Perbandingan metode Wavelet Daubechies dan MFCC sebagai ekstraksi ciri pada pengenalan fonem dengan Probabilistic Neural Network (PNN) sebagai Classifier

Abstract

Nowadays, the development of telecommunication research is rapidly increasing. One of the research is in sound area. Sound is a way for human to interact with computers, known as word recogniser. Word recogniser is a part of voice recogniser that make the computers possible to receive input from word that pronounced. Word that pronounced contain phonemes that arranged into sentence. Voice recognition technology can recognise and understand words that pronounce by digitalising it, and tuning the digital signal with certain pattern that has been saved in a hardware. The result from this word identification will be displayed into printed word. This research will compare between Wavelet Daubechies and MFCCas identity extraction on word recognition with (PNN) as pattern identifier. PNN is a pattern identifier that has high accuracy. The comparison between trained data and tested data in research is 75% : 25%. Tested data that has been used was vary, such as: testing with increase the noise (pure noise) and data with noise increasing from 30dB, 20dB and 10dB. The result from this research is that the identity extraction by using MFCC is much better than with Wavelet Daubechies. From the pure original data (without noise increasing) the accuracy is 92.3% and in data with noise increasing 30dB, noise 20dB and noise 10dB the accuracy is 50.96%, 26.92% and 19.23%.