HeardAI is a technology and advocacy project with the mission of making voice AI more broadly usable and equitable. We are initially focusing on including the voices of people who stutter in the development of speech recognition technology for everyday tech devices. Ultimately, our goal is to create technology that is accessible to all, including those with other types of speech differences.

Voice AI technology is everywhere, from Siri and Alexa to phone trees and online interviews. Unfortunately, it cannot accurately understand the speech of the more than 80 million people worldwide who stutter. Ensuring that this technology is accessible is crucial for minimizing added barriers for those who stutter.

In a recent survey, over 50% of people who stutter reported that current voice AI is “not at all” or only “minimally” beneficial in their day-to-day lives. Many find using voice AI frustrating, embarrassing, and even anxiety-provoking. Most people who stutter indicated that if voice AI were accessible, they would use it more, and their lives would be improved as a result.

While our initial focus is on benefiting people who stutter, we aim to expand our work to support all individuals with speech differences. Additionally, because everyone experiences speech disfluencies at times, HeardAI will ultimately benefit everyone!

The stuttering community is integral to our work. People who stutter serve as partners, advisors, and participants in design decisions. They contribute data for training our AI and provide feedback on all aspects of our project, ensuring that our solutions are effective and truly inclusive. Our leadership team and primary research assistants include people who stutter and allies, thereby reinforcing our commitment to the community and ensuring that their perspectives are represented at the highest level.

Our plans include expanding our work beyond the stuttering community to consider developing voice AI that works well for all speech variations.

Our team consists of researchers from Michigan State University, Western Michigan University, and Friends: The National Association of Young People Who Stutter, specializing in AI, stuttering, hiring and organizational psychology, accessibility, equity, stigma, and the lived experience of stuttering. Our partners include industry, legal experts, non-profit organizations, and advocacy groups.

These three organizations received the initial grant from the National Science Foundation to launch the work of HeardAI. Individuals from these organizations are involved because of their expertise in AI, stuttering, advocacy, and human resources, providing valuable insights and guidance to our project.

Many organizations support HeardAI because they recognize the importance of making voice AI accessible to all. Our mission is strengthened by the active partnership of the stuttering community, ensuring that their needs and perspectives are central to our development process. Additionally, we collaborate with tech companies and vendors to create practical and effective solutions for widespread use.

We would love to have your help in providing samples of stuttered speech, beta testing new products and features, providing feedback on usability and accessibility, and in other ways.

Please complete this form to indicate your interest and get involved.

Yes, we offer compensation for certain types of participant contributions, such as providing speech samples from eligible individuals. Details about compensation will be clearly disclosed in our communication to you before you participate. 

Please complete this form to indicate your interest and get involved.

AI uses data to “train” models. Your speech samples will be used to help the AI understand and accurately process stuttered speech.

Even with our preliminary work, we are already seeing improvements in the accuracy of our speech recognition system. Next, we need to gather a sufficiently large number of speech samples to reach the level of accuracy we aim for in our AI system.

Our algorithms analyze various features of speech to provide accurate transcriptions. By collecting a very large number of samples of stuttered speech, we can incorporate the unique characteristics of stuttering into our algorithms, making them more accurate than those developed using primarily non-stuttered speech.

Preliminary results show that our model has a significantly lower word error rate for stuttered speech compared to current products. It also performs better on metrics that consider the meaning of what the speaker is saying. Our focus on the needs of people who stutter and our close ties with the stuttering community enable us to improve upon related efforts and make voice AI more accessible from both social and technical perspectives.

The app that will understand stuttered speech is currently under development, but you can access the data collection app on a computer or cell phone.

We will provide you with access to the data collection app once you complete this form and indicate your interest in providing speech samples.

Please contact us at: heardai.team@heardai.ai. We look forward to hearing from you.