Audiovisual Synchronization on Seamless Interaction Teacher-forced
8.53LSE-DReconstructed face
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Reconstructed faceStreaming=false2026.06 | 8.53 | 12 | |
| Moshi-FaceStreaming=true2026.06 | 8.76 | 14 | |
| Moshi-Face w/o Face Transformer pre-trainingStreaming=true2026.06 | 9.53 | 13 | |
| Moshi-Face w/o t−1 face token inputStreaming=true2026.06 | 11.3 | 15 | |
| Random faceStreaming=false2026.06 | 11.7 | 13 | |
| Moshi-Face w/o full fine-tuningStreaming=true2026.06 | 11.8 | 16 |