AiSL

Real-time speech-to-sign translation

Role
Product designer
Origin
Amazon org-wide hackathon
Scope
0 → 1 product design

Deaf and hard-of-hearing employees in live meetings lacked access to the full context of the conversation, causing them to miss important information shared by speakers. The project was recognized for recognized for best UX and best accessible use of AI among 20+ competing projects.

"What other people don't realize is that though closed captions are helpful, they are not as good as ASL (American Sign Language). We are losing a lot of the emotional tone and nuances of the speaker."

— Viewer with hearing impairments
The proposal
An AI-powered real-time speech-to-sign language translation system that converts spoken audio into ASL using lifelike animated avatars, designed to operate within the timing constraints of natural conversation.

Design highlights

Latency-first interaction model
Latency-first interaction model Every design decision was tested against whether it preserved conversational flow, not just translation accuracy

Reflection: a hackathon timeline meant I optimized for proof-of-concept over edge cases — next iteration needs testing for International Sign (IS), background noise, and multi-speaker overlap.

All work TIBCO