Local Voice Wellness Screen
Records or uploads a 30–45s reading, extracts on-device vocal features (MFCC + spectral/energy measures), stores history locally, and generates a “signals to investigate + holistic practices” report. This is not a medical diagnosis.
1) Recording protocol (standard script)
Tip: Say the script in a natural conversational tone. Avoid whispering and avoid pushing loudness. Try to record in a quiet room.
2) Capture audio
Or upload audio
Supported: common browser audio formats (wav/mp3/m4a/ogg). Results may vary by codec.
Optional ML classifier (local model)
This app includes a slot to run an on-device TensorFlow.js classifier if you later load a model (no server). TF.js supports building audio classifiers and transfer learning workflows in-browser. [page:2]
Expected: model.json + shard .bin files. This v1 uses MFCC-summary features as input.