Speech Recognition on LibriSpeech (dev)
1.29WERFAdam
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| FAdamArchitecture=Conformer (600M), Pre-training=w2v-BERT, Fine-tuning protocol=Semi-supervised (LibriLight), Optimizer=FAdam2024.05 | 1.29 | — | — | |
| Adam (w2v-BERT paper)Architecture=Conformer (600M), Pre-training=w2v-BERT, Optimizer=Adam, Source=[53]2024.05 | 1.3 | — | — | |
| AdamArchitecture=Conformer (600M), Pre-training=w2v-BERT, Fine-tuning protocol=Semi-supervised (LibriLight), Optimizer=Adam2024.05 | 1.3 | — | — | |
| Gen3 Conformer XXLUnlabeled data (hrs)=60k, LM Fusion=No2020.10 | 1.3 | — | — | |
| Gen3 Conformer XXLUnlabeled data (hrs)=60k, LM Fusion=Yes2020.10 | 1.3 | — | — | |
| Gen3 Conformer XXL+Unlabeled data (hrs)=60k, LM Fusion=No2020.10 | 1.3 | — | — | |
| Gen3 Conformer XXL+Unlabeled data (hrs)=60k, LM Fusion=Yes2020.10 | 1.3 | — | — | |
| Pre-trained Conformer XXLUnlabeled data (hrs)=60k, LM Fusion=Yes2020.10 | 1.5 | — | — | |
| Gen4 ContextNetUnlabeled data (hrs)=60k, LM Fusion=No2020.10 | 1.6 | — | — | |
| Gen4 ContextNetUnlabeled data (hrs)=60k, LM Fusion=Yes2020.10 | 1.6 | — | — | |
| Gen4 Conformer LUnlabeled data (hrs)=60k, LM Fusion=No2020.10 | 1.6 | — | — | |
| Gen4 Conformer LUnlabeled data (hrs)=60k, LM Fusion=Yes2020.10 | 1.6 | — | — | |
| Pre-trained CTCUnlabeled data (hrs)=60k, LM Fusion=Yes2020.10 | 1.6 | — | — | |
| Pre-trained Conformer XLUnlabeled data (hrs)=60k, LM Fusion=Yes2020.10 | 1.6 | — | — | |
| Pre-trained Conformer XXLUnlabeled data (hrs)=60k, LM Fusion=No2020.10 | 1.6 | — | — | |
| Pre-trained Conformer XLUnlabeled data (hrs)=60k, LM Fusion=No2020.10 | 1.7 | — | — | |
| Conformer LUnlabeled data (hrs)=None, LM Fusion=No2020.10 | 1.9 | — | — | |
| Pre-trained CTCUnlabeled data (hrs)=60k, LM Fusion=No2020.10 | 2.1 | — | — | |
| ASR + LMSystem=ASR + LM2025.12 | 3.35 | — | — | |
| ASR + DLMSystem=ASR + DLM2025.12 | 3.4 | — | — | |
| GLAD-SOTParm.(M)=35.182026.07 | 3.5 | — | — | |
| SOT-SACTCParm.(M)=35.772026.07 | 3.6 | — | — | |
| H-SAGEParm.(M)=35.752026.07 | 3.6 | — | — | |
| SOT + Local MoLEParm.(M)=35.092026.07 | 3.7 | — | — | |
| ASR onlySystem=ASR only2025.12 | 4.08 | — | — | |
| SOTParm.(M)=36.072026.07 | 4.1 | — | — | |
| B2T connectionLayer configuration=Enc-Dec: 6L-6L2022.06 | — | 3.69 | 8.97 | |
| B2T connectionLayer configuration=Enc-Dec: 12L-6L2022.06 | — | 3.26 | 7.74 | |
| Centaurus Base (full SSM)Protocol=Online (streaming), no attention, Parameters (M)=12.4, FLOPs (G)=20.62025.01 | — | 5.9 | 13.1 | |
| Centaurus with causal conv-blockProtocol=Online (streaming), no attention, Parameters (M)=23.6, FLOPs (G)=43.72025.01 | — | 4.8 | 10.5 | |
| Centaurus with FFNProtocol=Online (streaming), no attention, Parameters (M)=18, FLOPs (G)=32.12025.01 | — | 5.2 | 11.3 | |
| Centaurus with Mamba macro-blockProtocol=Online (streaming), no attention, Parameters (M)=29.9, FLOPs (G)=46.92025.01 | — | 4.3 | 9.9 | |
| Full ConvProtocol=Offline (full-context)2025.01 | — | 3.1 | 9.9 | |
| Post-LNLayer configuration=Enc-Dec: 6L-6L2022.06 | — | 3.78 | 8.76 | |
| Pre-LNLayer configuration=Enc-Dec: 6L-6L2022.06 | — | 3.89 | 9.69 | |
| Pre-LNLayer configuration=Enc-Dec: 12L-6L2022.06 | — | 3.21 | 7.91 | |
| Wav2Vec2Protocol=Offline (full-context), Parameters (M)=94.4, FLOPs (G)=2852025.01 | — | 1.8 | 4.7 |