Spoken Question Answering on LLaMAQ
74.2Text ScoreFreeze-Omni
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Freeze-OmniBackbone LLM=Qwen2-7B-Instruct, Full-Duplex=true2026.05 | 74.2 | 56.2 | |
| GLM-4-VoiceBackbone LLM=GLM-4-9B-Base, Full-Duplex=false2026.05 | 64.7 | 50.7 | |
| MoshiBackbone LLM=Helium (7B), Full-Duplex=true2026.05 | 60.8 | 54.5 | |
| CF-DuplexBackbone LLM=Qwen3-1.7B, Full-Duplex=true2026.05 | 57.3 | 50.7 | |
| XA-DuplexBackbone LLM=Qwen3-1.7B, Full-Duplex=true2026.05 | 40.3 | 38.3 | |
| Speech-GPTBackbone LLM=LLaMA-13B, Full-Duplex=false2026.05 | 21 | — | |
| SALM-DuplexBackbone LLM=TinyLlama-1.1B-chat, Full-Duplex=true2026.05 | — | 51.3 |