Automatic Speech Recognition on TED-LIUM 2 (test)
6.8WER (HAT)ConMamba (All-in-One)*
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ConMamba (All-in-One)*Encoder=ConMamba, Strategy=All-in-One, Dual-mode support=Yes, Number of parameters=117.9M2025.12 | 6.8 | — | 6.8 | 7.5 | |
| Conformer (Single)Encoder=Conformer, Strategy=Single-mode, Number of parameters=576.3M2025.12 | 6.9 | — | 7.1 | 7.6 | |
| ConMamba (Single)Encoder=ConMamba, Strategy=Single-mode, Number of parameters=574.3M2025.12 | 6.9 | — | 7.1 | 7.7 | |
| ConMamba (All-in-One w/o Dual)Encoder=ConMamba, Strategy=All-in-One, Dual-mode support=No, Number of parameters=235.8M2025.12 | 6.9 | — | 6.9 | 7.6 | |
| ESPnet RNNtoken=BPE2019.09 | — | 0.11 | — | — | |
| ESPnet Transformertoken=BPE2019.09 | — | 0.081 | — | — | |
| Kaldi (s5)token=BPE2019.09 | — | 0.09 | — | — |