SISTEM PENGENALAN DIALEK MELALUI IDENTIFIKASI BAHASA DAERAH DI PULAU JAWA MENGGUNAKAN METODE MFCC (MEL-FREQUENCY CEPSTRUM COEFFICIENT) DAN ANFIS (ADAPTIVE NEURO FUZZY INFERENCE SYSTEM)

FAUZI, FAJAR MUHAMMAD (2021) SISTEM PENGENALAN DIALEK MELALUI IDENTIFIKASI BAHASA DAERAH DI PULAU JAWA MENGGUNAKAN METODE MFCC (MEL-FREQUENCY CEPSTRUM COEFFICIENT) DAN ANFIS (ADAPTIVE NEURO FUZZY INFERENCE SYSTEM). S1 thesis, Universitas Muhammadiyah Purwokerto.

Preview	Text Cover Fajar Muhammad Fauzi.pdf Download (3MB) | Preview
Preview	Text Bab I Fajar Muhammad Fauzi.pdf Download (591kB) | Preview
Preview	Text Bab II Fajar Muhammad Fauzi.pdf Download (1MB) | Preview
	Text Bab III Fajar Muhammad Fauzi.pdf Restricted to Repository staff only Download (987kB)
	Text Bab IV Fajar Muhammad Fauzi.pdf Restricted to Repository staff only Download (2MB)
	Text Bab V Fajar Muhammad Fauzi.pdf Restricted to Repository staff only Download (582kB)
Preview	Text Dapus Fajar Muhammad Fauzi.pdf Download (582kB) | Preview
	Text Lampiran Fajar Muhammad Fauzi.pdf Restricted to Repository staff only Download (884kB)

Abstract

Indonesia is a big country with a diversity or cultures and tribes, so Indonesia has many languages or different dialects in each region. Many voice signal processing research has been developed. Dialect Identification is an interesting topic has been developed. Speech Recognition with Mel-Frequency Cepstrum Coefficients (MFCC) and Adaptive Neuro-Fuzzy Inference System (ANFIS) methods for Batavian, Sundanese, Banyumasan, and Suroboyoan dialects of Javanese has been developed in this study. MFCC denotes a feature extraction procedure from a speaker's speech signal, in which the voice signal is converted into numerous feature vectors and then shown graphically. Software Python has been using to analyze and design sound pattern forms. The tests were carried out by designing software, collecting data, training the ANFIS system, creating a GUI, and testing data processing. The results of this study showed that MFCC can provide different feature values for each voice entered into the system, with Preemph = 0.99, Nfilt = 30, Nfft = 512, Winlen = 20ms, Winstep = 10ms, Numcep = 5, and Lowfreq = 100 as the MFCC parameters used in this study. The training model used with 72 data training and two membership functions of the Gaussian type resulted in an overall accuracy value of 46.4% with an average accuracy value of Betawi, 42.8%, Sundanese. 71.4%, Banyumas 42.8%, and Suroboyoan 28.5%.

Actions (login required)

View Item