Subject: Re: [AUDITORY] Feedback on features for music similarity From: Paul Arzelier <paul.arzelier@xxxxxxxx> Date: Wed, 31 Jul 2024 20:19:33 +0200Hi! Thanks for your answer, and appreciate the kind words! You definitely=20 make some good points: 1. I did NOT think about windowing the songs in "meaningful" moments,=20 but the beginning and the end make a lot of sense since we wanna make it=20 smooth. I'll explore and see what that yield. 3. I'm also thinking of letting users choose themselves, and see what=20 they like more. 5. It's super cool, I didn't know they exposed that! I'll definitely use=20 it to check that it matches at least e.g. the key the song is in. @xxxxxxxx - do you have a hunch of what they might say? One problem is=20 that musicians are not really concerned with the harsh reality of=20 digital signal processing :D Best Regards, Paul Le 30/07/2024 =C3=A0 17:54, Hem, Charles a =C3=A9crit=C2=A0: > Hi Paul, > > Very cool project!=C2=A0 I'm a PhD student at Harvard studying cochlear= =20 > implants on the human perception side of things, so I'm no audio=20 > specialist myself.=C2=A0 Trying to study music perception for cochlear=20 > implanted recipients brings me back to featurization of music=20 > frequently, though, so I have a couple non-expert thoughts for bullets=20 > 1, 3, and 5: > > 1:=C2=A0 Maybe it's adding too much complexity, but I wonder if=20 > splitting=C2=A0your feature analysis into more windows could be helpful= if=20 > you're most concerned about seamless transitions between songs.=C2=A0 F= or=20 > example, "start of song" (first 20 seconds, or something near that),=20 > "whole song", and "end of song" (last 20 seconds) could be used to=20 > give some extra weight to making sure that the "end of song" from the=20 > first song matches the "start of song" of the following song. Mainly=20 > for songs that are highly variable or have a drastically different intr= o. > > 3:=C2=A0 I haven't read the chroma paper you cited, but gut reaction fe= els=20 > like these features shouldn't outweigh tempo and loudness so much.=C2=A0= I=20 > think your four class division could be a good idea. > > 5:=C2=A0 Spotify song features are pretty easy to get using this:=20 > https://developer.spotify.com/documentation/web-api/reference/get-audio= -features.=20 > The features are much more abstract than your features, but in my=20 > experience, Spotify's smart playlists do a pretty good job.=C2=A0 I'm=20 > guessing these features are partially used in that, so could be a nice=20 > reference to compare your results to. > > Most of this probably isn't super helpful, but I'll keep an eye out=20 > for any updates about the project!=C2=A0 It could be a cool research to= ol=20 > to quantify people's music preferences one day, which would be an=20 > interesting use for me. > > Best, > Charles Hem