Examining the Influence of Speech Frame Size and Number of Cepstral Coefficients on the Spe.pdfVIP

Examining the Influence of Speech Frame Size and Number of Cepstral Coefficients on the Speech Recognition Performance Iosif Mporas, Todor Ganchev, Elias Kotinas, and Nikos Fakotakis Department of Electrical and Computer Engineering, University of Patras, 26500 Rion-Patras, Greece {imporas, tganchev, lias, fakotaki}@wcl.ee.upatras.gr Abstract In the present work we explore the influence of front-end setup on the speech recognition performance. Specifically, we study the dependence between specific parameters of the speech parameterization stage, such as speech frame size and number of Mel-frequency cepstral coefficients (MFCC), and the word error rate (WER). Our comparative evaluation is performed by employing the Sphinx-IV speech recognition engine and the TIMIT speech corpus. The experimental results demonstrated that a subset composed of the first 9 cepstral coefficients demonstrate lower WER than the commonly used 13 MFCC. On the other hand, the frame size turned out not to be a crucial parameter. 1. Introduction The contemporary speech recognition technology is based on statistical modelling of speech properties performed through powerful pattern recognition techniques such as the hidden Markov models (HMM) [1] and the Viterbi algorithm [2]. One problem that has not been solved with sufficient elegance, yet, is obtaining an efficient speech parameterization, which is to present the information carried by the speech signal in a compact form, so that it can