A Research of Speech Emotion Recognition Based on Deep Belief Network and SVM

Table 1

Hyperparameters and training statistics of the chosen DBN.

Number of hidden layers	5
Units per layer	50
Unsupervised learning rate	0.001
Supervised learning rate	0.01
Number of unsupervised epochs	50
Number of supervised epochs	475
Total training time (hours)	136
Classification accuracy	0.865