Is it possible to programmatically find out what sounds the microphone recorded?

Question

Sound processing

Good afternoon! There is a video surveillance system. If you add a directional microphone to it, then you can determine what kind of sound it was by recording the sound with any program? (for example, a car drove up, a dog barked, a drone flew over or a plane)

Answer 1

Elementary - take a neural network, train on a prepared sample, test, use.

Answer 2

As a baseline, I would try this solution:
- record the sound as a bunch of short chunks (there, 0.5 sec. for example)
- build spectrograms on these chunks (there will be a matrix of the form N * k, where k is the number of frequencies, N is the length of the recording)
- average spectrograms along the "record length" axis (we get vectors of size k) - we train a
classifier on these chunks and the labels assigned to them (perhaps - k-nearest neighbors? In any case, the problem in this form seems similar to the one being solved by him)
samples) perverted over more complex solutions - for example, as in my comment to kisaa's answer - and would compare the quality of the classification (on cross-validation, of course :-) ).