Automatic Speech Recognition on Chinese Dialect Scenario Aggregate
12Hallucination RateQwen3-ASR-1.7B
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen3-ASR-1.7BDecoding Mode=offline2026.04 | 12 | |
| OursDecoding Mode=offline, Beam Size=3, IA-SFT=true2026.04 | 12.2 | |
| Step-Audio2-miniDecoding Mode=offline2026.04 | 19.4 | |
| Ours (w/o IA-SFT)Decoding Mode=offline, Beam Size=3, IA-SFT=false2026.04 | 19.8 | |
| GLM-ASR-nanoDecoding Mode=offline2026.04 | 20.1 | |
| Fun-ASR-nanoDecoding Mode=offline2026.04 | 21.7 | |
| FireRedASR-LLMDecoding Mode=offline2026.04 | 22.8 | |
| Qwen3-Omni-InstDecoding Mode=offline2026.04 | 37 |