Transformer-Based Deep Learning Approach for ...

Transformer-Based Deep Learning Approach for Indian Sign Language Recognition

Author Name : Jay Joshi, Dhaval Patel

DOI: https://doi.org/10.56025/IJARESM.2023.1112232310

ABSTRACT This study introduces a novel approach to Indian sign language recognition using a transformer-based model, distinguishing itself in the field of machine learning applications. Here, proposed model is trained on custom dataset contains 30 distinct signs. for feature extraction advanced techniques such as MediaPipe employed, For classification transformer-based model architecture used. This model architecture includes a multi-head attention mechanism and layer normalization. Remarkably, the model achieves an accuracy of 94%, surpassing conventional methods like RNN (71%), a hybrid CNN-RNN approach (73%). The significance of these findings emerges not only from the transformer-based model's superior efficacy, but also from its potential to change communication aids for the deaf and hard-of-hearing. By efficiently interpreting sign language with greater accuracy, this model paves the way for more effective and inclusive communication tools, promising a significant impact in both technological and social contexts.