AI Seminar
    -------------------------------
    Tuesday, December 9th, 2003
    4:00 pm - 5:30 pm
    175 ATL (Large Conference Room)

    "Singing in vivo"

    Gregory H. Wakefield
    Department of Electrical Engineering and Computer Science
    University of Michigan




----------------------------------
In the 1950's, engineers at IBM predicted that human
dictation would be handled by computer within a decade. Fifty years
later, human dictation is still handled primarily by humans, while
research on speech has led to both highly specialized automatic
recognition systems as well as to a remarkable array of mathematical
tools for solving problems in a variety of disciplines. "Singing in
vivo" looks at the voice as an instrument for communicating music,
rather than language, and outlines several of the outstanding problems
in music information retrieval. While we will avoid making any
marketing predictions, we will present some of the current
state-of-the-art techniques that have been developed both here at
Michigan and at other research laboratories for extracting features of
vocal production which are specific to singing and music. Finally, we
will highlight several of the correspondences we find between singing
in vivo and other application domains. [Work supported by a grant from
the National Science Foundation.]