Creating a mood-scanner, Possible? Options

[odyssey]

My foobar2000 just went from Hardcore "Scooter" to gentle "Enya"... And I thought "We live in 2008, it MUST be possible to analyze audio data for it's energy and generally adapt a simple category for it, something like "sad/romantic/happy/energy".

I know this has been discussed several times already, and I'm not looking for a bunch of posts about this being impossible, but positive thoughts about how this could be done! This will never replace the need for genres and maybe BPM etc., but it could be a great addition to existing meta-data.

In a seperate thread we can discuss the user-interface that can handle all this

[Zoom]

I tend to agree that it is a subjective thing when it comes to determining the mood of a song and a database might work for the more obvious examples of mood (i.e. slow songs, or really fast ones) it's the songs that lie in the middle that would be argued over which side they should fall on. Perhaps an ideal solution would be an online database where people submitted a songs tags with a mood tag as well. The majority tag would be the default mood for that song, but you could have the option of seeing what other people had tagged as well. This would allow people who couldn't decide for themselves to have a somewhat reliable alternative which would be easy enough to change if they decided to do so.

This is definitely a topic worthy of discussion in my opinion. As someone who has a growing music collection, but very little in the way of classification tags, I think varied ways of classification outside of genre and rating would be a very good thing. This to me seems like one of the next steps in libraries of digital music which continue to grow. It creates something of a problem when you have over 10,000 songs at your fingertips but can't decide what you want to listen to at the moment. This may not be a problem for some, but the fact that there are already mood tags and mood indicators in the wild should say that it is a growing interest.

Done improperly I think this could do more harm than good. You get to a certain point, or at least I do, when you ask yourself; "How much data do I really need about my data?"

[SebastianG]

The MusicIP mixer is supposedly able to generate "good" playlists based on a small selection of songs that fits your mood. -- I never tried it though.

IIRC they use a big database that stores the "songs' genes" and the program requires some analyzing stage where the "genetic meta infos" are retrieved for all your songs (= scanning your collection's meta infos and requeting the "genes" from the server). Once you've done that you could at least theoretically create playlists with "similar" songs using "gene distance metrics".

The problem is that this service provider learns something about you because the server gets all the gimme-genes-for-tags requests (yay data mining!)

Cheers,
SG

This post has been edited by SebastianG: Feb 18 2008, 20:34