Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{207503,
        author = {Yash Savdekar and Gunjan Rane and Ketki Gokakkar and Sayali Navale and Prof. Mrs. Priyanka Deshpande},
        title = {Signverse: A Real-Time Indian Sign Language Recognition System Using Lightweight Spatio-Temporal Deep Learning},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {3},
        pages = {1461-1471},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=207503},
        abstract = {Effective communication between individuals who are hearing or speech impaired and those unfamiliar with sign language continues to pose a meaningful challenge in every-day social and professional contexts. While several gesture-recognition systems have been proposed over the years, their practical deployment is frequently constrained by heavy com-putational requirements, reliance on specialized hardware, or an inability to perform reliably under natural, unconstrained recording conditions. This paper presents Signverse, a real-time Indian Sign Language (ISL) recognition system designed around a lightweight spatio-temporal deep learning pipeline that runs on standard consumer hardware without any GPU acceleration. The system captures video through a webcam, extracts hand landmarks using MediaPipe, normalizes the resulting coordinate sequences, and feeds them into a Gated Recurrent Unit (GRU) network trained to classify eleven dynamic ISL gestures. A custom dataset of approximately 1,650 gesture sequences was assembled across multiple participants and diverse recording environments to encourage generalization. Under five-fold cross-validation, Signverse achieves a mean test accuracy of 96.8%, with a macro-averaged F1-score of 0.966, while maintaining real-time inference at roughly 28–30 frames per second on a standard laptop CPU. The results demonstrate that landmark-based temporal modeling provides a viable and computationally efficient pathway for accessible ISL interpretation, and the modular design of the pipeline makes it straightforward to extend toward larger vocabularies in future work.},
        keywords = {Indian Sign Language, gesture recognition, Medi-aPipe, GRU, LSTM, real-time inference, landmark extraction, as-sistive technology, human-computer interaction, spatio-temporal learning},
        month = {August},
        }