knagode/vocal-coach

Singing voice classifiers (WavLM embeddings + logistic regression)

One model per task (<task>.joblib). Each scores 1-second windows: frozen microsoft/wavlm-base-plus embeddings (mean over layers and time) → StandardScaler → LogisticRegression.

breathiness

Labels: breathy, clear. Trained on 8 clips (57 windows). Evaluation: leave-one-session-out

              precision    recall  f1-score   support

     breathy       1.00      0.75      0.86         4
       clear       0.80      1.00      0.89         4

    accuracy                           0.88         8
   macro avg       0.90      0.88      0.87         8
weighted avg       0.90      0.88      0.87         8