VoiceCheck AI

Inspiration

I’ve always been curious about how much our voice can tell about our health. During the pandemic, I came across research that showed cough sounds could help detect COVID. That made me wonder: if this works in research labs, why not try it on a phone? That’s how VoiceCheck AI started.

What it does

VoiceCheck AI is an iOS app that records a short cough, breath, or spoken phrase and analyzes it in seconds. It gives a simple risk level for respiratory issues and offers basic suggestions. I also added a small mental wellness feature: if you record how you feel, the app generates a short encouraging “future letter” to you.

How we built it

I built this app solo. The frontend is written in SwiftUI with AVFoundation for audio recording. I trained models on open respiratory datasets, then converted them into Core ML so everything runs locally on iPhone. The flow is kept simple: record → analyze → result.

Challenges we ran into

Building this alone meant I had to handle both coding and model training. Getting the model to work across noisy environments and different microphones was tough. Another challenge was making the app feel supportive without pretending to replace doctors.

Accomplishments that we're proud of

I managed to build a working iOS prototype in just a few days. The model runs locally with <1 second inference time, and the app provides clear, easy-to-understand feedback. Adding the “future letter” feature made the app feel more personal and human.

What we learned

I learned how much health information is hidden in voice and how powerful on-device AI can be. I also realized that simplicity and privacy-first design are key when building healthcare apps.

What's next for VoiceCheck AI

Next, I plan to collect more diverse real-world samples to improve accuracy, add multi-language support, and explore early screening for conditions like Parkinson’s disease. Integration with HealthKit and telehealth platforms is also on the roadmap.

Built With

• avfoundation
• core
• fastapi
• ml
• next.js
• pytorch
• swift
• swiftui