Sub-symbolic audio analysis is a kind of analysis that listens to the actual audio itself. A sub-symbolic music similarity engine makes recommendations to help users discover music based on what a given song actually sounds like.
The statistical method is an alternative and traditional approach. It is “tone deaf” in the sense that it has no idea what any of your music actually sounds like. If Bob likes songs A, B, and C; and Sue likes A and B too; and the system thinks Bob and Sue are pretty similar, it will suggest song C to Sue.
The system doesn’t listen to the song, but it still might have made a relevant recommendation. This works to a certain extent, but quickly starts to fall apart due to the cold start problem.
If a song is new to the system and nobody has rated it yet, it can’t recommend it to anyone. Or, if a user is new, it can’t make a recommendation either because the user hasn’t listened to any music yet.
