Isolated Word Recognition New Feature Extraction Techniques

This book explores the various feature extraction methods for isolated word recognition. Brief history of speech recognition and the classification techniques have been described in the book. We discussed the acoustic sound production and fundamental steps in feature extraction in details. In this book some of the challenges of speech recognition like high recognition rate, low feature dimension, computational efficiency and robustness to noise have been addressed. Apart from the comparison of existing feature extraction and classification methods, some of the new feature extraction techniques for Isolated Word Recognition (IWR) based on wavelet transform, power spectrum and bispectrum are proposed in the book. The detail experimentations were carried out using TI-46 database and Marathi digit database in clean and noisy conditions. The techniques developed provide high recognition rate, low feature dimensions and better performance in noisy conditions as compared to other existing algorithms.