You can't do much to improve the sound when the problem is what was not captured in the recording, i.e. the fundamental problem being the very restricted frequency and dynamic ranges of early recordings. Anything you do requires making assumptions about what might have been lost which, of course, you can never really know. You can try reducing noise but that isn't possible without the high likelihood of also removing some of what you actually want to keep.What will be nice is if they can use AI to improve the sound of early opera recordings. Now that is where the research should be focused. It is not perfect. You would not believe what AI thinks I say. I was writing to a friend to tell her I looked forward to seeing what her pastries were like and it wrote out " I look forward to seeing what her panties were like".