Here are two (mildly cherry-picked) random outputs as of right now:
https://drive.google.com/drive/folders/1nnC1a3APtfCIhsJtd6Q3dKYCw40mGz4m?usp=sharing
I’m running overnight, will share all 10000 randomly generated files when its done.
Looking forward to more renoise madness! Let me know what you think!
(If you are a developer, please let me know and I’ll share my current working code if you have ideas to improve it.)
that’s an interesting idea
Yeah, I’m working on stuff like this too, but maybe more pop oriented. Making a modular arrangement system that will also have a global randomizer button 
But it’s more interesting to use to quickly trial your way to the groove you want.
Latest example, but it’s gonna get better once switches/omits and symmetry are properly implemented: https://www.dropbox.com/scl/fi/55hu5r7kziao4wa0yxorv/generative-counterpoints.mp3?dl=0&rlkey=l2ywq7ihve3jyky6an1bn2ctz
What I’m really lacking is an algorithm that can determine accents from non-accents in an anonymous note stream (I need to understand exactly which factors needs to be known). AFAIK noone has solved this yet, so it’s kind of a holy grail. Then things start to get really interesting by interlocking the rhythms.
Maybe using a neural network here would be the way to go (for a generator), but requires a lot of training and manual weighting, I guess.
Aha, my favorite topic.
I will probably join
Generative or random ?
What are the rules ?
For those that are making random generators, can you share how this works? Do you have to link up the loaded sounds in renoise to the elements the tool can work with. Set which sample is the kick, the snare, hat, bass sound et cetera? So you will have to do preparation work beforehand, for it to make sense, not some semi smart name fetching thing where you can feed the content of a kit and it’ll figure out what is what through the name?
Yeah, I agree a neural network would help - it would be really cool to implement something like Text2Music by google, but text to renoise song instead!
Right now my method is really primitive - select random instruments from a folder, randomize the effect chain parameters, randomize the notes with a tiny bit of logic. They all come out sounding like Ambient tracks, I’m working on more diversity.
I’ve also written a neural-network based model for “classicalish” music based only on predicting notes in 4-part writing (https://soundcloud.com/user-332259139/transforming-transformers-2?in=user-332259139/sets/easy-listening), but I’m not sure how to integrate it into a model of other kinds of music, or even how to use it in conjunction with non-note commands and data.
Would love to hear your thoughts!
My current stance on generative music in general is that you need to work with seeds (hardcoded). Unless going for the blackbox AI/ML approach. My rationale for this is the following (personal opinion, true or false).
The events of all rhythmic counterpoints fall into either of these categories:
You can argue that an even pulse is a subset of euclidean pulses. Or even that all categories are a kind of linguistic phrase (even pulses being reminiscent of a child’s “da-da-da-da”).
The first category is imo mainly responsible for inducing visceral/physical response, as well as rhythmic context. Phrases, however, create call and response and adds intricacy by acting as a counterpoint. I like to describe the phrased parts of rhythm as being hypnotic (triggering our language center) rather than visceral or trance inducing. It’s interesting to note that you can create “pseudo-phrases” with combinations of two pulses, but this is a detour and quite limited imo.
Anyway… key point being: linguistic patterns of accents is something that is culturally conditioned rather than mathematical. They are evolved by convention, and imo needs to be hardcoded (or generated via some seed system). Otherwise it’s easy to end up with either chaotic arrangements or just electronic/trancey ones - depending on the algos used.