Speaking a New Language: Bringing My Voice Library to Italy

The Challenge

One in two children with cerebral palsy find speech difficult or impossible, a condition known as dysarthria. My Voice Library, which Kablamo originally designed and built with CPA in 2023, was developed to close a data gap that has historically held back assistive technology for these children:

• There simply hasn't been enough dysarthric voice and lip-movement data available for researchers to work with Dysarthria is a global condition, not one confined to English speakers, limiting the reach of any English-only research dataset CPA needed the platform extended to Italian to open the door for researchers in Italy to begin building a dataset of their own, not as a one-off translation job but as a durable foundation for future languages.
• The upgrade had to preserve strict data sovereignty, with voice and lip-movement recordings uploading directly from a user's device to region-specific storage kept isolated per country and researcher, never routed through CPA's own infrastructure. And CPA needed more than a one-off fix for Italian: they needed a foundation flexible enough that the next language wouldn't mean rebuilding the platform again.

The Approach

Kablamo brought product design, DevOps, and front and back-end engineering to the upgrade, working closely with a CPA-nominated product owner and with families, researchers, and users involved in co-design and validation throughout. Region-specific cloud storage was configured with cross-account access, so the platform can write a researcher's data into their own region, such as Milan for Italian users, without ever being able to read it back out, keeping every region's data isolated and compliant with local privacy requirements.

Claude Code carried the bulk of the multilingual refactor, working systematically through the existing codebase to build in language switching and region-aware data handling while keeping regression risk low. Alongside that core rebuild, it resolved existing bugs, added support for multiple recordings per sound to ease listener fatigue on repeated prompts, and helped the team evaluate speech-to-text options to find the best fit for Italian voice data.

Business Outcomes

As a not-for-profit, CPA needs every dollar of engineering spend to go as far as possible. The upgrade delivered value in four ways, starting with 100% of planned requirements delivered 25% ahead of schedule, at no extra cost.

• Speed to value: Every planned engineering requirement was delivered in six of the eight scoped weeks, 25% faster than planned. Rather than closing early, the team put the remaining two weeks (a quarter of the engagement) into UX polish, a more thorough testing pass and stronger technical documentation. CPA received more than the original scope within the same budget, with $0 in additional spend.

• Research reach: Researchers in Italy can now build their own dysarthric voice and lip-movement dataset on the platform. Each new language widens the pool of data available to assistive technology researchers, which is the gap My Voice Library was built to close.

• Scalable expansion: The multilingual framework turns future languages, other than right-to-left scripts, into largely a translation exercise rather than an architecture project. Adding a language is now expected to take around 2 weeks instead of 6. Expansion into a new country is now a funding decision for CPA, not a rebuild.

• Trust and compliance: Recordings upload directly from the user's device to isolated storage in the researcher's own region, and the platform can write data but never read it back. That means zero cross-region data transfers and zero read access to stored recordings. This gives CPA a data sovereignty model it can present to research partners and ethics bodies in each new country, removing a key barrier to future partnerships.