AI plugin uses neural networks to generate infinite drum samples - feedback welcome

VST, AU, AAX, etc. plug-in Virtual Instruments discussion
Topic Starter
24 posts since 18 Jun, 2022

Post Wed Jun 22, 2022 1:02 pm

jules99 wrote: Wed Jun 22, 2022 12:51 pm
ai_music_guy wrote: Wed Jun 22, 2022 10:43 am
jules99 wrote: Wed Jun 22, 2022 4:37 am
- A live mode for the generative music folks. Generate to samples in the background and have them change after x number of bars in each selected cell.
- A live mode for the generative music folks. Generate to samples in the background and have them change after x number of bars in each selected cell. [already on the to do list! "humanize" or "jitter" or something]

Thanks for the feedback!
Just to clarify, because I think you might have misunderstood. A live mode for generative purposes would mean that you load a set of generated samples in the plug-in and start sequencing them Or have a sequencer trigger them. Meanwhile, the plug-in would, in the background generate new samples for each cell. And you could set the number of times a sample is played until a sample is replaced. Or the number of bars until all samples change. Or just some.

But I'm not 100 percent sure if this would work since being connected to a server is not always feasible for live situations.
Oh, I understand. We've had brainstorms around this kind of thing. I think with the right kind of caching it should be feasible. Thanks for the suggestion!

1282 posts since 3 Dec, 2002

Post Wed Jun 22, 2022 3:58 pm

Synapse worked well here, was able to get some nice sounds out of it. Couple of things I'd like to see..
pitch adjustment for generated sounds
ability to duplicate a cell (to create slight variations for hats or varied snare hits)

Other than that, I like it.

1040 posts since 23 May, 2016

Post Thu Jun 23, 2022 3:14 am

ai_music_guy wrote: Wed Jun 22, 2022 4:46 am
atukao wrote: Wed Jun 22, 2022 3:06 am Will this be subscription only?
Yes - since we have model serving costs, we have to use a subscription model once we launch. We're thinking through how a free version might work, perhaps by asking users to help tag sounds, but that's a least a few months out after release.

We've also looked into bundling the models with the plugin to remove the model serving costs, but it's not feasible in the short term bc it'd be a significant reworking of our tech.
Thanks for the info, then I'm out of testing unfortunately. I dislike always on cloud connectivity, I even more dislike subscription only models.
I think it's a bit no-go to be dependent on a cloud service to be part of a plugin. Yes we are always on, but no, my studio is not always on and even I live in Germany, there are far more outages than you might think. Also when being on the road, I do not always have a good connection. Additionally when I was dependent on a product with cloud copy protection it didn't work because of a server outage (it was in the middle of the night in the developers country, so no support). I do not want to make myself dependent on more services of this kind.

User avatar
6007 posts since 9 Dec, 2008 from Berlin

Post Thu Jun 23, 2022 4:06 am

Nice concept, but I removed the plugin from my computer.
- I don't do subscription for plugins and the monthly prices look silly to me for what it does.
- I also do not want cloud-connectivity if I can avoid it.
Some of the generated sounds were nice enough, but I heard nothing I couldn't find similarly in my library and the variety in my test wasn't really that great either.
Probably an issue with same-same in - same-same out?
That seems to be the general issue with machine learning so far, you train it on stuff that already exists, so basically you get "interpolated" output, nothing really new.

But thanks for the opportunity to give it a try :-)


"Out beyond the ideas of wrongdoing and rightdoing, there is a field. I’ll meet you there." - Rumi
ScreenDream|Thomas Helzle 8) Twitter

User avatar
1844 posts since 17 Jun, 2005

Post Thu Jun 23, 2022 4:51 am

Ah this isn't a local solution at all and requires a subscription and cloud access at all times? The outcome (what this can do) is indeed great, but this aspect wasn't apparent in the initial post or the web site. Unfortunately, count me out as well - at least for the time being, as on the other hand, I see you are considering making this an actual local/offline tool as well.
Last edited by Guenon on Thu Jun 23, 2022 4:56 am, edited 1 time in total.

User avatar
116 posts since 23 Jan, 2022

Post Thu Jun 23, 2022 4:56 am


6715 posts since 28 Apr, 2004 from france

Post Thu Jun 23, 2022 7:27 am

Interesting instrument, but still a bit rough around the edges.

- I think that the generated sounds were usable, but I couldn't manage to make it create "unheard sounds" yet, there didn't had very surprising or unexpected results : I wish it could generate a wider range of sounds.
- Most of the generated sounds were roughly of the same duration : I tried a couple dozen times, but didn't succeed to create shorter or longer sounds. A regular ride cymbal sample is longer than the longest sample Synapse generated. I wish it could generate a wider range of durations.
- I wish we could specify "goals" or "directions" : "make the sound longer", "make the sound more bassy", "generate a rattly sound", "make just subtle changes"... Total random is ok, but sometimes you like some kind of directions, and it helps improving the workflow.
- I am surprised that there is no pitch parameter featured ?
- Is there a way to rename the pads?
- Can we save the samples/kits to our hard drive yet?
- Is there any multimode filter featured (LP/BP/HP/notch)?
- I wish we could load a couple custom samples (let's say 2 to 6) into a pad, and ask the instrument to morph them into something new.
- It seemed that most of the generated sounds were lacking some "punch" and sharpness ( (sharp transients, brightness), a bit like when you mess with spectral editing too much... Maybe it was only the first attempts I tried which generated such results, but as we don't have any tool to lead the generation/morphing process, I couldn't get any sounds which I would have loaded in my drumsampler of choice as the main kick or snare...
Last edited by sinkmusic on Thu Jun 23, 2022 7:41 am, edited 1 time in total.

2687 posts since 1 Sep, 2016

Post Thu Jun 23, 2022 7:35 am

Subscription only, oof, count me out.

User avatar
74 posts since 3 Jan, 2021

Post Fri Jun 24, 2022 12:23 am

Oh, I also expected this to run the model locally. Needing an always-on "phone home" feature is definitely inconvenient and unwanted on my systems, sorry.

Not only does it raise certain privacy and security issues, it also means you can't use it on slower/minimal connections. Or as said earlier for live performance (say you want to generate sounds on the fly).

For an audio product this is really not a very good systems architecture.

User avatar
39 posts since 13 Apr, 2009 from Belgium

Post Fri Jun 24, 2022 6:22 am

The idea is good, the pricing is completely unrealistic.
Compare what you get with this subscription to what you get with other subscriptions and it is game over.

Return to “Instruments”