This is a meta issue to track thoughts about the API of the feature extraction module and related issues.

Thoughts:

• make it functional to the extent possible -- avoid classes, standard scientific Python
• provide named feature sets; we can provide these through functions, e.g. vocalpy.features.sat.similarity_features or vocalpy.features.warbleR.spectro_analysis
• but at some point we're going to have re-duplication of features
• which is where it might make sense to use classes to represent named sets of features; we'd have a "functional" module that's the implementation of the features, then
• we might need to find a way to distinguish between frame-level features and segment-level features
  • SAT-style analyses basically extract each acoustic parameter per-frame
  • but I think warbleR takes a Raven selection table and then extracts scalar parameters per segment?
  • Or looking at it more, maybe warbleR is just taking whatever seewave gives them. More generally, I think it might be useful to have the ability to do both frame-level and segment-level feature extraction, with some way to specify how you do the reduction--I'm thinking of the pytorch reduction argument, but it could also be a callable so you say reduction=my_reduce_func and get e.g. the pitch at 20 ms or whatever you want
• nice to have: follow logic of librosa API: accept either audio (y, in our case Audio) or spectral representation (S, in our case a Spectrogram), and if audio then apply appropriate spectral transforms to get whatever representation we need to extract the features

Related issues, to fix for closing this one

• ENH: Add features from hybrid-vocal-classifier #89
• ENH: Add biosound spectral representation + acoustic features #27
• ENH: Add warbleR features and replicate warbleR.spectro_analysis function #30
• ENH: Add acoustic features from koe #93
• ENH: Implement FeatureExtractor #94