Modeling pause duration for Malayalam language TTS

Synthesized speech output from a Text To Speech (TTS) synthesis system requires the inclusion of prosodic features like pause durations, syllable prolongations etc. to improve intelligibility. These factors also make the synthesized speech more understandable to the physically disabled. This paper focuses on analyzing the factors that affect pause duration in Malayalam language. The factors that are analyzed in this paper are the position of the pause, the length of preceding phrase before and the length of succeeding phrase. Descriptive and inferential statistics tools were used to analyze the effect of these factors on pause duration. Using these factors, a pause duration prediction model is proposed in this paper. Inclusion of such prosodic features will make the output of TTS systems more intelligible.