An accessible web-based system for audio transcription and braille conversion

Abstract

This paper presents the development and evaluation of Transkiptor Web, a web application designed to automatically transcribe audio to text using artificial intelligence and subsequently convert the text to Braille. The system aims to promote the inclusion of people with visual impairments and deafblindness by providing access to printed or digital Braille content derived from audio sources. The system integrates deep learning-based speech recognition models (Whisper) and an automatic Braille conversion module (Gemini). Tests performed with Brazilian Portuguese audio samples demonstrated a word error rate (WER) below 3% under ideal conditions and an average Braille conversion accuracy of 98.7%. The platform is structured in modular components, including audio upload, transcription, Braille conversion, and export in accessible formats. The results indicate that the proposed solution can effectively contribute to digital inclusion and educational accessibility by allowing people with visual and hearing impairments to access recorded content. Future developments will focus on extending language support and integrating with assistive devices, allowing other groups to benefit from the transcriber.