6 | 3.73 | 0.97 | 0.9 | 0.11 | |
| Kanade 12.5Hzvariation=standard2026.01 | 6.4 | 2.3 | 3.83 | 0.95 | 0.66 | 0.19 | |
| Kanade 12.5Hzvariation=w/ change2026.01 | 7 | 3.1 | 3.83 | 0.94 | 0.67 | 0.22 | |
| WavTokenizervariation=standard2026.01 | 14.5 | 7.7 | 3.21 | 0.9 | 0.74 | 0.12 | |
| StableCodectrained on relevant data=true, variation=standard2026.01 | 14.8 | 7.2 | 4.08 | 0.93 | 0.81 | 0.12 | |
| WavTokenizervariation=w/ change2026.01 | 17.5 | 9.7 | 3.13 | 0.9 | 0.82 | 0.16 | |
| StableCodectrained on relevant data=true, variation=w/ change2026.01 | 18 | 9.3 | 4.03 | 0.92 | 0.84 | 0.12 |