Role of gender influence in vocal Hindi conversations: A study on speech emotion recognition

Recent times have been marked with the increasing demand for more intelligent human computer interfaces. By adding emotion recognition abilities, voice based interfaces can be made more human centric. As natural languages do not share similar acoustic-phonetic features and vary in production of speech sound, the emotion recognition accuracy gets affected with respect to the user's language. This work aims at studying the patterns of stress and intonation for emotional speech in Hindi (Indo-Aryan language) and analyzing the influence of gender on speech emotion recognition accuracy. The paper proposes a combined system for gender distinction and emotion recognition by extracting basic prosodic and spectral speech features and also compares three different classification algorithms. The performed experimentation over a Hindi emotional corpus reveals that 78% correct speech emotion recognition accuracy can obtained by adopting support vector machines for classification.