Technical Program

Paper Detail

Paper: PS-2B.32
Session: Poster Session 2B
Location: Symphony/Overture
Session Time: Friday, September 7, 19:30 - 21:30
Presentation Time:Friday, September 7, 19:30 - 21:30
Presentation: Poster
Paper Title: Convolutional Neural Network Achieves Human-level Accuracy in Music Genre Classification
Manuscript:  Click here to view manuscript
Authors: Mingwen Dong, Rutgers University, United States
Abstract: Music genre classification is one example of content-based analysis of music signals. Traditionally, human-engineered features were used to automatize this task and 61% accuracy has been achieved in the 10-genre classification. Here, we propose a method that achieves human-level accuracy (70%) in the same classification task. The method is inspired by knowledge of human perception study in music genre classification and the neurophysiology of the auditory system. It works by training a simple convolutional neural network (CNN) to classify short segments of music waveforms. During prediction, the genre of an unknown music is determined as the majority vote of all classified segments from a music waveform. The filters learned in the CNN qualitatively resemble the spectro-temporal receptive fields (STRF) in the auditory system and potentially provide insights about how human auditory system classifies music genre.