Siri Peddinti, a senior at Plano East Senior High School, has been named a national first-place winner at the 2025 Junior Science and Humanities Symposium (JSHS) for her research on detecting major depressive disorder using artificial intelligence and vocal analysis.
Held April 22–26 in Chantilly, Virginia, the 63rd annual JSHS brought together 239 of the top high school researchers from across the United States, Puerto Rico and Department of Defense Education Activity schools worldwide. Peddinti took first place in the Biomedical Sciences oral presentation category, earning a $12,000 scholarship.
Her project, titled “Vocalyze,” uses deep learning to analyze vocal patterns and word usage to detect signs of depression and track treatment progress.
“I was inspired to look into vocal inflections and depressive disorders after learning that the current diagnostic systems rely almost entirely on unstandardized questionnaires,” Peddinti tells Local Profile. “I knew from prior research that there are various surprising biomarkers that can indicate the presence of neurological disorders, and in my research, I found that vocal inflections were one of the most reliable markers.”
What Is JSHS?
The Junior Science and Humanities Symposium is a Department of Defense–sponsored program designed to encourage original STEM research among high school students. Finalists compete for scholarships at regional and national levels, with the national event offering more than $200,000 in total awards this year.
“Each year, the students at JSHS redefine the possibilities in STEM,” said Winnie Boyle, senior director of competitions, NSTA, in an official statement. “Their groundbreaking research and unwavering curiosity inspire us all, and we are honored to support their academic journeys with scholarships and awards.”
Peddinti was one of eight first-place oral presentation winners. Other top projects included an AI-powered system to assess honeybee health, a toolkit for monitoring Parkinson’s symptoms via telehealth, and new methods for identifying biomarkers in breast cancer.
How Does Vocalyze Work
Peddinti’s tool, Vocalyze, uses deep learning to identify major depressive disorder (MDD) through vocal acoustic inflections and sentiment analysis. The system listens to how a person speaks and analyzes both vocal tone and word choice, detecting subtle indicators often missed by traditional methods. It was trained on more than 187,000 audio samples from both MDD and non-MDD patients and tested on nearly 47,000 others. Paired with data from PHQ-9 questionnaires, the model achieved over 98% diagnostic accuracy.
Vocalyze combines two models: one to evaluate vocal tone and inflection, and another to analyze the words patients use. “The goal is to provide an automated, objective, scalable method to detect depression and effective monitoring without increasing pressure on patients or healthcare providers,” Peddinti said.
Peddinti is currently working on a mobile app version of the system that allows users to track their mental health daily through short voice recordings. Her long-term vision is to make early detection of MDD both scalable and practical, integrated into routine healthcare without placing additional burden on patients or doctors.
Challenges And Surprises
One of the biggest hurdles was cleaning background noise from audio recordings. “It affected how accurately the model could identify the prosodic features and tone,” Peddinti explained. After testing multiple preprocessing methods, she found one that significantly improved performance.
She also built a multi-output model capable of evaluating both depression presence and severity. “It required a lot of trial and error to fine-tune,” she said.
Setting Vocalyze Apart
Peddinti believes that what makes Vocalyze stand out is its ability to combine speech patterns with sentiment analysis in an entirely automated process. By removing the need for patient self-reporting — currently the standard in diagnosing depressive disorders — the tool avoids the subjectivity and bias that can accompany human evaluation. It also helps differentiate between MDD and other neurological conditions like bipolar disorder, where symptoms often overlap but manifest differently in vocal patterns.
“Vocalyze is non-invasive, user-friendly and cost-effective,” she said. “That means every patient visiting their primary care provider could be screened for depression quickly and accurately, before symptoms escalate or go unnoticed.”
Looking Ahead
Peddinti plans to continue her research at the University of Pennsylvania, where she’ll pursue dual degrees in bioengineering and business through the Jerome Fisher Management and Technology program. She hopes to expand Vocalyze with facial recognition or physiological data and work at the intersection of machine learning and neurology.
“I hope to develop affordable technology that can aid individuals with neurological disorders,” she said.
Advice For Aspiring STEM Researchers
“Just start where you’re interested,” Peddinti said. “You don't need to have all the answers to explore something meaningful—pick a problem that you are genuinely passionate about and research can become your most enjoyable pursuit. Research requires a lot of time and commitment, but it’s collaborative at its core. Surround yourself with people who share your passion and stay open to learning. Participating in science fairs at every level has had an immeasurable impact on my research and on me as a person.”
Abstract submissions for the 2024-2025 JSHS competition will open in the fall. For more information on JSHS, visit www.jshs.org.
Don't miss anything Local. Sign up for our free newsletter.