Phoneme Recognition on TIMIT (dev)
7.4PERwav2vec 2.0
Evaluation results
| Method | Links | |
|---|---|---|
| wav2vec 2.0Language Model=none, Model size=Large, Pre-training=LS-9602020.06 | 7.4 | |
| Masked-VPCEvaluation protocol=Fine-Tuning2025.12 | 9.5 | |
| vq-wav2vec2020.06 | 9.6 | |
| vq-wav2vec + BERT smallQuantization Method=Gumbel, Architecture=CNN-8L-PReLU-do0.7, BERT Pre-training=BERT small2019.10 | 9.64 | |
| vq-wav2vec + BERT smallQuantization Method=k-means, Architecture=CNN-8L-PReLU-do0.7, BERT Pre-training=BERT small2019.10 | 9.8 | |
| HuBERT ObjEvaluation protocol=Fine-Tuning2025.12 | 10.2 | |
| Masked-NCEEvaluation protocol=Fine-Tuning2025.12 | 11.2 | |
| CNN+BLSTMFeature=Raw-Wav, Architecture=CNN+BLSTM2026.06 | 11.3 | |
| SincNet+BLSTMFeature=Raw-Wav, Architecture=SincNet+BLSTM2026.06 | 11.3 | |
| FBank-83-WSJFeature=FBank-83-WSJ, Architecture=Best System in [2]2026.06 | 11.5 | |
| Sinc2Net+BLSTMFeature=Raw-Wav, Architecture=Sinc2Net+BLSTM2026.06 | 11.5 | |
| FBank-83Architecture=Best System in [2]2026.06 | 12.8 | |
| wav2vec2019.10 | 12.9 | |
| wav2vec2020.06 | 12.9 | |
| HMM over Time and Frequency Convolutional Net2015.06 | 13.9 | |
| CNN+BLSTMFeature=Raw-Wav–Proposed2026.06 | 13.9 | |
| Sinc2Net+BLSTMFeature=Raw-Wav–Proposed2026.06 | 13.9 | |
| SincNet+BLSTMFeature=Raw-Wav–Proposed2026.06 | 14.2 | |
| Raw-WavArchitecture=CNN2026.06 | 14.9 | |
| Raw-WavArchitecture=ParzNet2026.06 | 15 | |
| Raw-WavArchitecture=CGCNN2026.06 | 15.2 | |
| vq-wav2vecQuantization Method=Gumbel, Architecture=CNN-8L-PReLU-do0.7, BERT Pre-training=None2019.10 | 15.34 | |
| CNN + TD-filterbanksReference=Zeghidour et al., 2018a2019.04 | 15.6 | |
| CNN + TD-filterbanks2019.10 | 15.6 | |
| CNN + TD-filterbanks2020.06 | 15.6 | |
| vq-wav2vecQuantization Method=k-means, Architecture=CNN-8L-PReLU-do0.7, BERT Pre-training=None2019.10 | 15.65 | |
| Baseline + Conv. Features + Smooth FocusConvolutional Attention Features=true, Smooth Focus=true2015.06 | 15.8 | |
| Baseline Model2015.06 | 15.9 | |
| Baseline + Conv. FeaturesConvolutional Attention Features=true2015.06 | 16.1 | |
| Baseline (log-mel)2019.10 | 16.9 | |
| Raw-Wav (E2E)Architecture=SincNet2026.06 | 17.3 | |
| Raw-Wav (E2E)Architecture=CNN2026.06 | 18.9 | |
| 12L TransformerEvaluation protocol=Baseline2025.12 | 22.1 |