Formant frequency estimation of phonemes of Assamese speech

Abstract

Phonemes are the smallest distinguishable unit of speech signal. Formant frequency of a phoneme, the most fundamental concept in speech processing, differentiate one phoneme from another. Range of formant frequency of a particular phoneme can be used as a priori knowledge in various speech processing application. This paper describes a work done for estimating the formant frequencies of all consonant phonemes of Assamese language, which is a phonetically distinct language of North East India. The work uses the concept of pole or formant location determination from the linear prediction model of vocal tract. A discrete wavelet filter is used to segment phoneme from some typical Assamese word. The formant range thus estimated was later used in classification of Assamese phoneme family with the help of a Recurrent Neural Network.