How much potential remains for developing new non-AI audio plugins, business-wise?

[soundmodel]

There are audio processing techniques that aren't possible without AI such as timbre transfer. So AI (or really machine learning) does allow for new ways to create sounds and make music. But in the end it's just another technique for taking a digital signal and doing something interesting with it, and it has its own downsides and limitations.

This reminded me that speaking in terms of statistics theory machine learning and other models are just different lenses about the same thing. And in some cases a simple linear model is far more useful than a complex CNN or something.

In this sense the thread's question is flawed, because AI is not a successor to non-AI. It's an alternative to non-AI. But in statistics theory a linear model can also be AI.

So then I would answer the thread's question:

As much as there are useful applications for such models.

[whyterabbyt]

In this sense the thread's question is flawed, because AI is not a successor to non-AI. It's an alternative to non-AI. But in statistics theory a linear model can also be AI.

So then I would answer the thread's question:

As much as there are useful applications for such models.

Im sure the OP will be glad you helped them out there.

[kerfuffle]

In this sense the thread's question is flawed, because AI is not a successor to non-AI. It's an alternative to non-AI. But in statistics theory a linear model can also be AI.

In its original definition, AI is more of a philosophical thing than a specific set of techniques: Can computers do things that we traditionally consider to require human intelligence?

[soundmodel]

In this sense the thread's question is flawed, because AI is not a successor to non-AI. It's an alternative to non-AI. But in statistics theory a linear model can also be AI.

In its original definition, AI is more of a philosophical thing than a specific set of techniques: Can computers do things that we traditionally consider to require human intelligence?

Well SmartEQ seemed useful. But it doesn't replace the human at all to me. It just makes it a bit easier to get a good tone, if one's not an expert user of EQs. But IMO the suggestions the AI does are not very good. This kind of plug-in is easily replaced by just using presets in normal EQs as starting points.

I sort of think the same applies to e.g. MIDI orchestrators. The AI does not do a good job, but you can make it run (or you can just slam presets on the timeline) and then adjust the results.

The same certainly applies to "tone learners". I think they are not approximating a good tone, but just what the meters show.

[Lind0n]

In this sense the thread's question is flawed, because AI is not a successor to non-AI. It's an alternative to non-AI. But in statistics theory a linear model can also be AI.

In its original definition, AI is more of a philosophical thing than a specific set of techniques: Can computers do things that we traditionally consider to require human intelligence?

Well SmartEQ seemed useful. But it doesn't replace the human at all to me. It just makes it a bit easier to get a good tone, if one's not an expert user of EQs. But IMO the suggestions the AI does are not very good. This kind of plug-in is easily replaced by just using presets in normal EQs as starting points.

I sort of think the same applies to e.g. MIDI orchestrators. The AI does not do a good job, but you can make it run (or you can just slam presets on the timeline) and then adjust the results.

The same certainly applies to "tone learners". I think they are not approximating a good tone, but just what the meters show.

This. I think. So AI (Neural Nets and Machine learning) are good(arguably) at specific niche processes - but they cant "listen" to the audio - they are just running the numbers, and whilst this works for some things (e.g. guitar amp modelling) they lack the nuanced response we come to expect from an expert human, either at the wheel of the mix or the compiler. Yes AI will get better, faster, smarter but we will need a better "whole of brain" model than we currently have or is currently envisioned - so AI as a generalist useful tool, and thus replacement for tools built by experts seems not "doable" using the current AI models...

[soundmodel]

In this sense the thread's question is flawed, because AI is not a successor to non-AI. It's an alternative to non-AI. But in statistics theory a linear model can also be AI.

In its original definition, AI is more of a philosophical thing than a specific set of techniques: Can computers do things that we traditionally consider to require human intelligence?

Well SmartEQ seemed useful. But it doesn't replace the human at all to me. It just makes it a bit easier to get a good tone, if one's not an expert user of EQs. But IMO the suggestions the AI does are not very good. This kind of plug-in is easily replaced by just using presets in normal EQs as starting points.

I sort of think the same applies to e.g. MIDI orchestrators. The AI does not do a good job, but you can make it run (or you can just slam presets on the timeline) and then adjust the results.

The same certainly applies to "tone learners". I think they are not approximating a good tone, but just what the meters show.

This. I think. So AI (Neural Nets and Machine learning) are good(arguably) at specific niche processes - but they cant "listen" to the audio - they are just running the numbers, and whilst this works for some things (e.g. guitar amp modelling) they lack the nuanced response we come to expect from an expert human, either at the wheel of the mix or the compiler. Yes AI will get better, faster, smarter but we will need a better "whole of brain" model than we currently have or is currently envisioned - so AI as a generalist useful tool, and thus replacement for tools built by experts seems not "doable" using the current AI models...

TBH it's also questionable whether having an AI helps an expert musician at all, because the AI just throws a guess. It could be very far or near.

Somewhere where AI has been successful is mastering, but that is a whole lot easier domain, because you only need to adjust the full stereo track, not individual tracks, whose relative importance or meaning is really hard for an AI. Mastering also has far less room for gross errors.

[Fender19]

The industry is so saturated, and AI is getting better so fast.

Both true, but...

If I try to get into the industry at this point, will my plugin ideas be obsolete by the time I get to finish developing them?

That depends on what your ideas are and how fast you can develop them. Do you have the next "Soothe" or "Autotune" or are you making another generic compressor clone?

Is it too late to start an audio plugins business in 2024?

No. There is ALWAYS room for some new, great thing. Again, it depends on the answer above.

If you are already feeling defeated before starting (which it sounds like you might be) then invest your time elsewhere. If you have a new, great plugin idea then go for it!

[BertKoor]

Screenshot_20240321_213144_Gallery.jpg

[downtempo]

AI is good at doing high level audio effects. For example, taking a voice recording changing the language. AI is a bit of a buzzword right now, and, at times, it's sensationalized by the press. Broadly speaking though, I think predictions about the future are usually wrong.

[BertKoor]

You know what the biggest problem with pushing all-things-AI is? Wrong direction.

I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes.

[Widowsky]

You know what the biggest problem with pushing all-things-AI is? Wrong direction.

I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes.

Ahaha, so true.

The other day, I was reading the comments under an article about a DAW claiming to integrate AI at all levels of the workflow. Half hoped the software would be the engineer for them, the other half hoped the software would be the musician for them.

If the first half meets the second, there is no longer need for AI. Or perhaps an AI that brings them together…

[jupiter8]

So far,to me,the Ai offerings has been pretty underwhelming. Broadly 2 categories, dull as hall hell and not for me though useful for some people or nothing new but cheaper and more convinient.

Category 1: AI generate a whole song. Awh come on, what would i want that for? Useful for writing music for commercials mabe that needs fast turnaround cheaply maybe?
Stem separation, same but not as bad. Could see some potential uses but not for me. I want to creat something new.

Category 2: really useful for cloning reverbs, compressors and guitar amps. Sure having a SLO100 on full tilt without disturbing the neighbours is useful but it's not new. Has been done before.

So what's new? I could see AI assisted mastering being useful. Synplant and that new drum morphing thingie is cool. Not sure how much AI is in that though if any. Most things i imagine being useful could be done with AI but also without it i believe.

What about throwing a bunch of samples at a sampler and it automatically does all the mapping for you? What about feature extraction and resynthesis? Say you load a bassdrum and extract the transient, root sine the distortion and break those into separate parts and synthesize the sine, pitch envelope and distortion. Then you can form the bassdrum into a new one, longer, shorter, combine with transient from another sample, pitch shift everything apart from the transient etc. etc.

Give it some piano samples, extract the formants and synthesize that. Synthesize new samples inbetween the ones you have.

Could this be done with AI? Probably but also without,i believe. I believe the biggest help will propbably come from AI assisted coding but i have my doubts about that too.

Be creative, solve problems.

[Zaphod (giancarlo)]

short answer: quite a lot

[Cyforce]

In one direction AI could be also very beneficial - performance! Especially if one day in the future, plugins could use processing power from graphic cards. Similar maybe like the upscaling and frame generation in video games on modern graphic cards with specific calculator units. In the past, there was always the only way: decide to generate the images/frames in resolution xxxx:xxx etc... basically HD or higher like 4k and so on. The jump from an HD resolution (1920x1080) to 4k is a massive expense on performance. Nowadays, with new features designed for graphic cards with ai-related calculation units, the images can stay at HD (remaining a decent performance expense) while in parallel the other calculation unit performance an upscaling of the image in nearly the same quality to 4K, as purely native 4K - while in parallel generating frames in between the 30-120 frames per seconds, to the core calculation units "only" needs to perform a HD image - with more details etc, and only 20-40 frames a second, and the result will be 4k output with 40-80 frames per second, so a significant boost in performance and scale.

Something similar... could maybe come into the audio area, with AI calculation units(e.g. tensor cores on Nvidia cards) on graphic or DSP cards, performing massive oversampling in the signal processing like 16x, while the CPU of your system still uses only the performance drain of 2x oversampling and so on... or in general, using AI to make codes more efficient, draining less resources and so on.

[Widowsky]

Cory Doctorow has a pretty interesting take on AI, in addition to what's been said here:

https://pluralistic.net/2023/12/19/bubblenomics/#pop