-
-
Ui of vocal guard
Inspiration
Millions worldwide lack access to early and non-invasive health screening, especially for chronic diseases like COPD, stress, and neurological conditions. Inspired by recent breakthroughs in voice-based biomarkers and the desire to democratize healthcare, we set out to create an AI tool that enables anyone to check their health—using just their voice.
What it does
VocalGuard uses advanced AI to analyze 15-second voice recordings, instantly assessing risk for conditions such as COPD and stress. The platform provides clear, actionable feedback directly in a web or mobile app—making health screening fast, private, and accessible from anywhere.
How we built it
We combined Python, TensorFlow, librosa, FastAPI, and Docker to build a scalable backend and intuitive user interface. Our machine learning models (CNNs and Random Forests) were trained and validated on diverse, clinically-annotated datasets like NIH Bridge2AI-Voice and Coswara, ensuring real-world accuracy and robustness.
Challenges we ran into
Achieving high model accuracy across different languages, accents, and recording devices.
Ensuring privacy and HIPAA/GDPR compliance while rapidly processing sensitive voice data.
Creating a fast, lightweight pipeline that works in a real clinical or home setting.
Minimizing false positives/negatives to avoid unnecessary anxiety or false reassurance for users.
Accomplishments that we're proud of
Developed a fully working prototype with real-time voice analysis and instant health feedback.
Achieved 82% accuracy for COPD detection and nearly 77% for stress in pilot testing.
Built a privacy-first, cloud-ready stack with encrypted, transparent data handling.
Secured interest for a hospital pilot to further validate in actual healthcare workflows.
What we learned
Interdisciplinary teamwork—combining AI, clinical knowledge, and user-centered design—is essential for impactful health technology.
Robust data diversity is key to fairness and adoption in AI healthcare.
Clear, empathetic communication is crucial when delivering health insights to users.
What's next for VocalGuard
Launch our hospital pilot program and broaden condition coverage (e.g., Parkinson’s, depression).
Seek FDA clearance to position VocalGuard as a medical diagnostic tool.
Integrate with telemedicine and wearable health platforms for broader reach.
Continue improving data diversity, model explainability, and user experience to maximize impact and trust.
Log in or sign up for Devpost to join the conversation.