Descriptors... and textural taxonomies

Learning Resources
1

Thomas Grill has made some interesting research on the musical usage of descriptors. I’m starting to read through his papers, so here are the best of so far - I reckon @danieleghisi has some insights on this too…

https://grrrr.org/research/publications/paper-smc-2012/

The ICMC paper is good too, but I still have issues with the mapping at p.25 of the slides. Any comments welcome.

1 Like
2

I’ve not read the whole paper, but I remember when @a.harker showed some of this stuff during a CeReNeM talk a few years ago I found those graphics really confusing.

I don’t know if it’s a learning style thing, or dyslexic/color thing, but they are almost visually meaningless to me.

3

page 25 is a nice attempt to put some of the parameters into an essentially on-dimensional change of visual detail. Some are more intuitive than others : ordered-chaotic seems to me clearer than mapping brightness to high-low.

4

Out of the five ‘axises’ two are readable, and 2 make sense to me. All of them make sense in words, don’t read me wrong, I love them, especially if they are aurally significant (I need to hear them now) but the mapping:

  • high-low makes no sense (would take it for brightness)
  • ordered-chaotic makes sense but it could go further in messiness (with my music at least :wink:
  • tonal-noisy is cheesy as hell - we should have homogenous colour palettes vs rainbow (which also makes sense with the brightness with lightness since dark sound are muddled up so a dark rainbow would look perceptually more unified)
  • smooth coarse is almost ok - I want more circle on the left and more spikes on the right.
  • homo-heterogeneous is another contender for a colour palette variation

I really, really like the kind of data reduction he is attempting though, so let’s keep the discussion open and we might find something that kind of make sense!

Now, I really want to see them (and proximity) on a real corpus!

p

5

Even beyond that, and again, I’ve not read the paper so I don’t know if there’s a good reason for this, but of all the ways to visually communicate variance, subtle variations in a more-or-less grid of nearly identical shapes, is pretty poor.

Metaphor is probably unavoidable given what he is trying to do, but like @tremblap pointed out, some of the choices don’t make much sense (particularly with regards to color).

6

what is good for me in there is, once I’ve heard them, if the 5 dimensions are orthogonal and complementary and completely describe the spectral space in my head. If that’s the case (and I think that is what he pretends with the matrices from a very small group of specialist users), he did a fantastic job, and we can think of other metaphors for visualisation.

7

Hi everyone,

Can you drop me some links for your favourite resources (texts/papers) for audio descriptors/features. Perhaps it’s this link in the thread…

I’m working with a musicologist grad student (in ethnomusicology) who is looking at bias in music AI, in particular streaming selection etc.

So, anything that is broad/descriptive/less technical that you’ve found useful. Maybe @jacob.hart has some good suggestions :slight_smile:

Thanks!
L

8

I know @frederic.dufeu4 and Keitaro have to educate musicologists about that in their project, and I also know that @a.harker and @spluta were thinking about something more advanced… and I’m sure @groma and @weefuzzy will have their favourite resources too… so I’m eagerly looking here to see what will pop up and then we can try to put them in order of difficulty and completeness as a good resource.

I vaguely remember a website which was describing each descriptor but I cannot find it now… it was definitely not beginner stuff. Do you need to explain the whole idea of ‘machine listening’ in all its complexity? I find that this thing (https://medium.com/cochl/what-is-machine-listening-part-1-6fbdf2a3d892) is simple enough to explain how complicated that is for beginners, although the assumptions on music are a little simple but it is an entry text…

1 Like
9

Thanks for that! I’d actually be keen to see that website if you/someone can remember. Entry texts are good too. The student is an accomplished guitarist, oud player, teaching and working in Arabic, Latin American ensembles etc. and has already done some interesting work looking at Spotify recommendations of ‘world’ music and how it fails due to how musics are categorized. Super interesting work.

10

There was such a discussion at CTM festival last week around this person: https://www.youtube.com/watch?v=GzcWzblOiSs

a bit too note-based for me but still interesting

2 Likes
11

Seems like Eric Drott’s “Genre in the Age of Algorithms” would be a good resource.

3 Likes
12

Just got an update from the team who reminded me: https://musicinformationretrieval.com/ is a good intermediate resource I was thinking of.

3 Likes
13

thanks all! eric was also super helpful and sent something.

1 Like