🧠 VocClassifier — Infant Vocalization Classifier

Upload or record an infant vocalization clip (≤10s).
The model converts it to a Mel-spectrogram and predicts one of five classes.

Supported classes

  • Non-canonical (One) – early, immature speech-like sounds (example: InfantX.wav)
  • Other (Two) – unclassified/background/adult sounds (example: Other.wav)
  • Canonical (Three) – mature babbling with clear CV patterns
  • Cry (Four) – distress/discomfort vocalization
  • Laugh (Five) – playful/positive expression
Examples

How to use

  1. Click Upload (or record) an infant vocalization .wav.
  2. Press Submit.
  3. See the predicted probabilities for all five classes.

Model: ConvNeXt (FastAI + TIMM) trained on Mel-spectrogram images.
Train/Fine-tune on your own data: https://github.com/arunps12/VisionInfantNet