As soon as I saw how foundational speech AI models struggled with stuttered speech, I knew this problem could not be unique to people who stutter. Despite being 1% of the population, people who stutter are still being very much marginalized in tech and in the society. As a 4 years old organization with only two full time staff members, staging-7261-aimpowerorg.wpcomstaging.com/ can not change the trajectory of speech AI on our own. We need solidarity, we need allies, we need voices from all communities that are currently silenced by AI, we need the researchers and technologists to join us in this cause. We need a movement.

That’s what we did.

A community forms. Ideas turn into collaborations. And a workshop becomes a movement!

Since 2025, staging-7261-aimpowerorg.wpcomstaging.com/ has been bringing together an interdisciplinary, international community through our Speech AI for All workshops at ACM CHI, one of the world’s leading conferences on human-computer interaction.

What started as a conversation about accessibility and fairness in speech AI has grown into an active coalition working to make speech technology more inclusive, equitable, and empowering for everyone.

White board planing for first speech AI for all workshop at CHI 2025, by Kimi, Colin, Raja, and Shaomei.

Speech AI for All workshop 2025, group photo

🌱 Where It All Began: CHI 2025 in Yokohama, Japan

Our journey began on April 27, 2025, with the first Speech AI for All: Promoting Accessibility, Fairness, Inclusivity, and Equity workshop at CHI 2025 in Yokohama, Japan.

Together with collaborators from universities, industry, and disability communities, we brought together more than 40 participants to explore a fundamental question:

How can we build speech AI that embraces the full diversity of human communication, rather than expecting everyone to speak in the same way?

The workshop brought together perspectives that rarely share the same room: speech AI researchers, HCI scholars, linguists, disability studies experts, industry practitioners, and people whose voices and communication styles have historically been marginalized by technology.

Through thought-provoking keynotes, research posters, community panels, and small-group discussions, we explored not only the technical limitations of speech AI but also the social assumptions, power dynamics, and structural inequities behind them.

Accessibility and community participation were central to how we organized the event. We provided Japanese–English interpretation, sign language interpretation, live captioning, hybrid participation, and financial support for community speakers.

One participant reflected:

“I have never been to such an accessible event and having the opportunity to collaborate with communities that I wouldn’t otherwise be able to communicate with was truly eye opening.”

👉 Read our CHI 2025 workshop recap

📏 From Building Connections to Rethinking Measurements: CHI 2026 in Barcelona

One workshop was not enough!

Building on the connections and conversations sparked in Yokohama, we brought Speech AI for All back for a second workshop at CHI 2026 in Barcelona, Spain.

While our first workshop focused on bringing together diverse communities and perspectives, the second took a deeper dive into a concrete sociotechnical challenge: How should we measure and evaluate speech AI to ensure it works fairly and meaningfully for everyone?

Traditional speech AI evaluations often focus on technical metrics such as word error rate. But these numbers don’t necessarily capture people’s lived experiences, communication needs, or the ways speech technologies can reinforce existing marginalization.

Through lively, cross-disciplinary discussions, we explored different types of measurements and evaluation approaches that could better reflect the diversity of human communication, lived experiences, and marginalized identities. We asked what meaningful success looks like from the perspectives of people who use—and are affected by—speech AI, and how these perspectives can reshape the way technologies are evaluated.

The conversations reinforced an important insight: What we choose to measure shapes what we build. Making speech AI truly inclusive requires not only better technologies, but also rethinking whose experiences and values count when defining success.

[📸 PHOTO GALLERY: CHI 2026 workshop group photo, collaborative discussions, and brainstorming activities]

🤝 Beyond the Workshops: A Growing Speech AI for All Coalition

Perhaps the most exciting outcome of our workshops is what happened after everyone went home.

The conversations didn’t end when the conference rooms closed. Instead, they grew into an ongoing, cross-disciplinary coalition of people committed to transforming speech AI research and development.

Our community at a glance

• 💬 91 members in our active Slack community
• 🌍 Cross-disciplinary connections among researchers, technologists, advocates, and community members
• 📝 A collaboratively drafted position paper outlining a shared research agenda
• 🚀 Future workshops and convenings being co-organized for conferences including CHI and Interspeech 2027

Our Slack workspace has become a place to exchange research, share opportunities, find collaborators, and coordinate new initiatives across institutions and disciplines.

We are especially excited to see new collaborations emerging organically from the community. Rather than organizing isolated annual events, we’re building a space where relationships and collective work can continue year-round.

📝 Turning Conversations into a Shared Research Agenda

One major milestone of the coalition is our collaboratively drafted position paper, “A Cross-Community Agenda for Speech AI.”

Co-authored by 23 contributors across academia, industry, and disability-focused organizations, the paper brings together perspectives from technical speech and language research and sociotechnical fields such as human-computer interaction, accessibility, and responsible AI.

It challenges three fundamental assumptions in today’s speech AI research:

• Communication is more than words. Speech technologies need to account for the rich social, contextual, and multimodal nature of human communication.
• Voice is part of identity. Speech AI should respect how people express who they are, rather than treating differences as errors to be corrected.
• Accuracy isn’t everything. We need evaluation methods that capture real-world communication, user agency, and meaningful outcomes—not just conventional performance metrics.

The paper also proposes directions for research, system design, evaluation, and cross-community collaboration.

Our goal is not simply to make existing speech AI work better for more people. It is to rethink what speech AI should do, whom it should serve, and who gets to shape its future.

📄 Read the position paper: A Cross-Community Agenda for Speech AI

💜 Join Us in Making Speech AI for All!

Whether you’re a researcher, developer, designer, disability advocate, community organizer, or someone with lived experience of speech AI’s limitations, we’d love to connect.

Speech AI should work for the diversity of human voices—not the other way around.

Together, we can make that vision a reality.

Interested in joining the Speech AI for All coalition or collaborating on future activities? Reach out to contact@aimpower.org to get involved!