Duration Analysis for Malayalam Text-To-Speech Systems

Naturalness can be achieved in a text-to-speech (TTMP) by incorporating prosodic features which include duration of basic units, intonation patterns and stress. This paper presents the preliminary duration analysis required for a speech synthesis system for Malayalam, one among the 17 languages spoken in India. The statistical analysis of the duration of phonemes has yielded results, which can effectively be used for developing a duration model for the language. The durational patterns of vowels and consonants are compared. The study shows that, it is the duration of vowels that get significantly modified and hence duration modelling has to focus primarily on vowels. For vowels, the positional factors have more effect on duration than the contextual factors. The results obtained in this study can simplify a rather complex task of framing rules for duration variation.