TLFS23 Tamil language fingerspelling dataset

Area of Science:

• Computational Linguistics
• Computer Vision
• Deep Learning

Background:

• Tamil, an ancient language with 65 million speakers, presents unique communication challenges.
• The Speech and Hearing Impaired community requires specialized tools for effective communication.
• Existing vision-based translators lack comprehensive datasets for complex languages like Tamil.

Purpose of the Study:

• To introduce the Tamil Fingerspelling dataset (TLFS23) for research in computer vision and deep learning.
• To facilitate the development of automated translation systems for Tamil fingerspelling.
• To improve communication accessibility for Tamil-speaking individuals with speech and hearing impairments.

Main Methods:

• Collected a large-scale, labelled dataset of Tamil Fingerspelling gestures.
• The dataset comprises 2,55,155 images across 248 classes, representing 247 Tamil characters.
• Images were captured from 120 individuals of diverse age groups, ensuring varied representations.

Main Results:

• The TLFS23 dataset provides 1,000 images per unique finger flexion for each Tamil character.
• This comprehensive dataset enables robust training of deep learning models for fingerspelling recognition.
• The dataset is publicly available, promoting further research and development.

Conclusions:

• The TLFS23 dataset is a significant contribution to the field of assistive technology and computational linguistics.
• It paves the way for advanced vision-based translators for Tamil fingerspelling.
• This resource has the potential to bridge communication gaps for the Speech and Hearing Impaired Tamil-speaking population.