View single | DIVERSASIA - Best practices | Embracing diversity in ASIA through the adoption of Inclusive Open Practices

Title of practice:	Deep Lip Reading-A Deep Learning Based Lip-Reading Software for the Hearing Impaired
Author/developer:	Mohammed Abid Abrar; A. N. M. Nafiul Islam BUET, Dhaka, Bangladesh ; Mohammad Muntasir Hassan; Mohammad Tariqul Islam; Celia Shahnaz; Shaikh Anowar
Language:	English
Description of good practice:	The BUET team proposed a deep learning-based method for performing visual speech recognition on videos or sequences of color images with variable length. We employed convolutional neural networks (CNN) to this problem. The resulting model is being used to develop an app that can be run on all mobile devices. This can have huge humanitarian impact for the hearing-impaired. The accuracy obtained by our model is comparable to other works done on the MIRACLVC1 dataset. The accuracy can be further improved for phrases and sentences as contextual information plays a key role in real life lip reading. Nevertheless, this work and the accompanied mobile app can be of extreme benefit to the hearing-impaired community and will enable them to communicate in an easier and more natural way in more realworld settings.
Country where the practice is developed:	Bangladesh
URL to the material:	https://ieeexplore.ieee.org/document/9042439/authors#authors
Relevant file:
Type of practice:	Research, Publication
Group(s) targeted by the material:	Students Administrative staff Teaching staff Policy makers
The level of Creative Commons license:	No licensing infromation available
Can the practice be reused?:	No
What is the payment model for this material?:
What is the cost of using this material?:
What barriers does it help to overcome?:	Multiple barriers
Is there anything else you would like to add about this submitted good practice material?:
Tags: