Audio-Text Alignment on TEDlium v2 (dev and test)
0.31Mean DeviationCTC-Segmentation (Hybrid CTC/att. Transformer)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CTC-Segmentation (Hybrid CTC/att. Transformer)Architecture=Hybrid CTC/att. Transformer, Training Data=TEDlium v22020.07 | 0.31 | 0.85 | 88.8 | |
| CTC-Segmentation (Hybrid CTC/att. BLSTM)Architecture=Hybrid CTC/att. BLSTM, Training Data=TEDlium v22020.07 | 0.34 | 1.16 | 90.1 | |
| CTC-Segmentation (Hybrid CTC/att. Transformer)Architecture=Hybrid CTC/att. Transformer, Training Data=Librispeech2020.07 | 0.35 | 0.68 | 85.1 | |
| GentleAlignment Framework=HMM-based using kaldi2020.07 | 0.41 | 1.97 | 82 | |
| MAUSAlignment Framework=HMM-based using HTK2020.07 | 1.38 | 11.62 | 74.1 | |
| AeneasAlignment Framework=DTW-based2020.07 | 9.01 | 38.47 | 64.7 |