However,their desirable characteristics such asanalytic invertibility come atthe cost of restricting the functional forms.

This led to successful applications of Neural ODEs in machine learning.

In the last step, we implement Midi synthesizers to generate realistic music.Audeoconverts video to audio smoothly and clearly withonlyafewsetupconstraints.Weevaluate Audeoonpianoperformancevideos collected from YouTube and obtain that their generated music is of reasonable audio quality andcanbesuccessfully recognized withhighprecision bypopular music identification software. The source code with examples is available in a Githubrepository3.