seifukaleab/ethiopian-scales

Machine Learning framework for EMIR audio classification using MFCC, Chroma, Spectral Features, SVM, Random Forest, XGBoost, and SHAP-based explainability. ethiopian-scales/ ├── README.md ├── requirements.txt ├── .gitignore │ ├── notebooks/ ← the 5 phase notebooks (run, with outputs) │ ├── 01_preprocessing.ipynb │ ├── 02_feature_extraction.ipynb │ ├── 03_clustering.ipynb │ ├── 04_classification.ipynb │ └── 05_explainability.ipynb │ ├── src/ ← the .py scripts, grouped by phase │ ├── preprocessing.py │ ├── loader.py │ ├── inspect_data.py │ ├── features_logmel.py │ ├── features_mfcc.py │ ├── features_chroma.py │ ├── chroma_highlight.py │ ├── build_features.py │ ├── cluster_kmeans.py │ ├── cluster_chroma_only.py │ ├── visualize_pca.py │ ├── classify_svm.py │ ├── classify_rf.py │ ├── confusion_matrices.py │ ├── explain_feature_importance.py │ ├── explain_shap.py │ └── explain_lime.py │ ├── figures/ ← generated plots (committed, for the report) │ ├── features/ ← X.npy, y.npy, features.csv (see note) │ ├── Kaleabe Seifu APR Project Report.pdf ← the written deliverables │ │ │ └── data/ ← NOT committed (see .gitignore) └── raw/ ├── tizita/ └── bati/