VLibras — Brazil's Government-Run Sign-Language Avatar Embedded Across 120,000+ Public Websites

VLibras Suite is an open-source (GPL-3.0) machine-translation platform that converts written or spoken Portuguese into Brazilian Sign Language (Libras), rendered through a 3D avatar. It was developed from 2016 by the Digital Video Applications Laboratory (LAVID) at the Federal University of Paraíba, in partnership with the federal government — originally the Ministry of Planning, now the Secretariat of Digital Government under the Ministry of Management and Innovation in Public Services, together with the Ministry of Human Rights and Citizenship.

The suite ships as a browser plugin, desktop application, mobile app and a video-translation portal, and the Brazilian government's Digital Accessibility Guide recommends and embeds it as standard infrastructure across federal digital services, building on the legal basis of Law 10.436/2002 (which recognises Libras as an official language) and Decree 5.626/2005 (which mandates its use in public services). By 2023, official figures put its footprint at over 120,000 Brazilian websites carrying the widget, around 40 million daily accesses and roughly 3 million translations performed per month, drawing on a vocabulary of about 21,000 animated signs and a bilingual corpus of around 100,000 sentences. A companion tool, WikiLibras, lets the Deaf community crowdsource and validate new signs, and the project has been profiled internationally as one of the largest free sign-language translation infrastructures, including recognition by the Zero Project disability-innovation network.

Independent academic evaluation is more cautious than the usage statistics suggest. Brazilian usability and linguistics research (including eye-tracking studies of Deaf users interacting with Libras avatars, and analyses specifically of VLibras) documents recurring limits: avatars translate word-by-word rather than with full Libras grammar, non-manual markers (facial expression, mouth movements) that carry grammatical meaning in Libras are often flat or missing, and translation fidelity drops sharply on long or complex text. Researchers consistently frame VLibras as a valuable access layer that does not substitute for human interpreters on complex content, and find that feeding the system pre-simplified, plain-language text materially improves translation accuracy — a finding since used to refine guidance for public-sector content authors.

Read the full analysis: https://www.gov.br/governodigital/pt-br/acessibilidade-e-usuario/acessibilidade-digital