Enhancing Information Accessibility for Visually Impaired Users through BART Text Summarization and Text-to-Speech Integration in a Flutter-based Mobile Application

Visually impaired individuals face significant challenges in accessing digital information, especially due to the overwhelming amount of content online. Traditional reading methods such as braille are often insufficient, highlighting the need for more efficient solutions. This research develops a text summarization system for Indonesian news articles to enhance information accessibility for visually impaired users. The methodology involved collecting 1,000 news articles and preparing the data through tokenization and dataset splitting. The BART model, known for its strong performance in abstractive summarization, was used. The model was pretrained using a denoising autoencoder approach and then finetuned on the Indonesian dataset to optimize summarization performance. Post fine-tuning, the BART model showed significant improvements, with ROUGE-1, ROUGE-2, and ROUGE-L scores improving from $\mathrm{0. 4 9 8 6}, \mathrm{0. 4 0 8 5}$, and $\mathrm{0. 4 5 8 0}$ to $0.6436,0.5506$, and 0.6040, respectively. The system was implemented via an API using FastAPI and integrated with a Flutter-based mobile application. Testing showed that scraping times ranged from 17.34 to 54.09 seconds, while text summarization took between 20.20 and 23.65 seconds. The application was tested on platforms including an Android 14, demonstrating its effectiveness in real-world scenarios.