Skip to main content

Table 9 SER, error per class and average class error on the test partition for the BLSTM1PoolBLSTM2 RNN classifier (Avg pooling, N=10) trained with mixup augmentation with hyperparameter α combined with the HHM resegmentation

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

MixupSERClass error (%)Avg
  muspsmsn 
α=0.112.8814.3923.8718.6825.7020.66
α=0.211.8012.4622.8617.3424.3519.25
α=0.312.1414.8021.7117.3723.5419.36
No mixup12.4112.8722.5819.1624.9919.90