I want to create a piece of software which can recognize a song from a sound recording, in which the user of the software has sung or hummed a part of the song that he/she remembers, so kind of like SoundHound. For that, I need access to the songs that it should be able to recognize, since
- It should be able to match the recorder sound with the song, and
- It needs to train a model on the songs since it is machine learning based.
Can I get the access I need to basically any famous song (which the software should be able to recognize) for free or do I need to pay money for such access? How does the licensing of the songs affect this?