Communication powers connection, learning, and self-expression. For many children with cerebral palsy, even simple sentences become a guessing game — more than half of their spoken words can be unintelligible to those around them. Kablamo partnered with Cerebral Palsy Alliance and AWS as one team on a shared mission: build the world-first platform that would change that. The result was My Voice Library — a gamified, privacy-preserving voice data collection platform that creates personalized synthetic voice models from just 30 minutes of a child's own recordings. Delivered in 23 weeks. Winner of the 2023 Good Design Social Impact Award.
Industry
Government & public sector
Service Offerings
AI & Machine Learning Engineering | Data Design | Custom Software Development | UX, Research & Design
Technologies Used
ElevenLabs (professional voice cloning), AWS Lambda, DynamoDB, API Gateway, TypeScript, React
The barrier wasn't the technology.
It was the data. AI speech models and assistive communication tools had long hit a wall with dysarthric speech - the training data simply didn't exist. No one had ever built a high-quality dataset of children's dysarthric voices at scale. Without real samples capturing each child's unique speech patterns, AI models couldn't learn to understand or replicate them. As a result, most voice assistive technology was built for typical voices, and sounded like it. Cerebral Palsy Alliance had a vision for how to close the gap. They needed a partner with the technical depth and the human instinct to make it real - fast.
Kablamo, Cerebral Palsy Alliance, and AWS came together as one team - collaborators with shared skin in the game.
CPA brought deep knowledge of the children, families, and the research landscape. AWS brought cloud infrastructure. Kablamo brought the engineering and an approach that made the difference. The design challenge mattered as much as the technical one. Collecting voice data from children with complex communication needs had to feel like play, not a clinical process. The team built a gamified recording interface designed with the children in mind at every step, making the sessions engaging rather than arduous. Technically, the platform integrated ElevenLabs professional voice cloning, building personalized synthetic voice models from as little as 30 minutes of recorded audio. The AWS serverless architecture (Lambda, DynamoDB, API Gateway) ensured the platform was secure, privacy-preserving, and built to scale with the research community's needs. The project ran a full Discovery and Design phase with speech pathologists, engineers, and researchers before a single line of production code was written, ensuring what was built would actually work for the children using it.
A world-first: a high-quality dysarthric voice dataset created from real children's voices to provide the foundational resource the research community had never had. Personalized voice models built from just 30 minutes of audio, capturing emotional nuance and each child's individual speech characteristics. In pilot testing, trained speech pathologists could only distinguish the synthetic voice from the child's real voice 70% of the time — a remarkable result for first-generation models built on data that didn't exist before this project. Winner of the 2023 Good Design Social Impact Award. Published as an AWS case study. My Voice Library isn't a proof-of-concept. It's a living platform delivering measurable benefits for children, families, and researchers today, and laying the groundwork for what comes next: adaptive AAC devices, AI models finely tuned to dysarthric patterns, and communication tools that actually meet children where they are.
