AI model for Libras: an undergraduate thesis crosses the communication barrier with the deaf community
Daniela Soares de Oliveira and Danilo Domingo Vitoriano Silva defended at FCTE/UnB an undergraduate thesis building an AI model for Brazilian Sign Language translation — a technical response to a communication barrier affecting more than 2.3 million Brazilians.
Brasília, September 12, 2024
More than 2.3 million Brazilians have hearing impairment, and the Brazilian Sign Language (Libras) is essential to enable communication. The daily problem is the opposite: most of the population that does not master Libras hits blockers whenever they meet this community. This structural friction is what the undergraduate thesis defended in September 2024 by Daniela Soares de Oliveira and Danilo Domingo Vitoriano Silva addresses with machine learning, under the advising of Prof. George Marsicano Corrêa .
The work develops an AI model for automatic translation of Libras, tackling a classic computer-vision-for-sign-language problem: distinguishing similar gestures, sustaining accuracy under variable lighting and framing, and responding in time to enable dialogue. The approach ties machine learning to a directly social problem — the kind of work the academic community must sustain, because the market alone will not push it.
The result is input for developers willing to build inclusive applications — in public services, in customer support, in classrooms. A model, on its own, does not change the world; but the right model, placed in the right hands, changes who enters a dialogue and who is left outside.
The full text — written in Portuguese — is available at the UnB Undergraduate Theses Digital Library.
About CEDIS: The Center for Studies, Development, and Innovation in Software (CEDIS), linked to the University of Brasília, researches and develops innovative software solutions with strong presence in AI and NLP .