Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Transformer-based word level Bangla sign language recognition using relative quantization encoding.

Husne Ara Rubaiyeat1, Njayou Youssouf2, Md Kamrul Hasan2

  • 1Department of Computer Science, National University Bangladesh, Gazipur, 1704, Bangladesh.

Scientific Reports
|June 3, 2026
PubMed
Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Pilot Room-Level Acoustic and Physiological Monitoring of Respiratory Disturbance in Pigs Following Experimental <i>Klebsiella pneumoniae</i> Challenge.

Veterinary sciences·2026
Same author

4D Reconstruction of Fetal Left Ventricle from Echocardiography via 2.5D Radial Segmentation and Graph-Fourier Reconstruction.

IEEE transactions on medical imaging·2026
Same author

Towards trustworthy brain stroke diagnosis using a lightweight explainable deep learning framework for CT imaging.

Scientific reports·2026
Same author

Naturally Occurring CodY Variants Alter Ligand Binding, DNA Target Affinity, and Virulence in Clostridioides difficile.

Molecular microbiology·2026
Same author

Hydrophobic modification of a polyurethane sponge <i>via</i> emulsion polymerization for selective oil-water separation.

RSC advances·2026
Same author

Climate vulnerability of rural households in the central exposed coastal areas of Bangladesh.

Journal of environmental management·2026
Same journal

A tri-axis optomechanical accelerometer with plasmonic MIM waveguide and structural direction-dependent optical signatures.

Scientific reports·2026
Same journal

Holographic leaky-wave antennas with independently controlled multiple counter-rotating vortex beams.

Scientific reports·2026
Same journal

Differential associations of longitudinal hearing and vision trajectories with dementia and mild cognitive impairment in older adults.

Scientific reports·2026
Same journal

Abdominal obesity and leisure-time sedentary behavior in relation to gastroesophageal reflux disease risk: a prospective cohort study from the UK Biobank.

Scientific reports·2026
Same journal

Effect of nitrogen-rich COF incorporation on the structure and separation performance of polyamide nanofiltration membranes.

Scientific reports·2026
Same journal

Withanolide A inhibits hIAPP aggregation: An In silico, biophysical, and drosophila-based In vivo validation.

Scientific reports·2026
See all related articles
This summary is machine-generated.

This study introduces BdSLW401, a large Bangla Sign Language dataset, and Relative Quantization Encoding (RQE) to enhance transformer-based sign language recognition for low-resource languages, significantly reducing word error rates.

Area of Science:

  • Computer Science
  • Artificial Intelligence
  • Linguistics

Background:

  • Sign language recognition (SLR) faces challenges in low-resource languages due to data limitations and variations.
  • Existing transformer-based SLR models struggle with signer and viewpoint variability.

Purpose of the Study:

  • Introduce BdSLW401, a large-scale, multi-view Bangla Sign Language (BdSL) dataset.
  • Propose Relative Quantization Encoding (RQE) to improve transformer-based SLR performance.
  • Enhance model interpretability and address limitations of fixed quantization.

Main Methods:

  • Developed BdSLW401 dataset with 401 signs and over 100,000 video samples.
  • Introduced RQE, a structured embedding method anchoring landmarks and quantizing motion.
Keywords:
Bangla sign language datasetRelative quantization embeddingsTransformer-based SLR

Related Experiment Videos

  • Proposed RQE-SF variant for improved pose consistency.

    • Main Results:

    • Achieved significant Word Error Rate (WER) reductions: 44.3% in WLASL100, 21.0% in SignBD-200.
    • Demonstrated RQE's effectiveness on various benchmark datasets.
    • RQE improved model interpretability by focusing on key articulatory features.

    Conclusions:

    • BdSLW401 serves as a benchmark for low-resource SLR research.
    • RQE significantly advances transformer-based SLR for low-resource languages.
    • Adaptive encoding strategies are necessary for large-scale datasets.