Publication

Abstract : The work presents a low-cost, end-to-end system for digitizing printed books and enabling intelligent voice-based interaction using OCR, semantic retrieval, and Text-to-Speech (TTS). A Raspberry Pi with a camera module captures a page of a book placed on the system, which are processed using three OCR tools: Tessract, EasyOCR, and DocTR. The extracted text is semantically indexed using the all-MiniLM-L6-v2 Sentence Transformer model. The system has a set of servo motors and moving links which enable turning of pages in either direction for the capture of the next set of images. After capturing all the pages, the users can query the system, and the most relevant response is retrieved and read aloud using Google TTS. Performance is evaluated on 10 textbook chapters using Word Error Rate (WER) and Character Error Rate (CER). Tesseract achieved the best average accuracy with WER of 8.66% and CER of 7.39%.