Punjabi Speech Emotion Recognition using Prosodic, Spectral and Wavelet Features

Abstract

A rapid advancement can be seen in affective computing, speech emotion recognition (SER) and human computer interaction, SER has been getting more consideration. English, German and Indian are all popular languages that can be used for SER. Due to the lack of Punjabi speech emotion data, very few researchers have tried to implement Punjabi SER systems. Despite the fact that Punjabi is one of the most commonly spoken and understood languages over the web, there has not yet been a Punjabi speech emotion database. This work introduces a semi-natural Punjabi speech emotions database that was compiled from several TV series and Punjabi movies. This database contains utterances of male and female actors, and covers four emotions: happy, sad, neutral, and happy. The database is used to compute prosodic, wavelet and spectral features for emotion recognition. Wavelet parameters, in addition to the Mel frequency Cepstral (MFCC), which are widely used for SER, are also taken into account. The extracted features can also be used to classify emotions.