Distortion and AI

After all, SL is supposed to use AI functions to perform some actions. What is the development of AI for the distortion problems that appear on recordings due to overmodulation? Are there already programs that can analyze the audio material in such a way that they are able to eliminate the distortions afterwards?

You can already deal with some distortion. If a signal is relatively hard-clipped for example then you can have a software calculate what the peaks would have been without the clipping and just generate that as a new waveform. RX does this in the de-clip module.

Thanks, I understand the principle behind it, but it doesn’t convince me much. I probably expect too much. Are there other systems?

Distortion can happen in several places in the signal chain so the problems will appear differently. RX de-clip won’t work well with say an ‘overmodulated’ preamp for example, because it isn’t necessarily ‘pure clipping’ but rather added harmonics or other artifacts.

So, I guess my answer is that outside of RX de-clip I’m not aware of anything that undoes distortion.

I also think that distortion, like noise, is very difficult to tame just by principle. The human ear is able to hear and recognize individual voices or instruments out of a whole mix. Previous software only works with simple calculations (which are very fast, though), for example by following the edges of an oscillation computationally, and then calculating an intersection point above the clipping.
That’s why I imagine that an AI could definitely do more here. Just as SpectraLayers, for example, is capable of breaking down music into individual fragments, I could imagine that it would also be possible to get the “pure” sound without clipping by means of AI.

Got it.

Peter Jackson’s audio team’s propriety AI software was able to be trained to distinguish and separate out each individual Beatle’s voice - even from crappy stealth recordings by mics hidden in flower pots. And they did it using a ton of examples. I’d suppose you could feed an AI lots examples of each specific type of distortion both distorted and not. Eventually it would learn to identify different types of distortion & also what that sounds like before & after.

I hear you and am fascinated (where can I learn more? :blush:), but as someone who has listened to probably thousands of hours of Beatles voices, I would like to see how this Al Peter Jackson used performed on the “hard ones” (e.g., identifying/ separating voices in “I Don’t Want To Spoil The Party”, similarly for the lower harmony in the verse of “You Really Got A Hold on Me”, and of course who sang the "Ahh"s in the middle of “A Day In The Life” :smile:).

Another tough one - John’s melody and George’s harmony cross over each other multiple times in the verses of “Yes It Is”. I guess by exclusion one could say it must be George down there with John (since it’s clearly Paul up top), but figuring out who is singing each note is pretty challenging at times.

We’re learning more about Al’s tendency to confabulate when it is stumped …I wonder how PJ dealt with that, or if he was even aware of that back then?

Aside: Al … who is this guy in the news so much lately? Is it Al Bundy??

I don’t quite know whether to believe the story yet either. I would have to see that with my own ears and eyes. Are there even that many recordings of the Beatles singing individually to teach artificial intelligence? I once heard that many recordings were still made with overdubbing. When did the Beatles work with real multitrack recordings?
Only then was it possible to use the individual voices for training the
AI.

Train ticket prices, single Vs return - Singletrack Magazine

[sigh]

Well there are 2 different main flavors of what AI is used for. One is generative like ChatGPT where the AI generates content and this can result in it making stuff up. But SpectraLayers and Jackson’s team use the AI to learn how to distinguish one thing from another.(e.g. John from Paul) and for this kind of work there really isn’t any mechanism where it is even possible to make stuff up. Either the AI does the task correctly or not.

Oh yeah a ton of recordings. Jackson’s team had access to all the recordings that EMI has in their vault. So they could go in and find isolated vocal tracks on the master tapes. But they also trained it on speech as well as singing.

In one of the other AI threads here, someone posted this link a few months ago of Jackson’s team being interviewed about how they developed & used their tools plus a lot of other stuff about doing the project. It’s about an hour long but well worth it if you find this stuff interesting.

Thanks for sharing a link to that podcast. Just listened to it and found it super interesting.
And now I have a new podcast show to check out. Splendid!

AI is of course very interesting to everyone but I think very few people realize what it actually is doing to us as musicians/producers/composers and human beings?

We’re training AI to become better at everything what we as humans can do. We teach AI to sing like anyone we know. We teach AI to compose music for us.

And I’m sure right now people in the financial market are teaching AI to make better investment decisions?

Pretty soon (if not already!) AI will be better at everything we as humans can ever be?

So our existence as human beings eventually will become obsolete!

1 Like

My dishwasher is also better than me.
Not necessarily faster, but in the same time I can do other things. I guess it depends on what I see as the meaning of my life? With every major invention, work that had previously been done by people was simplified. Was that bad for people? Is that why humans died out?
Sure, in the first period after automatic looms were invented, quite a few people became poor and may have even starved to death. But today I am glad that I can buy decent fabrics without becoming poor myself because of the costs. In this respect, AI has nothing threatening for me.
There is another area that worries me more. That is the responsibility for damage caused by AI. If your AI car has to decide whether it should rather kill the old woman or the child in the baby carriage, if it finds no other way. My opinion is that you should not let an AI in a car on our roads. Not until the AI would be able to know everything that an adult human knows. Only when the AI itself can feel how it would be to be hit by a car. Only when it recognizes what it means for humans to be dead.
Since this will never happen in my opinion, I will be against the use of AI in situations where decisions are made about humans.
I see the matter quite differently with technical solutions, where AIs could definitely help us.

Yes, but this is of course a totally different matter. I would also welcome a robot to clean, vacuum my house and wash my car. This will give me more time for other things. But if these ‘other’ things are also done better by AI, what’s left for us to do? Think about it? If AI comes to a stage were it’s also going to be better at performing all our hobbies and passions, drawing, painting. Composing music, arranging, mixing and producing music. What’s actually going to left for us to do?

I’m not saying that we need to worry today or maybe even the next few years? But at the rate AI is currently developing IMO it could seriously endanger the meaning of our existence in the very near future.

They are already testing AI cars as you know. Who’s going to decide when it’s save? The government? The developers? The big investors? They don’t care about how many humans die? They just need a successful test that will document that the casualties will not be > than with conventional traffic. That will be the starting point for implementation.

How can you be sure about any of this? Government agencies all over the world are already using AI based algorithms in their systems to track down ‘potential fraudulent behavior’ and as a result target certain neighborhoods. So even if your completely honest the government assumes your guilty or at least a high risk just because you live in that neighborhood. So you’re now branded for life just because your name is on a list that was generated by an AI algorithm! It seems to be very easy to get on that list without you actually doing anything wrong and not even knowing about it? Until you need a loan, a mortgage or even a phone subscription. And then you discover it’s almost impossible to clear your name and reverse the damage it can do for the course of your entire life. It has happened to many people already and it most likely will happen again!

Yes, I agree that AI can help, assist and enrich our lives in multiple ways. But the dangers of it working against us are IMO much greater!