Speech-to-Speech Translation on TAT-S2ST En-Hokkien (dev)
55.5ASR BLEUSynthetic target*
Evaluation results
| Method | Links | |
|---|---|---|
| Synthetic target*Human Training Data=False, Weakly Training Data=False2022.11 | 55.5 | |
| Two-pass decoding (UnitY)Human Training Data=35-hr, Weakly Training Data=1.5k-hr2022.11 | 7.8 | |
| Three-stageHuman Training Data=35-hr, Weakly Training Data=1.5k-hr2022.11 | 7.5 | |
| Two-stageHuman Training Data=35-hr, Weakly Training Data=1.5k-hr2022.11 | 7.1 | |
| Single-pass decodingHuman Training Data=35-hr, Weakly Training Data=1.5k-hr2022.11 | 6.6 | |
| Two-pass decoding (UnitY)Human Training Data=35-hr, Weakly Training Data=False2022.11 | 0.9 | |
| Single-pass decodingHuman Training Data=35-hr, Weakly Training Data=False2022.11 | 0.1 |