ReadingStar 2.0
Helping children find their voice.
Overview
A closed-source project developed at UCL in collaboration with IBM, Intel, and the National Autistic Society. ReadingStar 2.0 is an application designed to support children with autism in developing their communication skills by encouraging them to sing along to AI-generated songs based on their interests and preferences. The project aims to create a fun and engaging way for children to practice vocalisation and improve their speech abilities. Everything the application does is done locally on the user's device, ensuring privacy and security for the children using it, as well as providing an offline experience that can be used on old and low-spec devices, making it accessible to a wider audience.
The Solution
We approached this problem by using IBM's Granite AI model to generate lyrics, Meta's MusicGen to create music, and packaged everything into a clean and user-friendly application built with Electron. This approach came with its challenges, such as optimising AI models to perform better on low-spec devices, beat matching, scoring the user's singing, all while ensuring that the application is accessible and enjoyable for children with autism. That's where our collaboration with the National Autistic Society was crucial, as they provided us with invaluable insights and feedback to ensure that our solution truly meets the needs of our target users.
I was responsible for the music generation, audio-utility functions, and the scoring system. Rory Byrne built the frontend and plugged all our subsystems together. Doruk Ersoy worked on the lyric generation and the structure of the backend. Sinan Sensurucu designed the beat-matching algorithm, tying the outputs from both models together and ensuring that the generated music and lyrics are in sync. As team leader, he also handled the business end of things.
Additional Details
- Biggest Milestone: presenting our project at the BETT conference on 22 January 2026.
- Currently working on: packaging.
- Gratitude to: our collaborators mentioned above, Professor Dean Mohamedally, and our TAs Ashley Tam and Yusuf Afifi.
