Effect of normalization method on classification of speech dysfluencies using LPC, LPCC and WLPCC

Abstract

The main aim of this paper is to discuss the enhancement of the classification performance on the speech dysfluencies, namely, prolongations and repetitions after employ statistical normalization (SN) on signal and cepstral mean variance normalization (CMVN) on the Linear Pridictive Coding (LPC), Linear Prediction Cepstral Coefficient (LPCC) and Weighted Linear Prediction Cepstral Coefficient (WLPCC) features. In addition, the ability of LPC, LPCC and WLPCC on the types of speech dysfluencies are compared. Linear Discriminant Analysis (LDA) is applied to classify the types of dysfluencies. Conventional validation method is use for evaluating the reliability of the classifier. The experimental investigation clarifies that the WLPCC features gives a superior result of 94.90% and outperforms the LPC and LPCC features.