Vowel Sound Recognition Using a Spectrum Envelope Feature Detection Method and Neural Network

摘要

We proposed vowel sound recognition using a spectrum envelope feature detection method. At first, we collected sound with a headset microphone and Windows sound recorder. I used disintegration Fourier transform and cepstrum analysis to create sound and extracted the characteristic point in the frequency axis of each vowel. I took the characteristic point from the first to the second. This is sample data of each vowel sound. With a neural network, I made a program to perform vowel sound recognition. As a result, I was able to finish the basic vowel recognition program. With this method two undulate mountains in the spectrum envelope graph of each voice data are selected. Each vowel sound can be categorized clearly in spite of the environment of the speaker indefiniteness. It is possible to improve the voice recognition in the speaker indefiniteness in the simple algorithm.