From: Segment boundary detection directed attention for online end-to-end speech recognition
Model | #Param (M) | PER (%) |
---|---|---|
Partial condition [12] | 3.1 | 20.8 |
Hard alignment with RL [14] | 6.8 | 20.5 |
Gaussian prediction attention [40] | 5.8 | 20.4 |
Hard monotonic attention [16] | 6.4 | 20.4 |
Stacked LSTM [15] | 1.0 | 20.0 |
CTC [41] | 3.8 | 19.6 |
Proposed method | 6.9 | 20.2 |
Soft attention* | 5.9 | 21.0 |
Soft attention bigger-E* | 7.5 | 20.8 |
Soft attention bigger-D* | 7.0 | 20.6 |