Ankita N Chadha, Mukesh A Zaveri, and Jignesh N. Sarvaiya
Complex Cepstrum, Classification, Disordered Speech Recognition, Regression
The society is fully aware of the needs of impaired people. One such impairment resulting due to articulatory or congenital conditions in human speech production mechanism produces a disordered speech. This disordered speech is difficult to interpret and leads to miscommunication. Thus, the speech enabled helping aid is a good alternative to existing devices designed to assist the impaired speakers. Taking this into consideration, this work proposes a spoken word recognition system using a General Regression Neural Network (GRNN) for disordered speech. The complex cepstrum based feature extraction is employed due to their ability to represent the speech production model with mixed phase response. The proposed system is compared with the state of the art techniques. The results are veri- fied using objective and subjective measures. The results suggest that the GRNN based parallel training recognition system performs better in terms of accuracy in contrast to baseline recognition algorithms.