T. Lidy, A. Schindler:
"Parallel convolutional neural networks for music genre and mood classification";
Report for Music Information Retrieval Evaluation eXchange (MIREX 2016);
Our approach to the MIREX 2016 Train/Test Classification Tasks for Genre, Mood and Composer detection is based on an approach combining Mel-spectrogram transformed audio and Convolutional Neural Networks (CNN). We utilize two different CNN architectures, a sequential one, and a parallel one, the latter aiming at capturing both temporal and timbral information in two different pipelines, which are merged on a later stage. In both cases, the crucial CNN parameters such as filter kernel sizes and pooling sizes were carefully chosen after a range of experiments.
Neural networks, Deep learning, Classification, Music, Audio, Convolutional Neural Networks
Electronic version of the publication:
Created from the Publication Database of the Vienna University of Technology.