From: malcolm@interval.com (Malcolm Slaney)
Date: Thu, 19 Oct 1995 17:05:58 -0700
Subject: Music Transcription and Auditory Processing at CCRMA
Message-Id: <v02130512acac9685d8ea@[199.170.106.94]>

The next couple of talks at the CCRMA Hearing Seminar address the signal
processing needed to model the cochlea and the auditory system.  The first
talk, by Rolf Woehrmann, addresses the use of wavelets to implement
auditory filtering.  The second talk, by Richard F. Lyon, talks about
filter's that more accurately reflect the phsyiological and psychophysical
data.

        Who:    Rolf Wohrmann (Visiting Scholar CCRMA)
        What:   Wavelets and Auditory Filters for Polyphonic Transcription
        When:   Thursday, October 26 at 11AM
        Where:  CCRMA Library, Top floor of the Knoll at Stanford

Transcribing music is one of the holy grails of music perception research.
How do we identify notes and tell when they start when multiple sounds are
played?  Can auditory research make the problem any easier?   Rolf's talk
will describe the preprocessing steps needed to build such a system.

See you at CCRMA next Thursday!

-- Malcolm



From: Rolf Wohrmann <rolf@ccrma.Stanford.EDU>
Subject: abstract

   Preprocessing for Automated Transcription of Polyphonic Music:
             Linking Wavelet Theory and Auditory Filtering

  - Rolf Woehrmann, Visiting Scholar/Composer, CCRMA -

[This research project is the result of an ongoing collaboration
with Ludger Solbach at the Technische Universitaet Hamburg-Harburg,
Germany. The talk itself is based on a paper we presented at the
ICMC in Banff as well as on a presentation at the CASA workshop at
the IJCAI in Montreal.]

        First I will present our method for the calculation of a
linear time-frequency distribution. We are using this as a
preprocessing step for automated transcription of polyphonic music
and for computational auditory scene analysis. Complex-valued
continuous wavelet transforms are normally computational expensive,
though using faster methods like the 'A Trous' algorithm. By linking
the gammatone filter auditory model to wavelet transforms we gain
the benefits of a faster computation and a more auditory oriented
preprocessing.
        A method for onset detection in polyphonic music will be
presented. Also, I will show some potential of our preprocessing
method for the detection and grouping of harmonics. Transcription
examples of western and non-western music will be given. Finally, I
derived score lists from the transcriptions and synthesized them.
Playing the results along with their original will show the
potential of our approach.
        I will briefly introduce the software used to generate
these examples, together with an application called AnnaLisa, which
is currently being developed for use especially in ethnomusicology
and to visualize unnotated music such as jazz, some electro-acoustic
and computer music, etc. Future directions of our research will
conclude my talk.