Conversational Intelligence on VoiceBench
4.57AlpacaEvalDSM-ASR+ Qwen2-7B-Instruct
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| DSM-ASR+ Qwen2-7B-InstructArchitecture=Naive pipeline2026.03 | 4.57 | 3.87 | 3.73 | 46.42 | 58.52 | 72.09 | 58.1 | 51.29 | 97.12 | 69.66 | |
| DuplexCascade2026.03 | 4.4 | 3.64 | 3.56 | 45.56 | 52.86 | 56.04 | 59.8 | 43.38 | 99.04 | 65.41 | |
| DuplexCascade-βCapability=Enables backchannels2026.03 | 4.38 | 3.7 | 3.53 | 46.35 | 52.84 | 58.46 | 59 | 44.45 | 99.03 | 65.81 | |
| Freeze-OmniData source=Results directly obtained from original paper2026.03 | 4.03 | 3.46 | 3.15 | 53.45 | 28.14 | 30.98 | 50.7 | 23.4 | 97.3 | 55.2 | |
| PersonaPlexData source=Inferred from official checkpoints2026.03 | 2.69 | 2.25 | 1.95 | 18.77 | 24.92 | 24.4 | 49.4 | 11.95 | 8.07 | 30.59 | |
| MoshiData source=Results directly obtained from original paper2026.03 | 2.01 | 1.6 | 1.3 | 15.64 | 24.04 | 25.93 | 47.4 | 10.12 | 44.23 | 29.51 |