Wednesday, January 10, 2007
I'm giving a talk at Yahoo! Research Berkeley on Friday January 12
I will present ways to analyze music and audio based on self-resemblance, using a "similarity matrix" representation inter-frame spectral similarity. Besides yielding an engaging visualization of musical structure, this approach can be used to segment and index audio by content. I will present the fundamental ideas, musical examples, and some interesting applications and offshoots, including the "beat spectrum," a novel way to characterize musical rhythm. I will also present a system for retrieving music by rhythmic similarity, and discuss applications such as audio abstracting and automatic podcast indexing. If time permits I will present some "minute madness" of other work such as VideoBots?, FlyAbout, and some current speculative projects.