LPC-based Neural Network For Automatic Speech Recognition
Zoheir Deiri, N. Bostros · 2005
An algorithm for automatic recognition of spoken words is presented in this study. A small vocabulary of 50 speaker-dependent i sol at- ed - words with short duration is implemented successfully. The algorithm is based on ex- tracting features (formants) f rom the speech signal and presenting them to a three-layer back-propagation artificial neural network for recogn i ti on. To implement and test the a lgorithm a microcomputer-based data acquisition system has been designed and constructed. The speech signal from a microphone is digitized and stored in the microcomputer where the LPC algorithm is applied and the formants are calculated. The formants are a pplied to the artificial neural network which is implemented as pattern classi- fier. The adaptation rule implemented in this network is the generalized least mean square (LMS) rule. The output of the network is the recognized word.