Question Answering on Llama Questions
75.3AccuracyDuplexPO
Evaluation Results
| Method | Links | |
|---|---|---|
| DuplexPOFull Duplex=true2026.07 | 75.3 | |
| Baichuan-AudioFull Duplex=false2026.07 | 74 | |
| SALMONN-omniFull Duplex=true2026.07 | 73.6 | |
| SFT BaselineFull Duplex=true2026.07 | 72 | |
| STITCH-RFull Duplex=false2026.07 | 70 | |
| Qwen2-AudioFull Duplex=false2026.07 | 69.7 | |
| Kimi-AudioFull Duplex=false2026.07 | 68.3 | |
| GLM-4-VoiceFull Duplex=false2026.07 | 65.7 | |
| Freeze-OmniFull Duplex=true2026.07 | 56.2 | |
| MoshiFull Duplex=true2026.07 | 54.5 | |
| SALM-DuplexFull Duplex=true2026.07 | 51.3 | |
| GLM-4-VoiceModality=S→S, # Params=9B, Full-duplex=✗2025.10 | 0.507 | |
| CT-Duplex w/ thkModality=S→S, # Params=1.7B, Full-duplex=✓2025.10 | 0.314 | |
| CT-Duplex w/o thkModality=S→S, # Params=1.7B, Full-duplex=✓2025.10 | 0.304 | |
| SpectronModality=S→T, # Params=1B, Full-duplex=✗2025.10 | 0.219 | |
| SpeechGPTModality=S→T, # Params=7B, Full-duplex=✗2025.10 | 0.216 | |
| MoshiModality=S→S, # Params=7B, Full-duplex=✓2025.10 | 0.21 | |
| SALM-Duplex*Modality=S→S, # Params=1.5B, Full-duplex=✓2025.10 | 0.15 | |
| TWISTModality=S→S, # Params=7B, Full-duplex=✗2025.10 | 0.04 |