Skip to main content

Table 5 Average class error and relative improvement over the baseline system for the RNN classifier with the combination and pooling block before the first BLSTM layer on the test partition

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

Config

Avg class error (%)

Rel. improvement (%)

Conv Pool BLSTM1,2

29.30

āˆ’ā€‰11.80

Conv BLSTM1,2

26.26

āˆ’ā€‰1.60

Pool BLSTM1,2

28.31

āˆ’ā€‰8.72