Bootleg with AI...

[Ancar]

Artificial intelligence could be a tool with which to try to improve the sound performance of some VU bootlegs. Taking inspiration from what the Beatles did with Now and Then. Do you think it could be feasible?
The evening of December 12, 68 at the Boston Tea Party, Gymnasium 68 and the Valleydale Ballroom would be worth working on in my opinion.

[Thatch]

If you mean being able to split the audio into different tracks (one for the guitar, one for the drums, one for the vocals etc) and then EQing the individual tracks and remixing them together to improve the clarity, then this can be done on live recordings but with mixed results. It usually doesn't work very well.

I tried it with the Cleveland October 1968 show - it didn't do a great job of working out what is vocal, what is guitar, etc. It added artifacts to the vocal track which made parts of it unlistenable.

It's a lot more effective with recordings that are already reasonably clear. I think Valleydale 1966 would be a non-starter, but I think it might work well with Gymnasium 1967, especially as I feel the vocals are too high in the mix. I'll give it a go sometime and will report back...!

[Elvis Plebsley]

I've used an online AI vocal remover on a couple of things with varying results. For the rehearsal jam on Andy Warhol From Tapes, it successfully removed two of the voiceover sections, but there is artifacting on the third. I also used it on a section of the Gerard Malanga reel from Chelsea Girls. This does leave a lot of artifacting, like background chatter, which is something that might be preferable to hearing the full Maria Mencken. It does seem to boost the background noise here and there. I tried the same section on another online vocal remover, but it was even less successful.

[Mark]

It's worth bearing in mind that it's a field that's advancing all the time at the moment. Of the off-the-peg online splitters, lalal.ai have just released a new algorithm for removing vocals and drums which to my mind is way ahead of anything else I've seen. Then there are the desktop based applications like RipX which are much more flexible and capable of even more impressive things, but I get the impression you really need to know what you're doing.

And as Thatch says, the recording needs to be half decent to start with. I had a quick play with Lalal and the new Little Sister recording but couldn't do much with it. On the other hand, something like Gymnasium 67 would be a great candidate.

[I am the table!]

If AI could be used to create a vocal-less Live MCMXCIII, that would be a Very Great Thing.

[jonas]

I have also often thought about this, but I think it might be too niche of a concept... You'd need an AI that is specialized for tape restoration. It would have to be trained with tons and tons of tape recordings of varying quality. Or some sort of AI "equalizer" that can restore the low/high end of an audio signal...

[Thatch]

It'll happen eventually - but how much can you add to the original before it no longer becomes a factual record of the gig?

[camforrester]

Brilliant!

If AI could be used to create a vocal-less Live MCMXCIII, that would be a Very Great Thing.

[Ancar]

Brilliant!

If AI could be used to create a vocal-less Live MCMXCIII, that would be a Very Great Thing.

[Ancar]

It'll happen eventually - but how much can you add to the original before it no longer becomes a factual record of the gig?

Without exaggerating of course. Enhance some sounds, clean the background noise. In my opinion there are some bootlegs that can improve a lot without distorting the authenticity. The one in the Gymnasium, in my opinion, would be perfect. It would be enough to lower your voice slightly.