Hybrid Approach of Feature Extraction and Vector Quantization in Speech Recognition

Abstract

This paper examines the speech recognition process. Speech recognition has two phases: the front end which comprises of preprocessing of the speech waveform and the back end which comprises of feature extraction and feature matching. In this review, we discuss some feature extraction techniques like MFCC, LPC, LPCC, PLP, and RASTA-PLP. As these techniques have some demerits stated in the paper, we discuss a hybrid approach of feature extraction with some combinations of the above techniques. Feature matching helps in the recognition part of speech recognition. It is done by comparing the feature vectors of the current user to the feature vectors stored in the database. It can be optimized by vector quantization (VQ) in order to speed up the recognition process.