Speech Processing on SUPERB (PER/WER focus)
3.21PERDinoSR
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DinoSRPre-training dataset=LibriSpeech 960h, Implementation=Original2025.12 | 3.21 | 4.71 | |
| DinoSRPre-training dataset=LibriSpeech 960h, Implementation=Our re-implementation2025.12 | 3.54 | 5.01 | |
| data2vec 2.0Pre-training dataset=LibriSpeech 960h, Implementation=Original2025.12 | 3.93 | 4.91 | |
| MR-HuBERT mono-basePre-training dataset=LibriSpeech 960h, Implementation=Original2025.12 | 4.16 | 5.76 | |
| data2vecPre-training dataset=LibriSpeech 960h, Implementation=Original2025.12 | 4.69 | 4.94 | |
| WavLM BasePre-training dataset=LibriSpeech 960h, Implementation=Original2025.12 | 4.84 | 6.21 | |
| SpidRPre-training dataset=LibriSpeech 960h2025.12 | 4.86 | 7.47 | |
| HuBERTPre-training dataset=LibriSpeech 960h, Implementation=Original2025.12 | 5.41 | 6.42 | |
| wav2vec 2.0Pre-training dataset=LibriSpeech 960h, Implementation=Original2025.12 | 5.74 | 6.43 |