Skip to main content

Table 8 SER, error per class and average class error on the test partition for the BLSTM1PoolBLSTM2 RNN classifier (Avg pooling, N=10) trained using mixup augmentation with hyperparameter Ī±

From: Multiclass audio segmentation based on recurrent neural networks for broadcast domain data

Mixup

SER

Class error (%)

Avg

Ā Ā 

mu

sp

sm

sn

Ā 

Ī±=0.1

14.84

15.21

27.99

23.05

29.34

23.90

Ī±=0.2

14.80

14.64

28.20

22.01

29.01

23.56

Ī±=0.3

14.81

16.03

26.32

23.89

28.22

23.62

No mixup

15.47

15.55

29.16

24.34

30.51

24.89