A Review of Front End and Back End Techniques for ASR

Abstract

Speech recognition is an alternative of typing on key-board. It is based on sound-analysis and converts the spelled words into the text. Last few decades have strengthened the foundation of ASR systems. This paper aims to provide an overview of the recognition process. Various feature extraction methods like MFCC, PLPCC etc. are reviewed here. These methods (MFCC and PLPCC) are compared on the basis of their way of processing the speech utterance. Connectionist approach to recognize the speech is explored. Finally experimental results are presented to show that how PLPCC provides more accuracy than MFCC as the number of coefficients increases.

Keywords: ASR, Hidden Markov Models, Mel Frequency Cepstral Coefficients, PLPCC, Time Delay Neural Networks, WER