Computational modelling of music similarity constitutes a key element for music information retrieval and recommendation systems. Similarity models and their analysis are also important for research in musicology and music perception. In this study, we test feature preprocessing with Restricted Boltzmann Machines in combination with established methods for learning distance measures. Our experiments show that this preprocessing improves the overall generalisation results of the trained models. We compare the effects of feature preprocessing on distance function learning using gradient ascent and support vector machines. The evaluation is performed using similarity data from the MagnaTagATune dataset, which allows a comparison of our results with previous studies.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=17107
(284KB)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
