Implementasi Speech-To-Text Dan Penerjemahan Bahasa Jawa Ke Bahasa Indonesia Dengan Arsitektur Berbasis Transformer Pada Aplikasi Android

Wiyoga, Athallah Narda (2025) Implementasi Speech-To-Text Dan Penerjemahan Bahasa Jawa Ke Bahasa Indonesia Dengan Arsitektur Berbasis Transformer Pada Aplikasi Android. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5027211041_Athallah Narda Wiyoga_BukuTA.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Bahasa Jawa merupakan bahasa daerah dengan jumlah penutur terbanyak di Indonesia. Namun, mahasiswa dari luar Pulau Jawa sering mengalami kesulitan dalam memahami bahasa ini, sehingga dibutuhkan solusi teknologi yang mendukung proses adaptasi linguistik. Penelitian ini mengembangkan aplikasi Android berbasis speech-to-text dan penerjemahan otomatis dari Bahasa Jawa ke Bahasa Indonesia menggunakan model transformer. Empat model Automatic Speech Recognition (ASR) diuji: Whisper-small, HuBERT-large, Wav2vec2-large, dan XLSR, menggunakan dataset OpenSLR. Hasil evaluasi menunjukkan bahwa HuBERT-large memiliki WER terendah 20,77%, tetapi mengalami masalah pada pengenalan fonem nasal. Whisper-small, meskipun memiliki WER lebih tinggi 25,77%, dipilih karena ukuran model yang kecil (967 MB), hasil transkripsi yang lebih stabil, serta arsitektur encoder-decoder yang mendukung akurasi pada fonem kompleks. Model Whisper-small diintegrasikan ke dalam aplikasi melalui FastAPI, dengan rata-rata waktu transkripsi 5–10 detik dan penerjemahan menggunakan Google Translate API versi tidak resmi dalam 5–7 detik. Pengujian oleh pengguna dari berbagai daerah menunjukkan performa yang akurat, stabil, dan mudah digunakan. Aplikasi ini berpotensi menjadi solusi praktis bagi mahasiswa non-Jawa serta berkontribusi pada pelestarian bahasa daerah melalui teknologi.
==============================================================================================================================
Javanese is one of the most widely spoken regional languages in Indonesia. However, students from outside Java often face difficulties in understanding the language, creating a need for technological solutions to support linguistic adaptation. This study developed an Android application integrating speech-to-text and automatic translation from Javanese to Indonesian using a transformer-based architecture. Four Automatic Speech Recognition (ASR) models were evaluated Whisper-small, HuBERT-large, Wav2vec2-large, and XLS-R using the OpenSLR dataset. Evaluation results showed that HuBERT-large achieved the lowest Word Error Rate (WER) at 20.77%, but encountered issues in recognizing nasal phonemes. Despite having a higher WER (25.77%), Whisper-small was selected due to its smaller model size (967 MB), more stable transcription output, and its encoder-decoder architecture that effectively handles complex phonemes. Whisper-small was integrated into the application via FastAPI, enabling real-time transcription within 5–10 seconds. Translation is handled using an unofficial version of the Google Translate API, with an average response time of 5–7 seconds. User testing across various regions confirmed the application’s accuracy, responsiveness, and ease of use. This application offers a practical solution for non-Javanese speakers and contributes to the preservation of regional languages through AI-based technology.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Android, Bahasa Jawa, Deep Learning, Speech Recognition, Android, Deep Learning, Javanese, Speech Recognition
Subjects:	P Language and Literature > PA Classical philology T Technology > T Technology (General) T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.83 Dynamic programming
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information Technology > 59201-(S1) Undergraduate Thesis
Depositing User:	Athallah Narda Wiyoga
Date Deposited:	29 Jul 2025 06:37
Last Modified:	29 Jul 2025 06:37
URI:	http://repository.its.ac.id/id/eprint/122527

Actions (login required)

View Item