Study of Acoustic Correlates Between Prosodic Features and Emotions in Marathi Language

Abstract

For designing intelligible and natural Text-to-Speech systems (TTS) researchers all over the world are studying relation between prosodic features and emotions in various languages. The objective of this research is to study acoustic correlates between prosodic features including duration, F0 and intensity for vowels and various emotions in Marathi language. Also, these prosodic features are analyzed for vowels in words non-final and final syllables. Pause durations occurring between words for various emotions are also studied. Marathi sentences spoken by a female speaker are recorded in five emotion styles: anger, fear, happy, sadness and neutral. For every sentence vowel mean duration, mean F0 and mean intensity are calculated. Long pauses, mid-pauses and short pauses for all the five emotions are analyzed. The ANOVA analysis for F0 and intensity with p<0.001 indicate that these parameters can be used as cue for various emotion styles for Marathi language. Short vowel duration and large number of short pauses are observed for angry emotion whereas long vowel duration and short, mid and long pauses are observed equally for happy emotion. Above observations related to prosodic features and pauses would be useful to model Marathi TTS system.