I am looking for methods that work in practice to determine a kind of acoustic similarity between different songs.
Most of the methods that I have seen so far (MFCC, etc.) seem to be aimed at finding only identical songs (for example, fingerprints, not music recommendations for music recognition). Although most recommendation systems seem to work with network data (listened songs) and tags.
Most Mpeg-7 audio descriptors also seem to be along this line. In addition, most of them are defined at the “extract this and that” level, but no one seems to actually use these functions and use them to calculate the similarity of the songs. But even an effective search for similar items ...
Tools like http://gjay.sourceforge.net/ and http://imms.luminal.org/ seem to use simple spectral analysis, file system locations, tags, and user input such as color and rating, manually assigned by the user, or frequency of listening to songs and skips.
So: what audio functions are calculated quickly enough for a common music collection and can be used to create interesting playlists and search for similar songs? Ideally, I’d like to feed an existing playlist and release a few songs that match this playlist.
So, I'm really interested in acoustic resemblance , not so much identification / fingerprinting. In fact, I just want to remove identical songs from the result, because I do not want them twice. And I'm also not looking for a request, singing. I don’t even have a microphone.
Oh, and I'm not looking for an online service . First of all, I don’t want to send all my data to Apple, etc. Secondly, I only want to receive recommendations from my own songs (now I don’t want to buy additional music until I have studied all my music. I even I didn’t turn all my CDs into mp3s ...), and secondly, my musical taste is not basic; I do not want the system to constantly advise Maria Carey.
Plus, of course, I am really interested in which methods work well and which do not ... Thanks for any recommendations from the relevant literature and methods.