March 7, 2025
Automatic Speech Recognition (ASR) systems have long struggled to accurately recognize and process stuttered speech, largely due to their training on fluency-biased datasets. As a result, people who stutter (PWS) frequently encounter challenges when using voice-activated technologies, automated transcription services, and other speech-based AI tools. This lack of inclusivity and accessibility also reinforces social and technological barriers for people with speech differences.
To address this issue, our project— funded by Patrick J. McGovern Foundation—aims to improve ASR models by creating a high-quality dataset of English stuttered speech. The goal is to collect approximately 50 hours of speech recordings that authentically represent the voices and speech patterns of people who stutter. Rather than approaching this as a purely data-driven initiative, we seek to center the agency and collective power of the stuttering community, ensuring that the dataset is created by PWS, with PWS, for PWS.
<aside> 💡
A core principle of this project is to design a data collection process that is non-extractive but empowering for both individual participants and the stuttering community as a whole. This means carefully considering factors such as participant recruitment, fair compensation, data governance, and long-term community impact.
</aside>
To inform our execution plan, we conducted interviews with eight stuttering advocates, gathering their insights on how to structure a data collection effort that is ethical, empowering, and inclusive.
This report synthesizes the perspectives of eight experts—including community leaders with lived experience of stuttering, academic researchers, and advocates—on key aspects of the stuttered data collection process. Their insights guide designing an approach that fosters trust, ensures fair participation, and ultimately contributes to a more representative and inclusive ASR technology for people who stutter.
We interviewed eight stuttering advocates to get diverse perspectives on the community-centered approach to the collection of stuttered speech data. Each person brings a unique perspective shaped by their lived experiences of stuttering, professional expertise, and advocacy efforts. Below is a detailed overview of their backgrounds.
Nan Bernstein Ratner: Nan is a professor in the Department of Hearing and Speech Sciences at the University of Maryland, College Park. Her primary areas of research are fluency development and disorder (stuttering), psycholinguistics and the role of adult input and interaction in child language development. Nan has been an expert in large-scale stuttered speech datasets, the co-founder of FluencyBank, one of the most widely used corpora for stuttering research.
Maya Chupkov: Maya is the founder of Proud Stutter, an award-winning podcast about changing how we understand and talk about stuttering, one conversation at a time. As a woman who stutters, she is shifting societal norms around stuttering and lifting up the voices of those with verbal differences. Maya is also a filmmaker who is creating a feature-length documentary about two Black men who stutter. The film weaves how these men come to terms with their disability, how their disability is compounded by race, and how it impacted them growing up.
Aidan Sank: Aidan is the Executive Director and co-founder of SPACE, a nonprofit organization working to create more space for stuttering and change the way the world listens. Aidan has been actively involved in multiple community-based initiatives aimed at increasing the visibility and agency of PWS. He has been working with the stuttering community for over 15 years and has a background in storytelling and the arts.
Naomi Rodgers: Naomi is a researcher at the University of Iowa, where she leads the Iowa Stuttering Lab and teaches courses in counseling, clinical methods, and stuttering therapy for speech-language pathology (SLP) students. She has been deeply involved in stuttering research, clinical intervention, and community support for years, frequently working with people across the spectrum of stuttering severity. Beyond academia, She runs a local chapter of the National Stuttering Association (NSA) in Iowa City. She is also an active volunteer for Friends: The National Association of Young People Who Stutter, regularly attending their annual conventions and supporting one-day workshops across the country.
Edmund Metzold: Edmund is a senior production support specialist in healthcare technology. Edmund is a person who stutters and has extensive involvement in the stuttering community, including leadership roles in the NSA Boston chapter and Passing Twice, an LGBTQ+ and stuttering support organization.
Jia Bin: Jia is a PhD student researcher in Communicative Sciences and Disorders at Michigan State University, specializing in stuttering research. As a Chinese woman who stutters, she has actively engaged in both American and international stuttering communities, advocating for greater visibility and support for people who stutter (PWS) across different cultural contexts. Jia founded and leads the Spartan Stuttering Group, which serves MSU students and faculty. She previously held the role of NSA Regional Coordinator, helping to expand and strengthen support networks for PWS. Jia also co-founded StammerTalk (口吃说), an online community for Chinese-speaking people who stutter.
Rong Gong: Rong is a person who stutters and a co-founder of the StammerTalk community. Rong is a research scientist at a large technology company. With extensive experience in stuttering advocacy and technical background in machine learning, Rong has led the development of AS-70: the largest Mandarin stuttered speech dataset for ASR and stuttering event detection.
Tracy Wang: Tracy is a person who stutters and an organizer of the StammerTalk community. She is a data scientist at a large retail company. She has also led the development of AS-70, the largest Mandarin stuttered speech dataset.
We conducted semi-structured interviews with stuttering advocates over Zoom and each interview lasted between 60–90 minutes. Interviews were recorded with participants’ permission. We provided each participant with a $50 Amazon e-gift card as a token of gratitude. The interview focused on the following key areas: