Deep Learning-based Analysis of Affective Computing for Marathi Corpus

Abstract

Speech Emotion Recognition (SER) has a broad variety of potential applications, including improving human-computer interaction in virtual reality and gaming environments, improving the identification and monitoring of mental health conditions, and improving the accuracy of chatbots and speech-based assistants. It must contend with cross-corpus SER, intonation and dialectal differences, as well as prosodic shifts brought on by factors like age, gender, locality, and religion. Deep Convolution Neural Network-based SER for Marathi is presented in this study. For the emotions of anger, happiness, sadness, and neutrality, our fresh Marathi data set includes 300 recordings of 15 speakers. On the basis of accuracy, precision, recall, and F1-score, the performance of the proposed DCNN is assessed using the new data set. The proposed scheme offers overall data accuracy of 0.4750, 0.4076, and 0.3927 for 5, 10, and 15 speakers, respectively, and overall accuracy of 0.6652, 0.6361, and 0.5800 for 5, 10, and 15 speakers, respectively, after feature extraction, which represents an improvement over the current state of the art used for SER for Marathi Corpus.