Perbandingan pemodelan Wavelet dan MFCC sebagai ekstraksi ciri pada pengenalan fonem dengan teknik jaringan syaraf tiruan sebagai classifier

Abstract

Researches on voice signals have been carried out using various signal processing methods, such as Linear Prediction Coding (LPC), Mel Frequency cepstrum coefficients (MFCC), and Neural Predictive Coding (NPC) which the whole method is based on Fourier transformation. Therefore, comparisons will be made with other approaches based on Wavelet transformation. This research would be the comparison of two feature extraction types, that are Wavelet daubechies and MFCC. Wavelet Transformation has become increasingly popular in signal processing as image and speech. Wavelet transformation has demonstrated good time-frequency localization properties and are appropiate tools for the analysis of non-stationary signals like speech. MFCC feature extraction that computes the cepstral coefficients by considering the human hearing. ANN multilayer perceptron, known as backpropagation used as classifier. From the research that has been done, it can be concluded that the use of Daubechies Wavelet methods as feature extraction on phoneme recognition is not better than MFCC. The accuracy of Daubechies Wavelet method with a 220 hidden neurons ANN learning has achived 36% rate as the best test results. While on phoneme recognition with MFCC method achieved 100% accuracy rate with a 240 hidden neuron learning ANN.