AI-powered presetmanager?

[Brecht Guillemyn]

Is there an AI presetmanager that lets you search sounds with natural language?

I’ve been wondering if something like this already exists.

Right now, managing presets is honestly painful. My library is full of things like “guitar 1” --> “guitar 100”, and finding the right sound means endlessly clicking through them, listening and deciding if they sound the way I want them to
What I’d love is a preset manager that uses AI (LLMs or similar) to let me search using natural language. For example, I could type:

“Give me a warm sounding guitar that feels like watching a sunset in the countryside”

…and it would return the 10 closest matching presets from my local library.

To be clear, I’m not looking for AI to generate sounds. I just want something that can intelligently browse (and maybe tag?) my existing presets based on their sonic characteristics.
In the age of AI I was wondering if a tool like this exists? Or if there are other people with the same problem/wish as me? I have been browsing the web, but alas.

Would love to hear your thoughts or suggestions.

[Jonas78]

if with presets you mean .wav / .mp3, flac et cetera sound files, then look into Sononym. If you mean vsti presets, I have no idea?

[Zeisner]

I just want something that can intelligently browse (and maybe tag?) my existing presets based on their sonic characteristics.

How about, uh... yourself?

[Resonant- Serpent]

I just want something that can intelligently browse (and maybe tag?) my existing presets based on their sonic characteristics.

How about, uh... yourself?

Yep. Scene tourists looking to grift and take advantage should always be shot down.

[Zeisner]

I mean, look at all the generative AI stuff. Underpaid incompetent **** tag training data pretty much randomly. No wonder why generative AI sucks. You need precision. The only way to get it is to do everything yourself, just like scientists. Carefully selected data tagged carefully, fed to a carefully tuned/programmed model. You don't need machine learning though, pretty much any simple database program already does the job. Like Excel or Calc. Add some filters and you're good. Not only will it be more precise but faster as well because you don't need to run some neuronal network.

[Brecht Guillemyn]

I just want something that can intelligently browse (and maybe tag?) my existing presets based on their sonic characteristics.

How about, uh... yourself?

Yep. Scene tourists looking to grift and take advantage should always be shot down.

This feels like a sensitive topic, even if it's not generative AI. I just started out using DAWs, VSTs, ... the other day and was curious for ways to streamline my workflow.

Hypothetically, would you also be against AI tagging if there was a guarantee the presets were tagged perfectly?

[Trancit]

...
Hypothetically, would you also be against AI tagging if there was a guarantee the presets were tagged perfectly?

This is what I think many people do not recognize any longer nowadays...

The only one making really sure that the presets are perfectly tagged in the sense that it is perfect for you...is you...

No other person...no wonky machine code...just you!

So if you like it or not... I fear you have to make your hands dirty and get the job done even if this might be a kind of offense for people in our new shiny "AI Age"...

[Zeisner]

Hypothetically, would you also be against AI tagging if there was a guarantee the presets were tagged perfectly?

Machine learning can never reach an accuracy of 100 % by design. You only get an approximation because there's no boolean math/logic involved. The only way to get 100 % accuracy is using standard statistic methods and doing everything yourself, using your categories based on your perception.

It's also much faster using a spreadsheet with category filters because you don't have to type in words anymore, you get to the desired results with just a few clicks instead of prompting for at least half an hour because your model is going crazy again (and still not finding that preset you're actually looking for).

[BertKoor]

The 'fact' that such a thing exists is based on an assumption: AI can categorise large collections of data.
While that might be true in some cases, it's false in general.

Some problems & wild assumptions here:
* The machine can load the preset and play with it for a while
* The machine has large quantities of other already categorised things to compare it with
* Somebody was willing to invest time & money to build it.

[aMUSEd]

The 'fact' that such a thing exists is based on an assumption: AI can categorise large collections of data.
While that might be true in some cases, it's false in general.

Some problems & wild assumptions here:
* The machine can load the preset and play with it for a while
* The machine has large quantities of other already categorised things to compare it with
* Somebody was willing to invest time & money to build it.

*The machine would know what is meant by '...that feels like watching a sunset in the countryside' (which is a subjective state known only to the author of that statement)

[ksandvik]

Ableton has a kind of half-way system of finding similar sounds, for example samples for kicks. However, I rather rely on happenstance finding interesting new sounds so I seldom use that feature. Same with a similar AI tool describing sounds.

[Zeisner]

(which is a subjective state known only to the author of that statement)

...which is the biggest problem with LLMs. This acts as additional noise layer that reduces accuracy by a large amount.