There are so many concert recordings of awesome performances that sound like crap because they are audience tapes.

Often before a performance recorded music is played and captured in the audience recordings.

Would it be possible to train a model on an archive of these concert audience recordings of studio recordings paired with the original studio recordings to develop a system to “clean up” audience recordings?