Table 3 SER, error per class and average class error for the RNN classifier on the test partition for the best frontend configurations and different number of stacked BLSTM layers
From: Multiclass audio segmentation based on recurrent neural networks for broadcast domain data
Layers | SER | Class error (%) | Avg | |||
|---|---|---|---|---|---|---|
mu | sp | sm | sn | |||
1 BLSTM | 16.25 | 16.82 | 30.00 | 26.75 | 32.07 | 26.41 |
2 BLSTM | 15.91 | 16.28 | 28.82 | 26.32 | 31.94 | 25.84 |
3 BLSTM | 16.02 | 15.71 | 27.46 | 30.09 | 30.74 | 26.00 |