Speech-to-speech instruction-following on InstructS2S Eval
3.47S2S AccLLAMA-Omni
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LLAMA-Omnimode=Zero-shot2026.02 | 3.47 | 3.99 | |
| SpeechGPTmode=Zero-shot2026.02 | 2.19 | 2.98 | |
| UniAudio 2.0 (Ours 3B)mode=Zero-shot, parameters=3B2026.02 | 2.16 | 3.66 | |
| Ablation (Ours 1B)mode=Zero-shot, parameters=1B2026.02 | 1.12 | 1.41 |