Gender Classification for Emotional Speech using GMFCC and Deep LSTM

Abstract

We have come to the point that one of the important aspects of the process speech emotion recognition is the gender classification. The correct classification of gender will improve the performance of Speech Emotion Recognition (SER) system towards its robustness. Here, we are specifically referring to Gammatone Mel Frequency Cepstral Coefficient (GMFCC) as a feature extraction method that extracts features from IITKGP-SESHC dataset, which is very crucial in deciding either male or female in gender classification. The well known classifier “Deep Long Short Term Memory (Deep LSTM)” is itself an important kind of Recurrent Neural Network (RNN) that handles the long-range dependencies more efficiently than the RNNs. The GMFCC feature has to pass through the Deep LSTM and get average percent gender identification accuracy of 98.3%.