EURASIP Journal on Audio, Speech, and Music Processing

Table 7 SER, error per class and average class error for the BLSTM₁PoolBLSTM₂ RNN classifier combined with the HHM resegmentation on the test partition for different pooling factors (N) and average pooling

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

N	SER	Class error (%)				Avg
		mu	sp	sm	sn
5	12.48	14.31	22.16	18.69	25.12	20.07
10	12.41	12.87	22.58	19.16	24.99	19.90
100	13.47	22.85	24.32	19.31	26.20	23.17
No pool	12.46	14.19	22.14	18.52	25.04	20.05

Back to article page