Scan your voice. Screen your health.

Detect early signs of diabetes, Parkinson's, depression, and heart conditions from a 20-second voice recording. Powered by peer-reviewed research. Runs entirely on your device.

Nothing leaves your device. Zero data sent to servers.

89% T2D detection accuracy
607 patients in training data
14 voice biomarkers analyzed
Published in Mayo Clinic Proceedings & PLOS Digital Health

Select Biological Sex

The voice biomarker models are sex-specific — diabetes manifests differently in male vs. female voices (pitch changes in women, intensity changes in men).

Based on Kaufman et al., 2023 — the models use different acoustic features per biological sex because T2D affects male and female voices through different mechanisms.

RECORDING 1 / 2
10

Read the text below at a natural pace

Pitch -- Hz
Intensity -- dB
HNR -- dB

First Recording Complete

For better accuracy, we'll record one more sample.
Read the same passage again at your natural pace.

Starting in 3...
0%

Analyzing Voice Biomarkers

Extracting pitch contour...
Pitch (F0)
Jitter
Shimmer
HNR
Intensity
MFCC
Spectral

Voice Biomarker Report

Metabolic Health

Type 2 Diabetes voice biomarkers

Based on: Jaycee Kaufman et al., "Acoustic Analysis and Prediction of Type 2 Diabetes Mellitus," Mayo Clinic Proceedings: Digital Health, 2023. Read Paper

Neurological Health

Parkinson's & neuromotor biomarkers

Based on: Research linking vocal fold perturbation measures to early-stage Parkinson's disease detection with up to 95% accuracy. Learn More

Mental Health

Depression & affect biomarkers

Based on: Kintsugi Voice research showing 20 seconds of speech can detect depression/anxiety biomarkers. Learn More

Cardiovascular Health

Heart & respiratory biomarkers

Based on: Studies linking voice features (pitch, breathiness) to fluid retention and cardiovascular conditions.

Important Disclaimer

This is a research demonstration, not a medical diagnostic tool. Results are based on published peer-reviewed research but are not intended to diagnose, treat, or prevent any disease. Consult a healthcare professional for medical concerns.

About the Science

VoiceScan extracts 14 acoustic features from your voice recording using signal processing techniques from computational paralinguistics. These features have been validated in multiple peer-reviewed studies:

  • Klick Labs / Mayo Clinic (2023): 267 participants, 89% accuracy for T2D detection using pitch, jitter, shimmer, and HNR features. Paper
  • Colive Voice / Luxembourg (2024): 607 participants, AUC 0.75 males / 0.71 females for T2D detection. Paper | Open Source Code
  • Google HeAR (2024): ViT-L model trained on 313M audio clips for health acoustic tasks. Paper
  • Kintsugi Voice: 20-second speech samples detect depression/anxiety, submitted for FDA De Novo clearance. Website

All processing happens entirely in your browser using the Web Audio API. No audio data is transmitted to any server.