Interesting discussion on corpus-based musicking

[tremblap]

Two people on this site have talked together for 90 minutes on corpus-based musicking, I think that should interest some people

@rodrigo.constanzo and @encanti it is fun to see your thoughts and tunes, congrats!

Ben Cantil - Particle Castle Bubble Party!

[MartinMartin]

Enlightening!

@rodrigo.constanzo @encanti
You talked about the art of creating corpuses. What would be an approach to dynamically include and exclude certain parts of a corpus?
With parts I mean not regions of the corpus (for that one must change the analyzed features of the target signal I suppose?)
more for example all the samples from a specific folder or time frame of the source signal.

Maybe this is not the right approach - but the goal would be to have some dynamic control over the selection of samples of an already constructed corpus… In a live performance for example.

[rodrigo.constanzo]

There’s obviously loads of ways you can go about it.

At a macro level, deciding what to include in a corpus at all, and how detailed/thorough it should be, will probably have the single biggest impact. For example, I’ve quite gotten into having relatively small corpora, so it ends up sounding kind of stuck and repeat-y in certain settings.

Beyond that, you can choose how you navigate the corpus. I really got used to using @a.harker’s entrymatcher, which allowed for very flexible querying of the corpus space, and you could chain loads of queries together (e.g. “the nearest match within only samples longer than 500ms, with a bright centroid, and no pitch confidence”). In FluCMa that kind of querying is not possible (in the same way, or at all in some cases), so I’m approaching things differently.

I know @tremblap likes pre-splitting the corpus up into different chunks and then querying within each of those sections. I think that’s quite handy, but not so fast for quick realtime use (i.e. ‘per query’ variations).

Beyond that you can get a bit more abstract with it. I’ve got a friend that likes building up a corpus by number of samples (e.g. samples 0-100 represent “section 1”, samples 101-200 represent “section 2” etc…) and then he will navigate through those sub selections of the larger corpus.

There’s also loads to be said about what descriptors you use too etc…, as that can have a gigantic impact on the matching you get as well.

[MartinMartin]

Thanks! I will explore these ideas…

Do you know of any other (than for example your sp.tools) shared patchers using the flucoma objects for c-cat synthesis?

Would be nice to learn about different approaches/concepts…

//or resources in general, videos etc.

[rodrigo.constanzo]

There’s some discussion (and patches from @tremblap) in this thread:

Another noob question on scaling non-overlaying corpora

Cool, that’s useful to see. A couple differences and things to note (for @agustinissidoro) with the approach in SP Tools (and C-C-Combine). In my approach I tend to load a single file (which is still possible in this example, but just replace-ing the buffer~ corpus instead of loading a folder) and then slice at a fixed interval, as opposed to onsets/slices. This is a large point of diversion as I end up with (usually (significantly)) more segments this way, as well as not necessarily having sl…

[tremblap]

Hey @MartinMartin

@MartinMartin

Do you know of any other shared patchers using the flucoma objects for c-cat synthesis?

I don’t know if you have read the various articles on the FluCoMa Learn website (https://learn.flucoma.org/) but many people have tried various ways indeed, and @jacob.hart article come with demo patches… I did some implementations in here, @jamesbradbury too… there are so many ways to match 2 bunch-of-sounds

There is also the thread about pedagogues here - in there you will find some ideas and some teaching material so you can find inspiration and start-up patches for audio mosaicking, in real-time and non-real-time… ideas of batch processing for instance.

I don’t know what you are looking for - the subject is vast! Let me know what you are trying to do, and what you have already done in any software so I can maybe a bit more specific.