Audio Perception on AIR-Bench Foundation
63.2Speech PerformanceQwen2.5-Omni + TCD
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Qwen2.5-Omni + TCDdecoding_method=Temporal Contrastive Decoding2026.04 | 63.2 | 74.5 | 66.7 | |
| Qwen2.5-Omni2026.04 | 61.8 | 71.6 | 64.8 | |
| Qwen-Audio2026.04 | 58.7 | 60.2 | 59.1 | |
| Whisper + GPT-42026.04 | 53.6 | — | — | |
| PandaGPT2026.04 | 39 | 43.6 | 40.4 | |
| SALMONN2026.04 | 37.8 | 33 | 36.3 | |
| BLSP2026.04 | 36.6 | 31.4 | 35 | |
| SpeechGPT2026.04 | 34.3 | 27.5 | 32.2 | |
| Next-GPT2026.04 | 33.6 | 32.2 | 33.1 |