Skip to main content

Table 10 Average class error and error per class obtained on the AlbayzĆ­n 2010 test partition for different systems proposed in the literature compared to our proposed RNN approaches

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

System

Class error (%)

Avg

Ā 

mu

sp

sm

sn

Ā 

Eval winner [65]

19.20

39.50

25.00

37.20

30.30

FA HMM [17]

18.80

23.70

*23.60

29.10

23.80

DCASE Baseline

19.03

25.58

23.59

29.52

25.18

RNN baseline [52]

14.19

22.14

18.82

25.04

20.05

RNN + Pool

12.87

22.58

19.16

24.99

19.90

RNN + Pool + mixup

12.46

22.86

17.34

24.35

19.25