I tried writing an program to do this ~10 years ago (I called it "Tweeter") using a system like Shazam. The problem is, Shazam can detect specific songs based o

I tried writing an program to do this ~10 years ago (I called it "Tweeter") using a system like Shazam. The problem is, Shazam can detect specific songs based on about a dozen different analysis points (found another use for poles and zeroes). But if the song is sped up or muffled, it differs too much from the source recording. Birdsongs are far more variable, even within the same species. Ironically, at the time I discovered Cornell was doing this same thing, but it looks like they finally got to a product by throwing an ML classifier at it. Very cool, ML is perfectly suited for this. The interesting thing I learned during my study was that there is an entire system of describing bird sounds with nonsense words ("skee-dlees chis chis chis") that goes back over a 100 years.