Optimization of some parameters in the speech-processing module developed for the speaker independent ASR system

Abstract

This paper deals with looking for an optimum parameterization in automatic speech recognition systems working with the speech transferred over a telephone channel. The performed experiments were supported by a large collection of training data provided from telephone calls of at least one thousand speakers. MFCC and PLP cepstral parameterizations were tested with the aim to find the optimal number of filters and coefficients. Temporal patterns describing several adjacent frames of a given frame were verified in connection with techniques ensuring feature extraction and decorelation of pattern space.

Keywords

• linear discriminate analysis
• optimization of front-end
• pattern space normalization
• speech parameterization