Noise reduction software often claims to be AI-based.
iZotope RX can sometimes do miraculous things. It's great for cleaning up dialogue tracks where there's a small amount of unwanted noise or too much reverb, and can recognise some very specific problems, such as mouth noises that actors often make. People make these noises all the time, but you can't hear them when you're listening to someone talk in the real world, but they can sound bad in recorded speech.
It still hasn't made the very expensive CEDAR noise reduction redundant.
Melodyne can convert audio to MIDI, but it's not yet possible for it to infer the pianist's pedalling. Maybe a little human intervention could help here.
In Steinberg's Nuendo, you can replace dialogue from one take with that of another, and it automatically realigns the syllables individually with the original dialogue. This guarantees that the new dialogue is in sync with the picture. It can also time-align different mics, such as boom and lavs (lapel mics). Doing this stuff manually is a lethal combination of tedious and stressful.