Music genre identification is useful not only for commercial applications like recommendation systems but also for deepening cultural engagement with music. Machine learning approaches have traditionally treated this as a “bucket sorting” problem, but music is experientially and culturally much more connected than discrete buckets.

Although there has been extensive prior work on identifying the genre of recorded music, this work has been criticized for its narrow use of deep learning tools, dataset limitations, and lack of engagement with genre theory.

This project explores whether a neural network can be trained to recognize genre with greater contextual sensitivity, informed by genre theory and inter-genre relationships.