A.I. to repair poor quality radio shows

I just thought about this yesterday, so it's a sure thing someone's already thought of it.  Is anyone using A.I. to clean up and repair poor quality radio shows?  

I think it would be ideal for shows that are beyond ordinary clean up methods.  If properly done, it could probably be used to fill in missing parts of shows (when a script is available, etc.)  It could probably even create an entire show that is missing if the script is available, but I'm not sure I'd personally think that's a good thing.  I'm a bit of a purist when it comes to OTR.

I particularly would like to see repairs done for recordings that are missing part of the audio spectrum (too much treble or too much bass) to the point it is difficult to understand, and repairs of shows where the original recording is too hot (overdriven).

I have seen some amazing cleanup of video that was done with Ai. Sure be nice to try out a few shows see how it goes. Only kicker is probably expensive to process those files unless you setup a local LLM on your custom gaming PC with a hot GPU in it. Not beyond the means at all.

That's an interesting possibility. Makes me wonder about things like sound effects and music, when it comes to creating episodes from scripts. But as for filtering poor quality existing shows, that seems like a real possibility. Wish I'd known this was coming a long time ago. Coulda saved a ton of money spent buying up show sets on Ebay, looking for better quality audio. But a lot of better quality shows were found that way. As well as quite a few "missing" episodes.

I found this.  Apparently the Internet Archive has a project to restore 78 RPM records using AI.  The dataset they're using might be a good starting point for OTR?  link: https://blog.archive.org/2023/04/24/ai-audio-challenge-audio-restoration-based-on-expert-examples/

Also, I read where SPERDVAC is beginning a large project to digitize their archive of radio shows. I wonder if AI is going to be a part of their plans.

For the record, I have an RTX 3060 Ti. Don't know if anyone here has something beefier, but if not, I'm happy to donate some computer time to experiments in this area.

@sun_lmao, Most all of my computer hardware is pretty basic.  I'd love for you to do some experimenting.  And please share what may work.    

I did a quick little experiment with an episode of Michael Shayne that had some light, but annoying clipping.  I used the freely available Adobe Podcast Enhancer (link below).  It did quite a good job of eliminating the clipping and preserving the voices.  However, it did a terrible, yet amusing thing to the music.  It tried to insert voices where the music was, so it came out as musical gibberish.  
If there is some AI application that cleans up music (and for free or very cheaply) I could mix the two tracks, keeping the best of each in my mixing software (Audacity.)

On another occasion I used a Shadow episode that was of very poor quality.  It did the same thing to the music and gave the actors a serious lisp.  I laughed so hard I almost peed my pants.  

link: https://podcast.adobe.com/enhance?ref=aitoptools

Corey Harker has recently been using AI to do light cleanup on the disc transfers that he offers in his quarterly buyers group. They sound quite good.

I experimented with this episode of "Blackstone Magic Detective."  
This is the original MP3 I worked with.  
The Ghost That Trapped a Killer original

After some conventional cleaning, I applied some AI and because the actors "lisped" somewhat, I mixed the two files together and the result is this:  
The Ghost That Trapped a Killer AI mix