End-to-End System Design for Music Style Transfer with Neural Networks
The objective for this research is to design an end-to-end performance style transfer system that can assist performers and composers by modeling the relationships between composition and performance. It leads to the central question for this research: can we learn the relationships between composition and performance and generate an expressive performance conditioned on specific composition styles? With the assistance of neural network techniques including recurrent neural networks, deep generative models and so on, we might be able to disentangle and separately model performance and composition styles, or jointly model them using an end-to-end network. Based on such model, the architecture can contribute to our understanding of music interpretation and collaborate with performers and composers.
C4DM theme affiliation:
Music Cognition, Audio Engineering