Feature Extraction for Speech Classification

Abstract

Study of phonological Processes, Speech recognition, Speech Synthesis and language learning requires Automatic classification of class of sounds and automatic identification of sound classes. This paper focuses on identifying features efficient in discriminating different classes of sound such as analyzing spectral features such as distinctive frequency components by Linear Productive Coding technique and vocal tract length. Artificial Neural network and Random Forest Classification Technique is used to check effectiveness of identified feature with 10-fold cross validation. The proposed system is also aimed at improving performance of phoneme recognition system.