Скачать или смотреть Eldorado Voice2Sign & Lenovo AI: Real-Time Brazilian Portuguese-to-Libras Avatar, Cloud GPU Pipeline

Voice2Sign, presented by Instituto Eldorado in partnership with Lenovo during AI For Good #aiforgood 2025 Geneva, converts spoken or written Brazilian Portuguese into Libras, the national sign language, by rendering a photorealistic avatar that signs the message on-screen. The prototype tackles the communication gap faced by Brazil’s 2.3 million Deaf citizens and demonstrates how configurable avatars, background colors and voice options improve inclusion across regions and age groups. https://www.eldorado.org.br

Under the hood, a transformer-based automatic-speech-recognition front end feeds a text-to-sign neural translation model that outputs tokenized gloss and non-manual markers. These parameters drive a real-time synthesis stack built on the open-source glTF rigging standard; final frames are composited on Lenovo ThinkSystem SR670 V2 GPU clusters or edge servers, depending on deployment. Current processing time per sentence ranges from two to five minutes, and the team is targeting sub-second latency through quantization and mixed-precision inference.

Achieving linguistic coverage required thousands of hours of motion-capture footage annotated with region-specific gloss variants—north-eastern dialects mark “doctor” differently from southern ones—so the researchers built an active-learning loop that flags unseen signs and retrains the model weekly. All raw video remains on-premise to meet Brazilian data-sovereignty rules, while a privacy-preserving API exposes only key-point vectors to the cloud, aligning with edge-AI principles and Lenovo’s Product Diversity Office guidelines for demographic equity.

Hospitals, banks and conference venues are piloting Voice2Sign to replace written captions for users with limited literacy, and YouTube integration is next on the roadmap. By fusing speech-to-sign, regional dialect adaptation and scalable compute, the project signals a practical path toward barrier-free audiovisual communication in the near future

My full AI For Good 2025 video coverage is here:    • AI For Good 2025 video coverage  

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here    • Filmed with DJI Pocket 3  

Join https://www.youtube.com/charbax/join for Early Access to my videos and to support my work, or you can Click the "Super Thanks" button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning:    • Daylight Computer CEO interview: Healthy C...