When is AI going to regenerate the lost data on recordings from cd or other digital source?

[antcollinet]

Can you reveal the maths that took you to your conclusion?

I did.

 

[pablolie]

AI will not ever be able to recreate the original. It doesn't have the data if you don't. It may also have several copyright issues. There is no science about recreating original recordings unless you *have* the original recording - which would mean you don't need AI anyhow.

The faith in AI as a magic tool that knows everthing about anything baffles me.

Is this really something anyone would want? AI enhanced original recordings based on whoever the hell sets the "improvement" parameters?

 

[Robin L]

There are only 88 notes on a piano, it will not take very long for AI to reproduce any piece of music. There are only 8 notes in an Octave. How many permutations of music can you produce with 8 notes? AI will be able to create a large finite number of recordings with those 8 notes. It's Rocket Science or probably not.

But those 88 notes could wind up being different notes depending on how the piano s tuned. A=440 is not sacrosanct. A was 421.6 in Mozart's time,

You would be surprised. First, it's 12 notes (actually 11 without the full octave). If we only consider 7 levels of loudness (ppp,pp,p,mf,f,ff,fff) and 8 durations (1/16, 1/8, 3/16, 1/4, 3/8, 1/2, 3/4, 1) that is 616 variations of a single tone. Then you can have different ways of phrasing (legato, staccato, pedal ...) and then there is tempo.
So for starters there are much more than 1000^20 ( =10^60) possible variations for a very short pieces of 20 notes inside of a single octave and that is without taking into account that you can play two- or three- or four-part or accords (though most of these don't conform to the main stream and that is what AI will produce: 100 mio times the same thing).

And don't forget that there's all those microtonal adjustments or the potential of slides between notes, meaning the possibilities are as close to infinite as anything in the physical realm. Consider that a symphony orchestra can have 100 string players attempt to play the same note, but never being able to play that note exactly at the correct pitch. Of course, that's why the string section of a symphony orchestra has such a rich sound.

 

[Robin L]

AI will not ever be able to recreate the original. It doesn't have the data if you don't. It may also have several copyright issues. There is no science about recreating original recordings unless you *have* the original recording - which would mean you don't need AI anyhow.

The faith in AI as a magic tool that knows everthing about anything baffles me.

Is this really something anyone would want? AI enhanced original recordings based on whoever the hell sets the "improvement" parameters?

That has already happened. Giles Martin has overseen the remixing of a number of Beatles recordings utilizing machine learning. The 2023 remix of "I Am the Walrus" on the reissue of the 1967-1970 compilation effectively re-writes the ending. The stereo imaging is much improved, but there are things going on towards the end that didn't happen on the original issue, like lots of what seem to be different radio broadcasts, much higher in the mix than before. On the other hand, some of the early tracks have improved definition. The Special Edition of Revolver from 2022 used the Mal de-mixing software. "She Said She Said" sounds noticeably different. My 2009 remaster has everything slightly distorted at the peaks, the 2022 re-mix is way too clean. Considering John's fondness for nasty distortion, would he approve of the "improved" sound? Would he have approved of the re-composed ending to "I am the Walrus"?

 

[Robin L]

You are exaggerating. The notes are the important part and there are only 8 of those. I rarely hear an original song these days. I believe that is because we have reached a tipping point in the fact that any selection of the 8 notes has already been played in another song.AI Could squeeze a few more out but repetition would occur by the nature of the limited scale.

The more you know . . .

Dunning–Kruger effect - Wikipedia

en.wikipedia.org

 

[thegeton]

I used to record and save vinyl in 24/96 at a point. Dirty Techno vinyl worn by DJ use, no less. My obsessed arse even did manual declicking using headphones and looking at the waveform, then repairing the clicks by redrawing the waveform by hand. Then normalise to full volume. Took about an hour for one track. Autodeclickers are flawed machine algorithms for amateurs!

That, is f**king hard core my friend...

 

[kemmler3D]

Some sort of AI could synthesize a sound field that resembles the one in the room where something was originally recorded, sure, maybe.

What are you going to play it back on? There's no playback system that can transpose a sound field from one room into another room. Even a full Atmos setup can't do that, just to the extent that your room resembles the studio.

Either way, as has been pointed out many times already... it won't be regeneration of anything having to do with the original, just simulation, synthesis, or interpolation, depending on what you want to call it.

I see where the concept is going - I have said before that the "final form" of music playback tech will involve recreating the sound field in 3D, inside your room, but that may require a room covered in transducers (wall, floor, and ceiling) that surpass anything we have today, as well as recording tech that I'm not sure exists yet either.

If we ever get playback systems that are capable of this, I imagine AI synthesis will be used as a type of "upscaling" of old recordings for those systems, but as far as I know we're pretty far off from having that kind of tech. We're only now seeing ANC systems that work in free air, and even then only in very limited spaces using a lot of specialized equipment.

The kind of thing envisioned in this thread would require very effective ANC (like Dirac ART but in 3D and full-range) and an array of transducers capable of producing an arbitrary sound field inside a specified volume.

Anyway, despite some of the off-base assertions made by OP in this thread, I don't think it's a dumb idea at the base of it, but there does seem to be some misconceptions weighing down the speculation here...

 

[pablolie]

That has already happened. Giles Martin has overseen the remixing of a number of Beatles recordings utilizing machine learning. The 2023 remix of "I Am the Walrus" on the reissue of the 1967-1970 compilation effectively re-writes the ending. The stereo imaging is much improved, but there are things going on towards the end that didn't happen on the original issue, like lots of what seem to be different radio broadcasts, much higher in the mix than before. On the other hand, some of the early tracks have improved definition. The Special Edition of Revolver from 2022 used the Mal de-mixing software. "She Said She Said" sounds noticeably different. My 2009 remaster has everything slightly distorted at the peaks, the 2022 re-mix is way too clean. Considering John's fondness for nasty distortion, would he approve of the "improved" sound? Would he have approved of the re-composed ending to "I am the Walrus"?

I don't overy care about any version of being a walrus. What you're saying seems to make my point. It's like photoshopping the umpteenth Madonna pic. Who cares about degrees of fakeness? You really ready to declare them "better"? They have a history of never being better, all those "remasters".

 

[olieb]

And don't forget that there's all those microtonal adjustments or the potential of slides between notes, meaning the possibilities are as close to infinite as anything in the physical realm. Consider that a symphony orchestra can have 100 string players attempt to play the same note, but never being able to play that note exactly at the correct pitch. Of course, that's why the string section of a symphony orchestra has such a rich sound.

Absolutely. I was actually more trying to give a rough and conservative lower boundary number. Leaving out the twefth note, the "no note" /rest.
The given number is closer to the number of possible piano scores than to the number of possible realisations with microdynamical and temporal variations, tunings, instruments, acoustics, ...... And other instruments have even more options to modify the sound. Let alone the option to have more than one performer and so on.
That would be a so much bigger number that it is becoming meaningless.
Even so, with only a few bars (20 notes) inside a single octave, the number of possibilities is about a billion times greater than the number of atoms in the earth!

 

[kemmler3D]

There are only 88 notes on a piano, it will not take very long for AI to reproduce any piece of music. There are only 8 notes in an Octave. How many permutations of music can you produce with 8 notes? AI will be able to create a large finite number of recordings with those 8 notes. It's Rocket Science or probably not.

If you limit it to just 12 notes and limit the length of the composition to just 10 notes, about 62 billion songs can be written.

(12 ^ 10)

If you allow rhythm as a differentiator between songs - meaning not just 10 notes with the same length and timing, it goes into quadrillions very quickly, quintillions if you allow chords.

Now consider all of those songs are just compositions, and could be played on any number of instruments any number of ways.

Even if there are only 1 quadrillion songs that can possibly be created (a catastrophically low estimate), if they are on average 4 minutes long, that totals up to 7.6 billion years, so even if you could generate all the music that could possibly exist, you couldn't listen to even a tiny fraction of it in your lifetime - if the whole human race did nothing but listen to these super simple songs for a year, no sleep, no eating, we could just barely get through them all.

Maybe put down the bong and pick up a calculator?

 

[Robin L]

Maybe put down the bong and pick up a calculator?

Stop harshing my buzz, dude!

 

[Titurel]

But those 88 notes could wind up being different notes depending on how the piano s tuned. A=440 is not sacrosanct. A was 421.6 in Mozart's time,

And don't forget that there's all those microtonal adjustments or the potential of slides between notes, meaning the possibilities are as close to infinite as anything in the physical realm. Consider that a symphony orchestra can have 100 string players attempt to play the same note, but never being able to play that note exactly at the correct pitch. Of course, that's why the string section of a symphony orchestra has such a rich sound.

Exactly! There is a naive assumption by non-musicians (and some, naive musicians) that we are confined to the tempered scale or at least constrained by it. Jazz is only one of the obvious manifestations of the inadequacy of that view. What are called 'blue notes' are simply pitches that don't exist in Western music's tempered scales; notes that live in the cracks between adjacent piano keys.

On top of that is the assumption that string based ensemble (orchestras, string quartets, etc.) typically play tempered scales, even when playing music composed "at the piano." They don't. This is also true for concert bands and any other ensembles not tied to tempered instruments.

Further, ignoring the significance of time and dynamic (2 near infinite variables) is likewise naive. Then reinforcement or negation of structural accents (hemiola, polyrhythms, etc), timbre, etc.... ad finitum. Then there is location - dueling timpani per Nielsen - antiphonal music per Gabrielli, et al.

And so on and on and... (Sorry. Back a ton of years ago, I nearly took a minor in music comp.)

 

[Ropeburn]

Maybe put down the bong and pick up a calculator?

Wtf you mean, bro?

Extra dirt purposefully not cleaned up for your viewing pleasure. To make a point.

(Hypex 252MP amplifier to the top right lol)

 

[Robin L]

Exactly! There is a naive assumption by non-musicians (and some, naive musicians) that we are confined to the tempered scale or at least constrained by it. Jazz is only one of the obvious manifestations of the inadequacy of that view. What are called 'blue notes' are simply pitches that don't exist in Western music's tempered scales; notes that live in the cracks between adjacent piano keys.

On top of that is the assumption that string based ensemble (orchestras, string quartets, etc.) typically play tempered scales, even when playing music composed "at the piano." They don't. This is also true for concert bands and any other ensembles not tied to tempered instruments.

Further, ignoring the significance of time and dynamic (2 near infinite variables) is likewise naive. Then reinforcement or negation of structural accents (hemiola, polyrhythms, etc), timbre, etc.... ad finitum. Then there is location - dueling timpani per Nielsen - antiphonal music per Gabrielli, et al.

And so on and on and... (Sorry. Back a ton of years ago, I nearly took a minor in music comp.)

I worked for eight happy years at the Musical Offering in Berkeley, California, now defunct, alas—a combination cafe and CD shop, focusing on "Early Music" meaning (mostly) Bach and pre-Bach. We also sold other classical titles. The proprietor, Joseph Spencer, was an internationally renowned Early Music expert, broadcasting a weekly program on public radio, presenting concerts and festivals and having his own CD label, mostly of harpsichord. Most of what he told me about different tuning systems of the Baroque and earlier times went over my head, but there were different tuning systems for keyboards depending on what keys one intends to use in a composition. This resulted in a lot of different tuning systems at the time. The Well Tempered Clavier isn't about equal temperament, it's about finding the right temperament.

 

[Robin L]

Wtf you mean, bro?

View attachment 514890

Extra dirt purposefully not cleaned up for your viewing pleasure. To make a point.

(Hypex 252MP amplifier to the top right lol)

You really need to clean that, bro.

 

[Chrispy]

It's not needed nor possible.

 

[Ropeburn]

You really need to clean that, bro.

Do you have any scientific reason to backup that request? No? Thought so. The dirt stays.

Instead, I'll revert to an earlier post in this thread. 88 notes by yourself.

What if I told you, 72 notes are enough? Infact that was an old challenge on https://battleofthebits.com/ - my avatar on there is this

...and I made this.

72notesong.mp3 | Powered by Box

app.box.com

A rich synthesizer patch. Played by no more than 72 notes of musical information. Challenge breeds creativity.

 

[Zgrado1970]

Do you have any scientific reason to backup that request? No? Thought so. The dirt stays.

Instead, I'll revert to an earlier post in this thread. 88 notes by yourself.

What if I told you, 72 notes are enough? Infact that was an old challenge on https://battleofthebits.com/ - my avatar on there is this View attachment 514893

...and I made this.

72notesong.mp3 | Powered by Box

app.box.com

A rich synthesizer patch. Played by no more than 72 notes of musical information. Challenge breeds creativity.

But doesn’t it taste like an old fish tank? Frequent cleaning allows you to taste the terps, man.

 

[stalepie2]

If it were true that only 8 notes mattered we wouldn't see an instrument with more than 8 notes.

But even if I were to change my calculation to only include the 12 notes in a single octave, the permutations for a song only drop to 3x10^1080.

You really need to get a grasp of how permutations can mathematically grow rapidly out of control when you are accounting for complexity before continuing with this - frankly ridiculous - thought experiment.

This is moving the goalposts, but the discussion is only meaningful if the music is enjoyable to human ears, something people want to listen to. I think that would greatly reduce the number of permutations and it would be found there would be little left for AI to produce that hadn't already been recorded before by people.

 

[Ropeburn]

This is moving the goalposts, but the discussion is only meaningful if the music is enjoyable to human ears, something people want to listen to. I think that would greatly reduce the number of permutations and it would be found there would be little left for AI to produce that hadn't already been recorded before by people.

Yes, I agree. Let's move the goalposts towards something no human would ever want to listen to. And thus, we will...