Expressive speech synthesis using prosodic modification for Marathi language

Abstract

This paper describes Prosodic modification in Neutral speech for generating expressive synthetic speech in Marathi Text-to-Speech (TTS) system. This paper focuses on voice modification technique maintaining acceptable quality and naturalness with reduced database. In this paper, a method to modify fundamental frequency contour is proposed for Marathi (language spoken in Maharashtra, India) TTS system. The emotion content in the speech is highly correlated to phonetic description and prosodic features such as fundamental frequency and time duration of that phone. Speech modification is done in time domain. Speech modification is done by a process of compressing the speech signal in time domain without changing its spectral features. For prosody generation, we have obtained a primary pitch curve for the word, based on the location followed by punctuation marks. Question mark and exclamation mark in the text are studied to modify prosody. TD-PSOLA (Time Domain Pitch Synchronous Overlap and Add) algorithm can improve the prosody of the synthesized speech. Prosodic changes can be possible by modifying different parameters like pitch level, pitch contour, pitch variation. The experimental results showed that the proposed prosody modification, based on pitch and duration modification can improve speech quality.