Feature importance for speech emotion recognition
-
FI_demo.ipynb
A demo containing the steps for finding feature importance od opensmile egemaps using the CREMA-D dataset. It includes loading and resampling audio files, extracting metadata, extracting egemaps features, speaker normalisation, label separation and calculating feature importance. Feature Importance is calculated using:
- Permutation Importance (model agnostic)
- Standard methods
coef_
for logistic regressioncoef_
for support vector machinefeature_importances_
for random forest
-
FI_all_emotions.ipynb
A demo to use the pipeline function for determining the feature importances for all emotions of the CREMA-D dataset (Angry, Happy, Fear, Sad and Disgust).