Gender Classification using MLP-based Speech Emotion Detection

Speech Emotion Detection (SED) is the extraction of the speaker's emotional state from his or her speech signal. There are a few universal emotions, such as Neutral, Anger, Happiness, and Sadness, that can be trained by any intelligent system with limited computational resources to identify as needed. In this work a gender classification model is proposed using an Multi-Layer Perceptron (MLP) based SED. Because spectral features contain emotional information, they are used in this work. The prosodic features that are used to model different emotions are fundamental frequency, loudness, pitch, speech intensity, and glottal parameters. Each utterance is analysed to extract potential features for the computational mapping of emotions and speech patterns. Classification of Gender is done by detecting pitch using the selected features. Here we used RAVDESS dataset which is trained and tested using a Convolution Neural Network. The most notable features like weight connectivity, local connectivity, and polling results are thoroughly trained and have an accuracy of 52.43 percent. To improve the model accuracy, we used the MLP classifier and feature extraction techniques namely, Mel-Frequency Cepstral Coefficients (MFCC) and Mel spectrogram, which resulted in an accuracy of 68.43 percent.