Automatic music recommendation has become an increasingly relevant problem in recent years, since a lot of music is now sold and consumed digitally. Most recommender systems rely on collaborative filtering. However, this approach suffers from the cold start problem: it fails when no usage data is available, so it is not effective for recommending new and unpopular songs. We propose to use a latent factor model for recommendation, and predict the latent factors from music audio using a deep convolutional neural network when they cannot be obtained from usage data.

Our demo processes music clips obtained from YouTube, selected by the user, and finds other clips with similar (predicted) usage patterns from a large database of 600,000 songs (a subset of the Million Song Dataset).