_extract_spectral_features(audio) using audio analysis library. | Feature | Unit |
| :--- | :--- |
| mfcc | 13 coefficients (list) |
| spectral_centroid | Hz |
| spectral_bandwidth | Hz |
| spectral_rolloff | Hz |
| spectral_flatness | 0–1 (tonal ↔ noise-like) |
| spectral_confidence | 0–1 | MFCCs power downstream classifiers (emotion, depression risk) and are also passed into the neurobehavioral fusion as feature contributions.