Modeling of vowel duration in Malayalam speech using probability distribution

Abstract

The first part of the work detailed in this paper, analyzes the probability distribution of vowels in Malayalam, an Indian language. Different probability distributions are fitted to the duration values of each vowel and the best fit is determined using quantile-quantile plot. The probability distribution of duration values, in accordance with the different factors affecting durations are also analyzed. In the second part, based on the results of the distribution fitting, a duration model is developed. It predicts duration as that, which maximizes the conditional probability, for the given set of factors. The model has the advantage that it can be expanded easily to include more factors, without affecting the existing structure. The performance of the model is evaluated using root mean square error (RMSE) and correlation between predicted duration and actual duration. The objective evaluation of the model gave an RMSE of 13.2 ms and a correlation of 0.80.