Music generation with variational recurrent autoencoder supported by history
This work addresses music generation for creative applications, but it appears incremental as it builds on existing variational autoencoder and recurrent network methods.
The authors tackled the problem of generating longer melodic patterns in music by introducing a variational autoencoder supported by history architecture combined with post-generation filtering, resulting in acoustically pleasing and melodically diverse music.
A new architecture of an artificial neural network that helps to generate longer melodic patterns is introduced alongside with methods for post-generation filtering. The proposed approach called variational autoencoder supported by history is based on a recurrent highway gated network combined with a variational autoencoder. Combination of this architecture with filtering heuristics allows generating pseudo-live acoustically pleasing and melodically diverse music.