benchmarks/rt60_regression/README.md

RT60 Regression Baseline

A fast baseline predicting RT60-like targets from RIR signals using lightweight features + RandomForest.

Usage

pip install soundfile numpy pandas scikit-learn
python benchmarks/rt60_regression/train_rt60.py             # auto target
python benchmarks/rt60_regression/train_rt60.py --target rt60

Default target order:
rt60, drr_db, c50_db, c80_db, band_rt60s.125, 250, 500, 1000, 2000, 4000

If you have no valid split, the script will carve 10% of train as an in-memory validation set.

Tips

• Ensure metadata/metadata.csv paths match your audio under data/ (or set RIRMEGA_DATA_DIR).
• To run on a tiny subset quickly, generate data-mini/ (see scripts/make_mini_subset.py) and copy this folder to a fresh repo/Space.