Skip to main content

Table 4 SER, error per class and average class error for the RNN classifier combined with the resegmentation module over test partition for the best feature configuration and different values of the down-sampling factor, L, and minimum segment length, Tmin

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

L,TminSERClass error (%)Avg
25, 1.25 s12.4914.5521.9919.0824.8820.13
35, 1.4 s12.4814.3122.2618.7025.1020.10
45, 0.9 s12.4614.1922.1418.8225.0420.05
55, 0.55 s12.5716.1222.0018.9424.9520.50