Systems Seminar
Beat Tracking and the Search for Regularity in Complex Data
Prof. Bill Sethares
UW ECE Department
Abstract
Humans are very good at identifying complex
patterns: the auditory system easily senses intricate
periodicities such as the rhythms that normally occur
in music and speech, the visual system readily grasps
the symmetries and repetitions inherent in textures
and tilings, the mind searches for simple underlying regularities to
explain phenomena that appear complex and irregular. Computers are
comparatively poor at locating such patterns.
Beat tracking methods attempt to automatically synchronize
to complex nonperiodic (yet repetitive) waveforms;
to create an algorithm that can "tap its foot"
in time with the music. The beat tracking problem
is important as a step in understanding how people
process temporal information and has direct applications
in the editing of audio/video data,
in synchronization of visuals with audio, in
audio information retrieval and in audio segmentation.
This talk discusses a method of determining the beat from
digitized audio that is based on two ideas.
The first is a method of data reduction
that creates a collection of ``rhythm tracks''
which represent the rhythmic structure of
the piece. Each track uses a different method of
(pre)processing the audio, and so provides a (somewhat)
independent representation of the beat.
The second idea is to model the rhythm tracks
(in simplified form) as a collection of random
variables with changing variances: the variance
is small when "between" the beats and large when
"on" the beat. Exploiting this simple stochastic
model of the rhythm tracks allows the beat detection
to proceed using Bayesian methods.
Audio examples are used to highlight the
strengths and weaknesses of the proposed method.
Time and Place: Wed., Nov. 17, at 3:30 pm in 4610 Engr. Hall.
SYSTEMS SEMINAR WEB PAGE:
http://homepages.cae.wisc.edu/~gubner/seminar/