Inhibition/Enhancement of Articulatory features - Which one is Dominant for Speech Recognition

Abstract

This paper presents a speech recognition technique based on inhibition/enhancement (In/En) of articulatory features (AFs) by determining the dominant factor between inhibition and enhancement. The proposed method comprises three stages - a) Multilayer Neural Networks (MLNs), b) In/En Network and c)Gram-Schmidt (GS) Orthogonalization. At first stage, the MLNs detects AFs and then In/En network is used to achieve categorical articulatory movement by enhancing peak patterns and inhibiting dip patterns. Finally GS algorithm decor relates the modified features to obtain orthogonalized feature vector before connecting with a hidden Markov model (HMM) based classifier. From the experiments based on phoneme recognition, it is shown that the enhancement of AFs affects more on Phoneme Recognition than Inhibition.