MUMBAI, India, Oct. 26 -- Intellectual Property India has published a patent application (202641123831 A) filed by Dr. Tamilarasan. S; Dr. Shrinivas Sirdeshpande; Dr Ramani G; K Anil Kumar; K Akhilesh Achari; Jyotshna Bharti; and Gowthami M on October 15, 2026, for Sign2voice: A Deep Learning-Based Real-Time Indian Sign Language Recognition And Multilingual Speech Translation System.
Inventors include Dr. Tamilarasan. S; Dr. Shrinivas Sirdeshpande; Dr Ramani G; K Anil Kumar; K Akhilesh Achari; Jyotshna Bharti; and Gowthami M.
The application for the patent was published on October 23, 2026, under issue no. 43/2026.
Abstract: ABSTRACT OF THE INVENTION The present invention provides a real-time Indian Sign Language recognition and multilingual speech translation system. Live RGB video is acquired and preprocessed, hand and upper-limb landmarks are extracted, and a 348-dimensional frame representation comprising a 294-dimensional hand block and a 54-dimensional upper-limb block is generated. Temporal samples are formed using a 30-frame sequence. A two-layer Bi-LSTM recognizes isolated ISL gesture classes, with an optional temporal-attention mechanism for weighting information across time steps.An optional transfer-learning configuration adapts an existing 352-dimensional IPN Hand source representation to a target-compatible 348-dimensional representation, trains a source temporal model, transfers selected learned parameters, and fine-tunes the model on ISL target data. In a particular experimental embodiment, INCLUDE-96 provides 96 target gesture classes.Confidence filtering and temporal processing stabilize predictions. Recognized ISL text may be translated into a selected Indian language and converted into speech through downstream translation and text-to-speech components. The system provides an integrated pathway from camera-based signing to stabilized recognition and multilingual spoken output.
Disclaimer: Curated by HT Syndication.
