Trusted Voice Project
Incubated at CPIT
Voice is how most of the world communicates — yet the voice data shaping AI today was overwhelmingly collected from a narrow slice of human speech: Anglophone, urban, Global-North-recorded. The communities whose voices are missing are the same communities most exposed to the harms AI is producing, and least equipped to participate in the technologies it is creating.
CPIT is incubating the Trusted Voice Project, a multi-region coalition for ethically sourced, culturally representative voice data across Africa, Latin America, and Asia. TVP brings together regional voice-data efforts, language communities, civil-society organizations, and policy partners under a shared charter, methodology, and governance framework — with the conviction that the people whose voices are at stake should be the people who anchor the data, the design, and the governance of how that data is used.
The project is structured around three pillars:
- A federated coalition of Global South regional partners, each retaining their own identity and operations, connected through a shared charter and convening cadence.
- A shared voice data methodology designed as a public good — capturing dialect, register, and linguistic texture, and built from day one to serve AI inclusion, cultural preservation, authentication, and synthetic-audio defense.
- A standards and policy workstream that translates coalition learnings into transferable frameworks for election commissions, telecom regulators, central banks, and international standards bodies.
CPIT anchors the project as fiscal sponsor, legal home, ethical overseer, and secretariat — providing the connective infrastructure that lets regional efforts contribute to, and benefit from, a shared Global South framework.
The Trusted Voice Project is led by Pamposh Raina, Responsible AI Advisor at CPIT.
The Trusted Voice Project is currently in its foundational phase. We welcome conversations with regional partners, funders, and researchers.
Get in touch → contact@cpitpbc.com
