Public AI Network Seminar Series - May 2025

Public AI Network Seminar Series - May 2025
Photo by Marcos Luiz Photograph / Unsplash

It was a privilege to talk to the Public AI Network! https://publicai.network/

The Public AI Network (PAINT) is a coalition working to bring about public AI. We aim to:

  • Ensure public capacity-building is part of the conversation about AI design, policy, and funding
  • Make it easier to build public AI by coordinating research efforts in the ML community
  • Support policymakers and technical teams seeking to implement public AI
  • Organize the broader movement for public AI
Public AI Seminar
A research seminar to study public AI and other forms of public interest AI.
Public AI NetworkPublic AI Network

I'm sharing an AI-supported read-out of my talk below. The content is extracted from my slide deck by Claude.

Mozilla Common Voice: Technology That Speaks Your Language

In a world where voice-enabled AI systems increasingly shape our daily interactions, one fundamental question emerges: who gets to participate in building the future of speech technology? Mozilla Common Voice offers a compelling answer—everyone should.

What Is Common Voice?

Launched in 2017, Mozilla Common Voice represents a community-centred approach to building AI training data. Rather than leaving voice data collection to corporations or market forces, Common Voice operates as self-service community infrastructure.

The project embodies several key principles:

  • Community-led governance: From nomenclature and domains to representation and mobilization approaches, communities drive every aspect of the process
  • Public domain commitment: All data is released under CC0 license, ensuring maximum accessibility
  • Digital public good status: Officially registered as a Digital Public Good, reinforcing its mission to serve global communities
  • Radical inclusivity: The platform prioritizes languages and dialects that commercial entities might overlook

Who Gets Involved, and Why?

Common Voice attracts a diverse coalition of participants, each with distinct motivations:

Open Source Advocates

Technology enthusiasts and developers who recognized that proprietary data was monopolizing innovation. They ask: "Who can afford the vendors?" Their solution: democratize the data itself.

Language Activists and Researchers

Government researchers, linguists, and cultural preservationists working to revitalize intergenerational language transfer and preserve cultural heritage. For them, Common Voice offers tools to ensure their languages don't disappear in the digital age.

Local Innovators

Start-ups, nonprofits, and developers with voice-enabled feature needs driven by regional markets, literacy considerations, or personal passion projects. These builders need data that reflects their communities' actual speech patterns.

How It Works: A Seven-Step Community Process

Common Voice operates through a crowd-sourced pipeline:

  1. Language Request: Community members request new language support
  2. Website Localization: Volunteers translate the platform interface
  3. Sentence Collection: Communities gather appropriate text for voice recording
  4. New Language Launch: Mozilla activates the Common Voice site for that language
  5. Voice Contribution: Community members record themselves reading sentences
  6. Voice Validation: Other community members validate the quality of recordings
  7. Dataset Release: Mozilla publishes updated datasets every three months

This process ensures that each language implementation reflects genuine community needs and cultural nuances.

Real-World Impact: The MabelAI Case Study

People often ask us who uses Common Voice data. Hundreds of thousands of people! Here's a quick case study from our downloader community.

MabelAI's mission:

  • Preventing preventable deaths due to language barriers in healthcare settings
  • Ensuring privacy through fully confidential, on-device transcription and translation
  • Enabling offline functionality for use in remote areas without network connectivity

As MabelAI founder Karolina Sjöberg Jabbar explains: "MabelAI is uniquely compatible with data security requirements in healthcare. It can also function without a network, allowing for use in fieldwork in remote areas. We are honored to both use and support the Mozilla Common Voice project, as we share the same mission of inclusive speech AI."

Impressive Scale and Reach

The numbers speak to Common Voice's global impact:

  • 300 languages supported across the platform
  • Over 5.1 million downloads of voice datasets
  • Continuous growth in both language diversity and community participation

Addressing the Hard Questions

Common Voice doesn't shy away from complex ethical considerations surrounding AI public goods. The global NLP community expresses varied perspectives:

Concerns about exploitation:

  • "Why should I create data for megacorps to hoover up?"
  • "My people have been exploited for hundreds of years—we are done giving up our heritage to western orgs"

Privacy and safety worries:

  • "What if my voice gets cloned? How do I know that I am safe?"

Economic realities:

  • "I can hardly pay for my family's food—I cannot do anything for free"

Community empowerment motivations:

  • "This is my data, an investment in my children's future—it's not for sale"
  • "We don't want to be left behind—my grandchildren cannot speak with me—we want tools to help them learn"

Philosophical differences:

  • "Restrictions just limit the good this can do. There will always be the odd bad egg"

These diverse viewpoints highlight the complexity of building truly inclusive AI systems while respecting community autonomy and addressing legitimate concerns.

Evolution and Future Directions

Common Voice continues evolving:

Beyond Open Licensing

The project now offers community licenses and frameworks including Creative Commons variations, Nwulite, Esuthu, and custom licensing options, giving communities more control over their data usage.

Beyond Scripted Monologue

Future developments include spontaneous speech collection, code-switching examples, dialogue datasets, and domain-specific small language model datasets.

Microservices & Open APIs

Common Voice is transitioning toward infrastructure-as-a-service, enabling white-labeling and customization for specific community needs.

Mozilla Data Collective

This emerging initiative follows a "Create. Curate. Control." philosophy, providing hosting, sharing, and governance frameworks for datasets from multiple communities beyond just Mozilla's efforts.

The Bigger Picture: Democratizing AI

Mozilla Common Voice represents more than a voice dataset—it's a proof of concept for community-driven AI development. In an era where AI capabilities increasingly determine economic and social opportunities, Common Voice demonstrates that inclusive technology development is both possible and practical.

The project challenges us to reconsider fundamental questions: Who should control the data that powers AI systems? How can we ensure that technological advancement serves all communities, not just those with the most resources? What does genuine community consent look like in AI development?

By putting communities in the driver's seat, Common Voice offers a pathway toward AI systems that truly speak everyone's language—literally and figuratively.

Interested in contributing to Mozilla Common Voice or learning more about community-driven AI development? Visit the Common Voice website or reach out to explore how your community can participate in building more inclusive speech technology.