Vowel recognition from telephonic speech using MFCCs and Gaussian mixture models

Abstract

This paper presents vowel recognition from speech using mel frequency cepstral coefficients (MFCCs). In this work, microphone recorded speech and telephonic speech are used for conducting vowel recognition studies. The vowels considered for recognition are from Hindi alphabet namely (a), (i), (u), (e), (ai), (o) and (au). Gaussian mixture models are used for developing vowel recognition models. Vowel recognition performance for microphone recorded speech and telephonic speech are 91.4% and 84.2% respectively. © 2012 Springer-Verlag.