Projected Features for Hindi Speech Recognition System

Abstract

Automatic speech recognition (ASR) is a technology that allows a computer to identify the words uttered by a person using a
microphone or telephone. The processing required is divided into two parts: the signal processing front end and statistical framework of hidden
Markov model (HMM) at back-end for pattern classification. This paper presents a comparative study of different types of feature reduction
techniques in the context of Hindi language. Experimental results show a significant improvement in ASR performance by using extended MFPLP
(PLP derived from Mel scale filter bank) feature extraction technique at front-end with the help of Hetroscedastic linear discriminant
analysis (HLDA) projection scheme. All the investigations are based on the experiments conducted in typical field conditions using standard
close talking microphone.

Keywords: ASR; feature reduction; Hindi; PCA; LDA; HLDA