Comparison of vowel feature extraction on time and frequency domain

Abstract

Vowel recognition is the most important part of the speech recognition process. Most spoken speeches must contain vowels to be sounded. It needs a method that can separate a vowel with another. The methods of the feature extraction on time domain, frequency, cepstrum, and fourier are several basic methods that can be used. This paper compares features of the strengths of the feature of zero crossing rate, energy, spectral centroid, spectral spread, spectral entropy, harmonic ratio, fundamental frequency, cepstrum, and fourier to separate and recognize vowels of a, i, u, e, and o in Indonesian dialect/language. The results show that the spectral spread feature that is one of the features in the frequency domain has the most accurate ability to recognize vowels tested compared to other features.