Deep learning, web app · Grades 10 to 11
Sign to Bengali
Real-time sign-language recognition from a webcam, translated into Bengali and English text.
- output languages, Bengali and English
- 2
- output languages, Bengali and English
- webcam recognition in the browser
- Live
- webcam recognition in the browser
- across lighting, hand orientation and background
- Robust
- across lighting, hand orientation and background
01 The person
Why it exists.
Most sign-language tools stop at English. In Kolkata the language at home is Bengali. My app recognises gesture sequences from video and outputs Bengali text with transliteration, so a signer and a non-signer can talk without an interpreter. In August 2026 I donated it, with my employment-aggregator platform, to the Academy of Skill Development, which acknowledged both in writing.
02 How it works
The mechanism.
- 01MobileNetV2 CNN for spatial features per frame
- 02Bidirectional LSTM over fixed-length frame sequences for temporal context
- 03Pre-processing, augmentation and segmentation of video
- 04Flask web app with live webcam recognition, translation and transliteration
Stack
- MobileNetV2 + BiLSTM
- TensorFlow, OpenCV
- Flask