Wave Alchemy - Triaz Native - No Kontakt -- Free for current owners!

VST, AU, AAX, CLAP, etc. Plugin Virtual Instruments Discussion
RELATED
PRODUCTS

Post

imrae wrote: Fri Feb 23, 2024 1:01 am
I'm a lot more interested in getting the directory-based thing working properly than replacing it with AI voodoo that makes mistakes. It is already easy to drag in samples from Sononym :-)

AI/Machine learning isn't necessarily "voodoo".

Granted, there's a lot of people using it as a marketing buzzword but it's an incredibly useful tech for these kinds of categorisation tasks.

For example, Sononym itself uses AI/ML for categorisation. As do Atlas and XO.

If you have tens or hundreds of thousands of samples then AI is a great solution.

Post

kraster wrote: Fri Feb 23, 2024 2:52 am
imrae wrote: Fri Feb 23, 2024 1:01 am
I'm a lot more interested in getting the directory-based thing working properly than replacing it with AI voodoo that makes mistakes. It is already easy to drag in samples from Sononym :-)

AI/Machine learning isn't necessarily "voodoo".

Granted, there's a lot of people using it as a marketing buzzword but it's an incredibly useful tech for these kinds of categorisation tasks.

For example, Sononym itself uses AI/ML for categorisation. As do Atlas and XO.

If you have tens or hundreds of thousands of samples then AI is a great solution.
kraster's absolutely right here - categorization of samples is a pretty straightforward ML application, and can be super helpful with huge data bases. As there are a lot of pre-tagged, large sample data bases existing, one does not even have to hire coders to do the annotation, because it has been done already by either the companies themselves or the users... What a trained model would do: it would categorize new samples based on the model, derived from the training data - so identify a kick drum, tom, hi hat, gong, whatever, and maybe even their tonality and atmosphere. These models can be really good, not much worse than humans in such a case (and more dependable/consistent), and would probably categorize better than a rough folder structure - as a sample could have a lot more descriptors/tags.

Also, one could use the extracted features to generate new samples, so this could be the basis of a new instrument with infinite samples. (And indeed, one may even infer more complex physical properties from instrument data, to come up with physical models of instruments - if there are enough data points for the model...). For musicians, AI opens many exciting options, and we will see a lot of useful applications in the future for sure. And if it reduces the tedious tagging of my ten trillion samples, I would be super happy! :clap:

Post

This isn't about how Triaz handles your ten trillion sample library, it's about how it handles small, named sample packs that are already well-organised.

AI is a good solution to "I am drowning in samples, find all my phattest snares that sound like this progressive house reference track". It's a poor solution for the "what kind of drum is 909/kick/lofi/03 lofi kick 3.wav and what tags might be useful" problem.

Post

imrae wrote: Fri Feb 23, 2024 8:19 am This isn't about how Triaz handles your ten trillion sample library, it's about how it handles small, named sample packs that are already well-organised.
Yup- and the 15 000 drum samples that come with Triaz not exactly small....
X32 Desk, i9 PC, S49MK2, Studio One, BWS, Live 12. PUSH 3 SA, Osmose, Summit, Pro 3, Prophet8, Syntakt, Digitone, Drumlogue, OP1-F, Eurorack, TD27 Drums, Nord Drum3P, Guitars, Basses, Amps and of course lots of pedals!

Post

Also… AI isn’t likely to detect all the tags they I already know… for instance, to tag the tape saturated version from the normal one. If a developer gives you that, would be nice to use it.

Input jsons for ‘directing’ the import (reducing names, extracting tags) seems like a good idea, like someone said it is kind of for advanced users, but if advanced users share them, even less technical users can benefit…

Post

chroma wrote: Fri Feb 23, 2024 8:29 am Also… AI isn’t likely to detect all the tags they I already know… for instance, to tag the tape saturated version from the normal one. If a developer gives you that, would be nice to use it.
Yes, that's absolutely possible, depending on the amount of the data used for the learning process, how well it differentiates according to the features of the sample, the quality of the annotation etc. Saturated samples do have recurring features that differentiate them from others, and therefore we as humans are able to detect the differences - and ML as well (with proper training). In my day job, I work with ML for other tasks, and you'd be surprised how well it works. For small developers, it's probably more a question of time, money and input data available, but this is not a problem "by principle".
imrae wrote: Fri Feb 23, 2024 8:19 am This isn't about how Triaz handles your ten trillion sample library, it's about how it handles small, named sample packs that are already well-organised.

AI is a good solution to "I am drowning in samples, find all my phattest snares that sound like this progressive house reference track". It's a poor solution for the "what kind of drum is 909/kick/lofi/03 lofi kick 3.wav and what tags might be useful" problem.
No, that is a misunderstanding. You can use a pre-trained model to categorize new stuff you throw at it. The training data for annotation has to be solid (that's what the developer has to take care of), but then the categorisation later on - based on that model - could be done on a single sample. So it's actually a great solution for the problem you describe. I guess that Wave Alchemy are aiming exactly for that - a model that will be able to auto-categorise new material based on their preceding training. Which would be fabulous! :tu: But hey - they could answer that with more authority than me. :hihi:

Post

It's a waste of CPU if you threw away metadata that is right there. We are a long way from AI identifying that tape saturation was used more reliably than the people who recorded it to tape.

Post

... it would, if there was reliable meta data for all your samples... which is not the case for most of the samples I collected over the last 20 years. :?:

Whatever, I give up to convince you that this makes a lot of sense - it seems nobody here at KvR has that use case (i.e. tons of un-tagged, un-categorized samples). Just wondering why people are using Atlas or XO then.

P.S.: Just a few years ago, people said that an AI won't be able to reliably categorize and analyze language, or produce texts itself. And now LLMs do this extremely well, and for some tasks, much better than most people out there.

Post

I wouldn't mind a compact view of the sample Browser which can show more entries at the same space. It is rather large

Compact means not less information but just a reduced font size and line height to see more entries.

Post

midi_transmission wrote: Fri Feb 23, 2024 10:10 am I wouldn't mind a compact view of the sample Browser which can show more entries at the same space. It is rather large

Compact means not less information but just a reduced font size and line height to see more entries.
Yep, great idea.

Post

Can anyone show me how do you symlink for Triaz?

I try,

mklink /J "E:\Drum Kits" "C:\Users\username\Documents\Triaz\User Samples"

but I get, "can't create a file that is already there" (or something similar, don't know the direct translation because it's in another language)

What I'm doing wrong?

Post

Fornicras wrote: Fri Feb 23, 2024 12:34 pm Can anyone show me how do you symlink for Triaz?

I try,

mklink /J "E:\Drum Kits" "C:\Users\username\Documents\Triaz\User Samples"

but I get, "can't create a file that is already there" (or something similar, don't know the direct translation because it's in another language)

What I'm doing wrong?
You need to delete the "C:\Users\username\Documents\Triaz\User Samples" folder first, because the symlink will re-create it as a symlink instead. But check first that the folder is empty ;)
Find my (music) related software projects here: github.com/Fannon

Post

Fornicras wrote: Fri Feb 23, 2024 12:34 pm Can anyone show me how do you symlink for Triaz?

I try,

mklink /J "E:\Drum Kits" "C:\Users\username\Documents\Triaz\User Samples"

but I get, "can't create a file that is already there" (or something similar, don't know the direct translation because it's in another language)

What I'm doing wrong?
Just going from memory of when I did it the other day...

Is the order not the wrong way round? Shouldn't it be C drive first, then E?
"Like toilet paper, we're on a roll."

Post

Yes, sorry for overlooking it. This is btw. how I symlink my favorites & presets to OneDrive (I do this with many plugins / programs). This is for PowerShell. But still, you need to first delete or move the file / folder on the left side:

Code: Select all

cmd /c mklink /j "%userprofile%\Documents\Triaz\Presets" "%userprofile%\OneDrive\Musik\Triaz\Presets"
cmd /c mklink "%userprofile%\Documents\Triaz\favorites.json" "%userprofile%\OneDrive\Musik\Triaz\favorites.json"
Find my (music) related software projects here: github.com/Fannon

Post

For some reason Triaz just doesnt start most of the time with Bitwig when I hit play. This is with Sync enabled. If I click in the Triaz channel area, it starts to play.

Totally odd, not sure what the deal is there.

If you sequence it with midi from the DAW how does that work? Do you just use an empty sequence basically?

Return to “Instruments”