Upload a CSV to retrain the nativity classifier. This will download
audio, extract embeddings, and train from scratch.
Required CSV columns: dp_id, audio_url,
nativity_status, language
Embeddings are stored on the server only and are not exposed in the
public repository. Training typically takes 15-45 minutes depending on
dataset size.
⚠ On Hugging Face Spaces, storage is ephemeral. Trained models will be lost
when the Space restarts. Use the Download Model button after training
to save your model locally, then re-upload the files to the repository.