Tamil Murasu featured MEANT alongside student projects using artificial intelligence for social impact.
← All workDell InnovateFest 2026 · Second runner-up
MEANT
Helping AAC users reply before the conversation moves on.
People who use AAC often lose their turn while they put a reply together. MEANT suggests a few likely replies, and the user still decides what gets said. We worked with SPD Ltd, Singapore.
- My role
- UI and UX, the Singaporean TTS voice with a teammate, and presenting the build
- Result
- Second runner-up, polytechnic category · S$3,000
- Stack
- On-device AI on a Dell GB10 · Singaporean TTS
How it works
What happens when it’s their turn to reply.
- Holding the turnTurn Claim shows the other person that a reply is coming.
- SuggestionsA Dell GB10 makes the suggestions on the device itself, not online.
- They decideThey pick a suggestion, type, or use their board, and it’s only spoken when they tap.
The voice
A Singaporean voice.
Most text-to-speech voices don’t sound Singaporean, so the user ends up sounding like someone else. I worked on a Singaporean TTS voice with a teammate.
The tablet app with the hawker-centre demo words. The code is private.
Without the AI
If the AI goes down, the board and typing still work.
What I did
The screens people use, and getting the live demo to work.
Colour-coded boards
Each type of word has its own colour and an openly licensed symbol.
Search
If the suggestions are wrong, you can search every word instead.
Setup first
The screen ruler and mic permission happen during setup, not in the middle of a conversation.
The live demo
I scripted our turn on stage and fixed a stuck mic, a silent send and a black camera.
All from my commits to MEANT’s private repo.
In the press
Tamil Murasu featured MEANT.
MoreSP School of Computing’s post ↗
MEANT’s symbols: Mulberry Symbols © Steve Lee, CC BY-SA 4.0