PHONEMES CLASSIFICATION USING ARTIFICIAL NEURAL NETWORK BASED ON SPEECH SPECTRUM FEATURES

A METHOD FOR CLASSIFY PHONEMES AND/OR SYLLABLES PRONOUNCED BY HEARING-IMPAIRED PEOPLE UTILIZING ARTIFICIAL NEURAL NETWORK BASED ON SPEECH SPECTRUM FEATURES. USER DETAILS OR INFORMATION ARE STORED PRIOR TO THE SYSTEM (1). PHONEMES AND/OR SYLLABLES ARE PRESENTING TO USER (2) AND THE PRONUNCIATION OR UTTERANCE IS RECORDED VIA AN AUDIO INPUT DEVICE (3). PERCEPTUAL LINEAR PREDICTION (PLP) IS APPLIED FOR ANALYZING THE PRONOUNCED TARGET SYLLABLES AND A SPEECH SPECTRUM IS PRODUCED (4). SPEECH SPECTRUM IS SEGMENTING TO LOCATE THE CONSONANT-TO-VOWEL (CV) TRANSITION BORDER AND TO REMOVE NON-SYLLABLE SIGNAL (5). SEGMENTED SPEECH SPECTRUM IS USING AS INPUT DATA FOR IDENTIFICATION BY THE MULTILAYER PERCEPTRON (MLP) NETWORK (6). ACCOMPANYING DRAWING: