Page 62 - AI for Good-Innovate for Impact

AI for Good-Innovate for Impact

Use Case – 11: Libras Project

Country: Brazil

Organization: Lenovo

Contact person: Hildebrando Lima, hlima1@ lenovo .com

11�1� Use case summary table

Domain Accessibility

Problem to be addressed Communication with the hearing impaired
Key aspects of the solution Sign language (Libras) to text (Brazilian language), (text to sign
language) using avatars
Technology keywords Classifier Model、Digital Avatar

Data availability Private

Metadata (type of data) Articulation points, Movements, Vectors
Model Training and fine Classifier
tuning
Testbeds or pilot deploy- PC support using sign language, via Lenovo portal
ments

11�2� Use case description

11�2�1� Description

Introduction: Communication barriers for the hearing-impaired pose significant challenges in
daily interactions, hindering their ability to engage effectively with others. Sign language serves
as a primary means of communication for individuals using Libras (Brazilian Sign Language),
but the lack of widespread understanding can lead to misunderstandings and isolation. This
use case addresses these challenges by leveraging AI-driven technology to facilitate seamless
communication between individuals using sign language and those using spoken language.

Solution Overview: The solution focuses on developing a robust translation tool that translates
sign language (Libras) into Brazilian text and audio, and vice versa, using digital avatars[2].
Using a classifier model and digital avatars, the solution accurately captures and translates
sign language gestures into written and spoken language, bridging the communication gap
between the hearing impaired and the broader community. Access to private data, including
vectors from individuals using Libras, enables the training and fine-tuning of the classifier
model to ensure accurate translation. GPU resources are utilized for efficient processing,
while metadata such as articulation points and movements enhance translation accuracy. Pilot
deployments, including PC support via the Lenovo portal, validate the solution's effectiveness
in real-world scenarios.

46