BdSL-SPOTER: A Transformer-Based Framework for Bengali Sign Language Recognition with Cultural Adaptation

2025-11-15

AI Summary

To address the challenges of strong cultural specificity and scarce annotated data in Bangla Sign Language (BdSL) recognition, this paper proposes a lightweight pose-driven Transformer framework. Methodologically: (i) a culturally adaptive gesture preprocessing module is introduced to capture region-specific BdSL characteristics; (ii) an optimized learnable positional encoding scheme is designed to enhance spatiotemporal modeling of skeletal keypoints; and (iii) a curriculum learning strategy is integrated to improve generalization and convergence speed under low-data regimes. The architecture employs only four compact Transformer encoder layers, significantly reducing parameter count and FLOPs. Evaluated on the BdSLW60 benchmark, the model achieves 97.92% Top-1 accuracy while accelerating inference by 32% over prior approaches. This work delivers an efficient, deployable solution for low-resource sign language recognition, balancing accuracy, computational efficiency, and cultural adaptability.

Technology Category

• Computer Vision: Biometrics, Face, Gesture & Pose
• Machine Learning: Multimodal Learning
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

We introduce BdSL-SPOTER, a pose-based transformer framework for accurate and efficient recognition of Bengali Sign Language (BdSL). BdSL-SPOTER extends the SPOTER paradigm with cultural specific preprocessing and a compact four-layer transformer encoder featuring optimized learnable positional encodings, while employing curriculum learning to enhance generalization on limited data and accelerate convergence. On the BdSLW60 benchmark, it achieves 97.92% Top-1 validation accuracy, representing a 22.82% improvement over the Bi-LSTM baseline, all while keeping computational costs low. With its reduced number of parameters, lower FLOPs, and higher FPS, BdSL-SPOTER provides a practical framework for real-world accessibility applications and serves as a scalable model for other low-resource regional sign languages.

Problem

• Recognizing Bengali Sign Language with cultural adaptation
• Improving accuracy on limited data using curriculum learning
• Providing efficient framework for low-resource sign languages

Innovation

• Transformer-based pose framework for sign language recognition
• Cultural preprocessing and compact four-layer encoder design
• Curriculum learning enhances generalization on limited data