Automatic Emotion Recognition Model Using DBN Model

Abstract

Machine learning techniques can now identify human emotions from voice signals, but their poorer detection
accuracies in practical applications stem from their lack of rich representation capability. Speech signals include
numerous levels of representations that can be automatically discovered by deep belief networks (DBN). This research
proposes an ensemble of random deep belief networks (RDBN) technique for speech emotion recognition in order to
fully utilize its benefits. After first extracting the input voice signal's low-level properties, it uses them to create a
large number of random subspaces. After that, each random subspace is given to DBN, which uses the higher-level
features as the classifier's input to produce an emotion label. The final emotion label for the input speech signal is
then determined by fusing all of the outputted emotion labels using majority voting. Experimental findings on
benchmark speech emotion datasets demonstrate that RDBN outperforms the comparison techniques in terms of
speech emotion recognition accuracy.