APLIKASI PENCIRIAN DENGAN LINEAR PREDICTIVE CODING UNTUK PEMBELAJARAN PENGUCAPAN NAMA HEWAN DALAM BAHASA INGGRIS MENGGUNAKAN JARINGAN SARAF TIRUAN PROPAGASI BALIK

Sari

Abstract

In this research designed a recognition system for learning the pronunciation of the word animal names in English. Original speech signal sample at 8000 Hz pick out a small portion For voice parameter extraction process used method Linear Predictive Coding (LPC​​) to obtain cepstral coefficients. LPC cepstral coefficients are transformed into the frequency domain with Fast Fourier Transform (FFT). For decision making process of the introduction and use Neural Networks (NN) back propagation. Testing is done using the data train, according to a database of test data and test data do not fit database. While the networks do a variation of 3, 4 and 5 hidden layers respectively for 1, 2 and 3 the number of syllables said. Based on the results of testing training data, the recognition rate for each variation of each network the number of syllables showed no difference in test results, the percentage was 99% for the 1 syllable, 98.5% for the 2 syllables and 100% for 3 syllables. Test data suitable for testing the database, the highest recognition rate for type 1 syllable is a network with 4 hidden layers using a variation of the percentage is 85%, whereas type 2 syllables highest recognition rate using a variation of 5 hidden layers with the correct percentage of 75% and 81.67 % for type 3 syllables using 5 hidden layers. While the test results do not fit the test database, the highest recognition rate for type 1 syllable is a network with 4 hidden layers using a variation of the percentage is 15.83% while the type 2 syllables highest recognition rate using a variation of 3 hidden layers with percentage correct, 20.83% and 33.33% for type 3 syllables using 3 and 4 hidden layers.

Keywords : Linear Predictive Coding, Fast Fourier Transform, Neural Network, Backpropagation.