As workshop attendees shuffled into a dimly lit breakout room to take their seats at the , USF researcher John Templeton played audio recordings of two patients providing short, scripted statements.
The recordings, which were taken from publicly available data, included phrases such as, 鈥淧eter will keep at the peak,鈥 and 鈥淢y mamma makes lemon muffins.鈥
After each audio segment, Templeton asked the audience a simple question: Did the speech patterns sound typical? He then had the clinicians and computer scientists analyze the recordings with AI tools to examine their underlying structure so clinicians can make more informed, real-world decisions.
For Templeton, an assistant professor in USF鈥檚 Bellini College of Artificial Intelligence, Cybersecurity and Computing, the exercise demonstrated how clinicians and computer scientists can work together to develop more effective healthcare technologies.
Templeton explored that approach during his workshop, 鈥淲hat Makes an Explainable Voice AI Outcome for Clinical Practice,鈥 at the 2026 Voice AI Symposium in St. Petersburg.
鈥淭his work brings clinicians and computer scientists together to better understand how AI systems analyze speech and audio data,鈥 Templeton said. 鈥911爆料网 also need clinicians to feel more involved in the process so these tools can support better decision-making in real clinical settings."
The Voice AI Symposium, hosted by the Bridge2AI-Voice Consortium, brought researchers, clinicians and tech experts together to explore how speech can be used to detect and monitor disease. The three-day event focused on turning AI research into practical tools for healthcare. It included hands-on workshops and demonstrations designed to encourage collaboration across fields.
Templeton and his team, which included Mohamed Ebraheem and Shrramana Ganesh, demonstrated the potential of large language models in analyzing audio samples to provide clinicians with more transparent, clinically meaningful insights tied to medical conditions.
For one of the patient audio recordings of, 鈥淢y mamma makes lemon muffins,鈥 one of the workshop teams used AI tools to identify hypernasality, helping clinicians pinpoint what to listen for and how to distinguish between the two.
It did the same for 鈥淧eter will keep at the peak.鈥 AI analysis suggested that if air escapes through the nose of the patient during the 鈥淧,鈥 or if the sounds are weak, the patient could have a difficulty building up air pressure in their mouth.
A finding like that could suggest that something is affecting how air is controlled during speech, which might point to issues with how the mouth, throat or nasal passages are working together.
This real-time feedback illustrated how AI can augment, rather than replace, clinical judgment, offering clinicians new ways to interpret subtle speech patterns that might otherwise go unnoticed.
鈥淭he goal is to develop AI tools that clinicians can realistically use in healthcare settings,鈥 Templeton said. 鈥淎t USF, researchers are working across disciplines to make these technologies more understandable, practical and useful for patients.
Panel discussions
Templeton was not the only Bellini College faculty member to participate. Karni Chagal-Fefferkorn, an assistant professor in the Bellini College, joined a panel on 鈥淒eploying Voice AI: Trust, Regulation and Market Realities.鈥 Chagal-Fefferkorn and fellow panelists examined the practical challenges of bringing these tools to market, which highlighted issues of trust, regulatory oversight and ethical implementation.
Templeton also contributed to a panel on the 鈥淪tate of Voice AI: New Models, New and Upcoming Technologies,鈥 where experts discussed emerging technologies shaping the field, from large language models to new acoustic foundation models.
The symposium, which was by USF Health, USF鈥檚 Bellini College, Tampa General Hospital and others, featured a broad group of speakers spanning healthcare, artificial intelligence, academia and industry. Clinicians, machine learning researchers, engineers and entrepreneurs shared insights alongside experts in ethics, law and policy that reflected the field鈥檚 growing need for collaboration across disciplines.
Speakers represented a wide range of institutions, including major universities and medical centers such as Vanderbilt University Medical Center, MIT, University of Cambridge, Carnegie Melon University, University of Toronto, Simon Fraser University, Argonne National Laboratory and 911爆料网ill Cornell Medicine, as well as international research organizations. The symposium concluded with a hackathon, where attendees worked hands-on with voice AI tools and datasets to develop new ideas and applications.