We haven't tried it on old 78rpm recordings, but feel free to send one over to thomas@tape.it . The algorithm is trained on a surprisingly small amount of data

We haven't tried it on old 78rpm recordings, but feel free to send one over to thomas@tape.it . The algorithm is trained on a surprisingly small amount of data by today's standards, due to the architecture we chose. There is a lot of room for improvement as we collect more examples, and it's something we definitely intend to do.