Acoustic analysis and perception of emotions in hindi speech using words and sentences

Abstract

This paper presents a study related to perceptual evaluation of emotions expressed in Hindi speech and their acoustic prosodic correlates. Six emotions i.e. Neutral, Happiness, Sad, Fear, Anger and Surprise were selected for the present study. For this purpose, a database of fifteen continuous sentences and isolated words i.e. Hindi digits ‘शून्य, एक, दो, तीन चार, पाँच, छः, सात, आठ, नौ’ spoken by five males and five females repeated three times by each speaker, was created. Understanding the difference between acoustic features for different emotions is important for computer recognition and their classification. For the analysis of acoustic features which changes correspondingly with emotions were analyzed using PRATT speech processing software tool. Human perception experiment shows that the overall recognition of the emotions is about 60% for continuous sentences and 53% for isolated words. It is found that anger has highest intensity followed by neutral, happiness, surprise, sad and fear. Some differences were observed in case of continuous sentences. The dynamic changes in the pitch and intensity in these utterances have been analyzed.