Table 5 Important configurations for WSJ experiments
From: A new joint CTC-attention-based speech recognition model with multi-level multi-head attention
System | CTC encoder layers | Attention encoder layers | Decoder layers | MTL λ |
|---|---|---|---|---|
K1 | 2 | 3 | 1 | 0.2 |
K2 | 2 | 3 | 1 | 0.1 |
K3 | 3 | 4 | 2 | 0.1 |
K4 | 3 | 4 | 2 | 0.1 |