MUMBAI, India, Sept. 14 -- Intellectual Property India has published a patent application (202641106626 A) filed by Nitte Meenakshi Institute Of Technology, Nitte Deemed To Be University, Bengaluru; Tejaswini N P; Sharath Nathaniel; Surya S; and Satvik Daruvuru on September 04, 2026, for Real-Time Bidirectional Sign Language Converter System For Mute And Deaf Persons Using A Hybrid Deep Learning Architecture.
Inventors include Nitte Meenakshi Institute Of Technology, Nitte Deemed To Be University; Tejaswini N P; Sharath Nathaniel; Surya S; and Satvik Daruvuru.
The application for the patent was published on September 11, 2026, under issue no. 37/2026.
Abstract: ABSTRACT OF THE INVENTION Title: REAL-TIME BIDIRECTIONAL SIGN LANGUAGE CONVERTER SYSTEM FOR MUTE AND DEAF PERSONS USING A HYBRID DEEP LEARNING ARCHITECTURE The invention relates to a real-time, bidirectional, vision-only sign language converter system (Figure 1) that simultaneously supports American Sign Language (ASL) and Indian Sign Language (ISL) and provides both Sign-to-Text-and-Speech and Text-to-Sign translation through a single integrated interface, thereby bridging the communication gap between deaf and mute persons and hearing persons without requiring any wearable sensors, data gloves or specialised hardware. The system comprises a video input module (101), a mode selector (102) and a language selector (103), a hand-landmark detection module (104), an ASL sub-pipeline having a Region-of-Interest extractor (105), a multi-orientation test-time augmenter (106) and a MobileNetV2-based Convolutional Neural Network classifier (107), and an ISL sub-pipeline having a feature constructor (108) that builds a novel 86-dimensional relational vector from wrist-relative coordinates of both hands together with an inter-hand displacement vector, a z-score normaliser (109) and a Multi-Layer Perceptron classifier (110) of funnel architecture 256-128-64, the outputs being stabilised by a sliding-window temporal smoother (111) and rendered through a unified Streamlit interface (114) by a text-and-speech output module (113) employing a Google Text-to-Speech engine and a sign-display module (115) driven by a hierarchical text-to-sign lookup engine (112) with a fingerspelling fallback. The present invention ensures low-latency, low-cost, robust and stable real-time recognition for both ASL and ISL on commodity hardware, position-invariant and background-invariant classification of complex two-handed ISL gestures through its 86-dimensional relational feature vector, and truly bidirectional communication between the deaf/mute community and the hearing population through a single, accessible graphical user interface as illustrated in
Disclaimer: Curated by HT Syndication.
