Automatic Speech Recognition on LS Clean
1.49WERREAD
Evaluation Results
| Method | Links | |
|---|---|---|
| READCombination Strategy=ROVER integrated with segment-level combination2026.06 | 1.49 | |
| ROVERStrategy=ROVER baseline2026.06 | 1.51 | |
| READCombination Strategy=Segment-level2026.06 | 1.66 | |
| NeMoCategory=Candidate Model2026.06 | 1.67 | |
| READCombination Strategy=Sentence-level2026.06 | 1.67 | |
| Qwen2.5-OmniCategory=Candidate Model2026.06 | 1.74 | |
| Qwen2-AudioType=AR2026.01 | 1.8 | |
| MinMoType=AR2026.01 | 1.8 | |
| LFM2-Audio-1.5B# Total Params=1.5B2025.11 | 2.03 | |
| Qwen-Audio2026.01 | 2.04 | |
| Phi-4-MultimodalType=AR2026.01 | 2.1 | |
| Whisper-large-v3-turbo# Total Params=1.5B2025.11 | 2.1 | |
| SpeechVerse2026.01 | 2.1 | |
| SALMONN2026.01 | 2.1 | |
| Qwen2.5-Omni-3B# Total Params=5B2025.11 | 2.14 | |
| Prompt Projectorprompt=best, tuning=LoRA2026.01 | 2.16 | |
| Whisper large-v3Category=Candidate Model2026.06 | 2.2 | |
| E2E-TTS-Stage2Training Stage=22026.06 | 2.22 | |
| w/o LLMAblation=Without LLM2026.06 | 2.23 | |
| E2E-TTS-Stage1Training Stage=12026.06 | 2.27 | |
| Q-Former2026.01 | 2.28 | |
| Prompt Projectorprompt=best, tuning=frozen2026.01 | 2.28 | |
| Prompt Projectorprompt=base, tuning=LoRA2026.01 | 2.33 | |
| Prompt Projectorprompt=base, tuning=frozen2026.01 | 2.34 | |
| Vanilla LLM-ASRprompt=best, tuning=LoRA2026.01 | 2.44 | |
| Vanilla LLM-ASRprompt=base, tuning=LoRA2026.01 | 2.46 | |
| Whisper-large-V32026.06 | 2.56 | |
| SLM2026.01 | 2.6 | |
| Vanilla LLM-ASRprompt=best, tuning=frozen2026.01 | 2.72 | |
| Whisper mediumCategory=Candidate Model2026.06 | 2.79 | |
| Vanilla LLM-ASRprompt=empty, tuning=frozen2026.01 | 2.84 | |
| w/o LLM and LFMAblation=Without LLM and LFM2026.06 | 2.96 | |
| DIFFUSPEECHType=Diff.2026.01 | 3 | |
| Vanilla LLM-ASRprompt=base, tuning=frozen2026.01 | 3.09 | |
| SenseVoice-small2026.06 | 3.15 | |
| Llama-Omni2Type=AR2026.01 | 3.5 | |
| MoshiType=AR2026.01 | 5.5 | |
| SpiritLMType=AR2026.01 | 6 | |
| SpeechGPTType=AR2026.01 | 11 |