Overlap Handling Evaluation on Full-Duplex-Bench Talking to Other v1.5
0.96STOIMoshi (Fine-tuned)
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Moshi (Fine-tuned)Model version=Fine-tuned, Training Data=Sommelier2026.03 | 0.96 | 3.3 | 20.26 | 2.3 | 146.4 | 96.4 | 17 | -51 | 16.3 | 18 | 16 | 0 | 0.63 | |
| Moshi (Base)Model version=Base2026.03 | 0.9 | 2.55 | 12.64 | 2.34 | 124.5 | 96.6 | 16 | -49.1 | 16.2 | 15 | 18 | 4 | 0.63 |