Some recent audio/video projects that use generative processes to yield surprising results. "Temporide" does a pixel-by-pixel delay on a video, showing many time lapses simultaneously. Spectral splicing, morphing, and reconstitution creates new audio based out of what you feed it. And "Ghost Jockey" generates a continuous stream of mashup audio and video.
Daniel Iglesia makes electronic music and video, or more accurately, is lazy and creates machines that do it for him.
----
In this talk, I will present my work on sound source separation with applications for music. Music is repetitious in nature and this repetition actually informs the source separation process. I derive an automated statistical approach based entirely on repetitive structure to separate sound sources. In addition, spectrograms contain time-frequency structure. This structure may be factored into note-like components containing a spectral shape modulated by an amplitude envelope. When multiple spectrograms are available, I show how to incorporate this additional spatial information to separate components and combine them to form the original source signals.
Mitchell Parry is a Ph.D. candidate in the College of Computing working with Irfan Essa. His research interests include source separation, signal processing, visualization and music information retrieval.
