I had a (probably crazy) idea for a project and I was wondering if you all think it would be in any way possible. I'm interested in analyzing sounds made by different types of animals (for example bird songs or the croaking sounds made by different species of frog) and looking for relationships between the sound and other factors like habitat or taxonomic classification. If I could obtain digital recordings, would it be possible to feed the data into a clustering algorithm in such a way that it could identify and compare important characteristics? My fear is that most algorithms would prioritize extraneous details related to the quality of the recording or the type of compression over potentially meaningful factors like pitch, timbre, and pattern.
A Google search uncovered a few possibly related articles:
This one proposes a k-medioids approach but seems to focus on computer engineered sound waves rather than real life recordings.
This one uses a hierarchical algorithm and has a lot of good discussion on data cleansing and extracting "Low-level descriptors" to use as potential model features, however the focus is on classifying music which is obviously a much richer feature space than what I'm considering.
I have very limited knowledge in this area and would have to do a lot more research on data prep, feature creation, model selection, and so on but just wanted to ask if this is completely insane before I go too far down this rabbit hole.