SynTalk fuses camera vision and a sensor glove to translate sign language
AlphaBox05 · reddit · 2026-11-07
Developer Garv Goel built SynTalk, an assistive sign-language system combining wearable hardware, computer vision and ML to convert gestures into text and speech. The key engineering challenge: merging two very different inputs — MediaPipe hand landmarks from a camera and flex-sensor readings from a glove — solved by keeping their feature extraction and classifiers separate behind a shared interface. He also adds Hindsight as a persistent memory layer for cross-session context and personalization. Code is on GitHub with a live demo site; full write-up on Medium.
Related event: Developer Open-Sources SynTalk Sign-Language-to-Speech System(3 posts)→
More from Embodied
- Jim Fan teases 'something extraordinary' CoRL keynote at Austin on Nov 9 — DrJimFan · 2026-11-08
- Removing negative-advantage samples drops robot zip-tie success from 86% to 60% — micoolcho · 2026-11-08
- PIVOT, a robot policy improvement method from deployment experience, accepted to CoRL 2026 — ZeYanjie · 2026-11-08
- Robotics roundup: Agility IPOs with $465M, Bedrock raises $600M Series C — DJiafei · 2026-11-08
- MuJoCo 4 released: Studio & Filament renderer, Deformables, Speculative Contacts — bousmalis · 2026-11-08
- Unitree factory tour sparks debate: China's robot factories are still human-powered — garrytan · 2026-11-08