Full-duplex speech interaction on Full-Duplex-Bench v1
93.4TOR (Synthetic Pause)dGSLM
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| dGSLM2026.06 | 93.4 | 93.5 | 69.1 | 1.5 | 93.4 | 97.5 | 0.352 | 91.7 | 0.201 | 2.531 | |
| Freeze-Omni2026.06 | 64.2 | 48.1 | 63.6 | 0.1 | 99.7 | 33.6 | 0.953 | 86.7 | 3.615 | 1.409 | |
| ASPIRin2026.06 | 48.2 | 48.6 | — | — | — | 76.5 | 0.273 | 94.1 | 3.734 | 0.992 | |
| PersonaPlex2026.06 | 48.2 | 44.4 | 18.2 | 4.6 | 84.1 | 95.8 | 0.219 | 94 | 4.5 | 0.271 | |
| Moshi2026.06 | 44.5 | 52.8 | 25.5 | 7.4 | 82.4 | 73.9 | 0.162 | 92 | 3.44 | 1.377 | |
| PersonaPlex + RL (Seamless)Base Model=PersonaPlex, Training Data=Seamless2026.06 | 35 | 35.6 | 7.3 | 11.2 | 78.6 | 97.5 | 0.086 | 99.5 | 4.533 | 0.223 | |
| PersonaPlex + RL (Fisher)Base Model=PersonaPlex, Training Data=Fisher2026.06 | 32.8 | 36.1 | 12.7 | 12.2 | 78.3 | 95 | 0.079 | 100 | 4.52 | 0.187 | |
| Moshi + RL (Seamless)Base Model=Moshi, Training Data=Seamless2026.06 | 30.7 | 46.3 | 14.5 | 10.1 | 79.4 | 95.8 | 0.16 | 100 | 3.63 | 0.409 | |
| Moshi + RL (Fisher)Base Model=Moshi, Training Data=Fisher2026.06 | 22.6 | 41.7 | 9.1 | 9.5 | 78.9 | 96.6 | 0.121 | 100 | 3.575 | 0.461 |