Paralinguistic speech understanding on AIR-Bench Speech (test)
71.45Emotion AccAZEROS
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| AZEROSSystem Paradigm=End-to-end Speech-LLM2025.12 | 71.45 | 86.75 | 61.3 | 84.8 | 73.6 | 85.6 | 77.25 | 8.28 | |
| DeSTA2.5System Paradigm=End-to-end Speech-LLM2025.12 | 64.3 | 84.24 | 65.6 | 97.3 | 65.2 | 83.7 | 76.72 | 7.57 | |
| Qwen3-Omni-30BSystem Paradigm=End-to-end Speech-LLM2025.12 | 62.2 | 91.11 | 36.9 | 97.7 | 80.4 | 90.7 | 76.5 | 7.85 | |
| Gemini2.5-proSystem Paradigm=End-to-end Speech-LLM2025.12 | 60.7 | 90.7 | 34.1 | 99.1 | 68.5 | 92.2 | 74.22 | 8.52 | |
| Whisper + GPT-4oSystem Paradigm=Cascaded System, ASR Module=Whisper-large-v3, LLM Backbone=GPT-4o2025.12 | 59.5 | 21.9 | 41.1 | 96.8 | 69.8 | 87.7 | 62.8 | 7.54 | |
| Qwen2.5-OmniSystem Paradigm=End-to-end Speech-LLM2025.12 | 54.85 | 89.76 | 44.8 | 89.7 | 79.7 | 88.6 | 74.57 | 6.97 | |
| Whisper + Qwen2.5System Paradigm=Cascaded System, ASR Module=Whisper-large-v3, LLM Backbone=Qwen2.5-7B-Instruct2025.12 | 50.8 | 28.36 | 36.4 | 88 | 73.6 | 82.7 | 59.98 | 7.34 | |
| GPT-4oSystem Paradigm=End-to-end Speech-LLM2025.12 | 49.1 | — | — | 76 | 61.6 | 85.8 | — | 7.53 | |
| Qwen2-AudioSystem Paradigm=End-to-end Speech-LLM2025.12 | 48.15 | 64.71 | 23.1 | 77.8 | 87 | 84.7 | 64.24 | 7.2 | |
| SALMONNSystem Paradigm=End-to-end Speech-LLM2025.12 | 29.9 | 35.5 | 48.7 | 28.1 | 51.7 | 36.7 | 38.43 | 6.16 | |
| GLM-4-VoiceSystem Paradigm=End-to-end Speech-LLM2025.12 | 22.95 | 23.91 | 18.7 | 25.4 | 27.9 | 21.1 | 23.33 | 5.53 |