Pengenalan Huruf Vokal Berdasarkan Gerakan Wajah Menggunakan CNN-LSTM

Rahman, Muhammad Daffa Abiyyu (2024) Pengenalan Huruf Vokal Berdasarkan Gerakan Wajah Menggunakan CNN-LSTM. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022211015-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2026. Download (2MB) | Request a copy

Abstract

Deteksi gerakan wajah untuk menentukan huruf vokal yang sedang diucapkan oleh seorang subjek saat ini masih perlu dieksplorasi. Ditemukan adanya korelasi antara gerakan wajah dan akustik suara yang dihasilkan menunjukkan bahwasanya memungkinkan untuk menentukan huruf yang diucapkan oleh seorang individu berdasarkan gerakan wajahnya. Memanfaatkan teknologi Convolutional Neural Network dan Long-Short Term Memory memungkinkan untuk mengenali fitur gambar-gambar dalam sekuential waktu tertentu, huruf vokal yang diucapkan oleh seorang subjek dapat dipelajari oleh komputer untuk mengenali huruf vokal. Karena kompleksitas bahasa, penelitian ini difokuskan untuk huruf vokal Bahasa Indonesia. Penelitian ini mengambil pendekatan pemanfaatan gerakan wajah dengan model CNN-LSTM dengan akurasi 65.31% terhadap data yang tidak diketahui. Saat melakukan perbandingan model CNN-LSTM dengan model alternatif yang terdiri atas model CNN dan model CNN-BiLSTM, ditemukan akurasi terbaik dimiliki model CNN-BiLSTM dengan 72.12% akurasi.
=================================================================================================================================
Detection of facial movement to determine the vowel being spoken by a subject currently still needs to be explored. A correlation was found between facial movements and the acoustic sound produced indicates that it is possible to determine the letters spoken by an individual based on their facial movements. Utilizing Convolutional Neural Network and Long-Short Term Memory technology makes it possible to recognize image features in a certain time sequence, vowels spoken by a subject can be learned by a computer to recognize vowels. Due to the complexity of languages, this research is focused on Indonesian Language vowels. This research takes the approach of utilizing facial movements with CNN-LSTM resulting in 65.31% accuracy on unknown data. When compared with alternative models such as CNN and CNN-BiLSTM, the highest accuracy is owned by the CNN-BiLSTM models with 72.12% accuracy.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Huruf Vokal, Gerakan Wajah, Convolutional Neural Network, Long-Short Term Memory, Vowels, Facial Movement
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7882.P3 Pattern recognition systems
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Muhammad Daffa Abiyyu Rahman
Date Deposited:	24 Jan 2024 06:27
Last Modified:	24 Jan 2024 06:27
URI:	http://repository.its.ac.id/id/eprint/105598

Actions (login required)

View Item