Well, you could certainly just download from archive.org for the first-time usage of any particular song and then cache them from there. Like being your own CDN.
This is a good idea. I did this with about 38,000 books in the public domain and hosted and indexed the text files on the page so I could do quick full text searches of all of them: https://www.locserendipity.com/Google.html. Interestingly, it is not nearly as good as a RegEx search of a directory of the same content, revealing some inconsistencies in their search algorithm (at least for custom search engines).
