Text-Dependent versus Text-Independent speech emotion recognition

Abstract

The communication between individual and equipment is through speech emotion recognition which plays a vital role and is very exigent to handle. Today, this filed has become an important area of research. It has wide range of applications. This paper analyzes the performance of emotion recognition for eight speakers. Indian Institute of Technology Kharagpur Simulated Hindi Emotional Speech Corpus (IITKGP-SEHSC) emotional speech corpora used for emotions recognition. The sentiments under surveillance for this study are anger, fear, happy, neutral, sarcastic, and surprise. The categorization is prepared using Gaussian mixture model (GMM). Mel-frequency cepstral coefficients (MFCCs) attributes have been used for defining the emotions. We have extracted the percentage of accuracy of emotion for both text-dependent data and text-independent data. We also observed that emotion recognition performance depends on text and speaker. We found that the percentage of accuracy of text-dependent data is more than the text-independent data.