Given a musical song, automatically analyze it to determine all its musical characteristics, e.g., styles. Deconstruction. Inspired by various rhythmic musical styles. This seems relatively straightforward with machine learning, though analogous to the difficult object recognition problem (in a complicated scene) in machine vision.
Various companies, Pandora, Google, have tried this in hopes of suggesting more music that a person will like, but have found it is not too effective. Collaborative filtering works better.
The new tack is instead to provide it intended as a teaching tool: find me more examples of this musical "thing", an educational tool to help people compose more music.
Generalize the idea to automatically analyzing other genres of art. Text, i.e., literary fiction. 2D paintings and photographs. Movies (training examples could be tropes enumerated in TvTropes), though we also want things like cinematic techniques like camera work if the tool is to educate filmmakers.
No comments :
Post a Comment