Analysis of Acoustic Correlates of Marathi Prosodic Features for Human-Machine Interaction

Abstract

The prosodic feature study gives insight into the suprasegmental properties of speech. Researchers have been studying acoustic features of prosody worldwide to design a model for prosody in various languages to make human-machine interaction natural. This study focuses on the interface between linguistics and technology, particularly for less studied languages like Marathi, to improve human-machine interaction applications by investigating acoustic correlates of various prosodic features of Marathi emotional utterances. The Marathi language, spoken in the Indian states of Maharashtra and Goa, has received relatively little attention for prosody. Due to a lack of appropriate resources, studying the prosodic features of the Marathi language was difficult. With the help of seven male and four female actors, we created a database of 440 Marathi utterances in the four main emotion styles—anger, happiness, fear, and neutral. The acoustic correlates of prosody were calculated from the recorded dataset using PRAAT software. According to this study, the prosodic characteristics of Marathi emotional utterances alter in response to emotions. However, anger and happiness; and fear and neutral emotions share somewhat similar acoustic properties.