2017/07/14 by Gaëtan Hadjeres, Frank Nielsen, Hadjeres, Gaëtan +3 · 2 citations
Computer Science · Neuroscience · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Music Technology and Sound Studies #Music and Audio Processing #Neuroscience and Music Perception
paper · pdf · doi:10.48550/arxiv.1707.04588
openalex publication_date 2017/07/14 · openalex created_date 2022/10/06 · openalex updated_date 2026/07/28
VAEs (Variational AutoEncoders) have proved to be powerful in the context of\ndensity modeling and have been used in a variety of contexts for creative\npurposes. In many settings, the data we model possesses continuous attributes\nthat we would like to take into account at generation time. We propose in this\npaper GLSR-VAE, a Geodesic Latent Space Regularization for the Variational\nAutoEncoder architecture and its generalizations which allows a fine control on\nthe embedding of the data into the latent space. When augmenting the VAE loss\nwith this regularization, changes in the learned latent space reflects changes\nof the attributes of the data. This deeper understanding of the VAE latent\nspace structure offers the possibility to modulate the attributes of the\ngenerated data in a continuous way. We demonstrate its efficiency on a\nmonophonic music generation task where we manage to generate variations of\ndiscrete sequences in an intended and playful way.\n