Voice Stress Analyzer

Inspiration

My times public speaking and noticing how often I stutter and generally become nervous in that setting.

What it does

It essentially takes a voice clip and tracks subtle signs of vocal stress and fatigue, such as a high pitched voice.

How we built it

Built with Python and FastAPI. Core pitch/jitter analysis uses NumPy and SciPy for autocorrelation-based signal processing, librosa for audio loading, and the Anthropic API for generating plain-language summaries.

Challenges we ran into

I think the big challenge is coming up with a metric to identify what vocal stress and fatigue is and how to measure it

Accomplishments that we're proud of

I'm proud that I caught a subtle bug that could've quietly broken my results: pitch estimates kept spiking to values at the edge of my search range — artifacts from silence, not real vocal data. Rather than trust the first working output, I validated it against a spectrogram, traced the problem, and built a filtering step to remove those artifacts before they skewed my jitter calculations.

I'm proud that when something looked "good enough," I pushed to verify it was actually correct.

What we learned

What it's like to be in a hackathon

What's next for Voice Stress Analyzer

I would like to expand this so that the feedback is more personalized (i.e. if a person stutters too much then the detector can output a message saying that they need to slow down).

Built With

• api
• fastapi
• python