RASTA-PLP For Speech Recognition of Articulatory Handicapped People

Abstract

This paper describes an approach of isolated word recognition for normal people and articulatory handicapped people using relative spectral and cepstral perceptual linear predictive (RASTA-PLP) feature extraction techniques. Recognition is carried out using a minimum distance classifier. The results of RASTA-PLP Cepstral coefficients and RASTA-PLP spectral coefficients are compared. The result for normal persons shows that the recognition accuracy is 75.11% from features of RASTA-PLP Cepstral coefficients as compared to 62.11% calculated from RASTA-PLP spectral coefficients. The result for articulatory handicapped persons shows that the recognition accuracy is 45.60% from features of RASTA-PLP Cepstral coefficients as compared to 38.50% calculated from RASTA-PLP spectral coefficients.