Pengenalan Ejaan Alfabet Fonik Menggunakan Pembelajaran Mesin dengan Ekstraksi MFCC

Razak, Farzana Afifah (2024) Pengenalan Ejaan Alfabet Fonik Menggunakan Pembelajaran Mesin dengan Ekstraksi MFCC. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025201130_Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (3MB) | Request a copy

Abstract

Fonik adalah metode pendidikan yang membantu anak-anak memahami hubungan antara ejaan dan suara. Fonik melibatkan dua pendekatan utama, yaitu synthetic Phonics dan analytic Phonics, yang masing-masing memiliki cara berbeda dalam mengajarkan anak membaca dan mengeja kata. Penelitian ini mengusulkan pemanfaatan teknologi pengenalan suara untuk mengidentifikasi pengucapan Phonics. Dalam tugas akhir ini, dikembangkan model untuk memprediksi ejaan huruf fonik menggunakan pembelajaran mesin seperti Support Vector Machine (SVM), K-Nearest Neighbor (KNN), dan Random Forest dengan ekstraksi MelFrequency Cepstral Coefficients (MFCC). Proses penelitian mencakup pengumpulan data, augmentasi data melalui penambahan noise dan time shifting secara acak dan konstan. Serta ekstraksi MFCC yang dilakukan baik secara manual maupun menggunakan library Librosa. Langkah-langkah lainnya meliputi penyimpanan label dan fitur, standarisasi data, dan encoding untuk mengoptimalkan performa model. Hasil eksperimen yang diperoleh menunjukkan bahwa model SVM dengan augmentasi data konstan dan ekstraksi MFCC menggunakan library Librosa memberikan performa terbaik dengan akurasi mencapai 91,95%. Hasil ini menunjukkan keefektifan dan kehandalan metode yang digunakan dalam pengenalan ejaan alfabet Phonics, memberikan kontribusi yang signifikan. Selain itu, data non-native dengan model SVM mencapai akurasi 90,53% yang memberikan performa lebih baik dibanding dengan data native. Hal ini dipengaruhi oleh banyaknya jumlah data, cara mengumpulkan data, serta cara mengucapkan ejaan
========================================================================================================================
Phonics is an educational method that helps children understand the relationship between spelling and sound. Phonics involves two main approaches, namely synthetic Phonics and analytic Phonics, each of which has a different way of teaching children to read and spell words. This research proposes the utilization of speech recognition technology to identify Phonics pronunciation. In this final project, a model is developed to predict the spelling of Phonics letters using machine learning such as Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Random Forest with Mel-Frequency Cepstral Coefficients (MFCC) extraction. The research process includes data collection, Data Augmentation through the addition of noise and time shifting randomly and constantly, and MFCC extraction which is done both manually and using Librosa library. Other steps include label and feature storage, data normalization, and encoding to optimize model performance. The experimental results obtained show that the SVM model with constant data augmentation and MFCC extraction using the Librosa library provides the best performance with an accuracy of 91.95%. This result shows the effectiveness and reliability of the method used in Phonics alphabet spelling recognition, making a significant contribution. In addition, non-native data with the SVM model achieved an accuracy of 90.53% which provides better performance compared to native data. This is influenced by the amount of data, how the data is collected, and how the spellings are pronounced.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Augmentasi Data, Fonik, Mel-Frequency Cepstral Coefficients (MFCC), Data Augmentation, Mel-Frequency Cepstral Coefficients (MFCC), Phonics
Subjects:	T Technology > T Technology (General)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Farzana Afifah Razak
Date Deposited:	01 Aug 2024 07:09
Last Modified:	17 Sep 2024 06:16
URI:	http://repository.its.ac.id/id/eprint/110166

Actions (login required)

View Item