ShobdoOCR — Bangla-English OCR for Bangladeshi Documents

ShobdoOCR is a word-level OCR system designed for Bangladeshi government documents including NID cards, birth certificates, land deeds, and invoices. It handles mixed Bengali and English text using a classifier-first dual-recognizer architecture — a lightweight 23K-parameter script classifier (99.82% accuracy) routes each detected word to either a Bengali CRNN or English CRNN recognizer, returning per-word bounding boxes, recognized text, and script labels.

Part of the DocReader BD intelligent document understanding system.

Install

pip install --index-url https://test.pypi.org/simple/ \
            --extra-index-url https://pypi.org/simple/ \
            shobdoocr==0.1.1

Note: shobdoocr is currently hosted on TestPyPI (test registry). Dependencies are fetched from the official PyPI automatically.

Usage

from shobdoocr import OCR

ocr = OCR()  # models download automatically (~80MB)

# Plain text
text = ocr.read_text("nid_card.jpg")
print(text)

# Word-level output with bounding boxes and script labels
results = ocr.read("nid_card.jpg")
for word in results:
    print(word['text'], word['script'], word['box'])

Downloads last month: -; Downloads are not tracked for this model. How to track