eduzhai > Applied Sciences > Engineering >

Generative Modelling for Controllable Audio Synthesis of Expressive Piano Performance

  • Save

... pages left unread,continue reading

Document pages: 3 pages

Abstract: We present a controllable neural audio synthesizer based on Gaussian MixtureVariational Autoencoders (GM-VAE), which can generate realistic pianoperformances in the audio domain that closely follows temporal conditions oftwo essential style features for piano performances: articulation and dynamics.We demonstrate how the model is able to apply fine-grained style morphing overthe course of synthesizing the audio. This is based on conditions which arelatent variables that can be sampled from the prior or inferred from otherpieces. One of the envisioned use cases is to inspire creative and brand newinterpretations for existing pieces of piano music.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...