Share Email Print
cover

Proceedings Paper

Model-based approach to partial tracking for musical transcription
Author(s): Andrew Sterian; Gregory H. Wakefield
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

We present a new method for musical partial tracking in the context of musical transcription using a time-frequency Kalman filter structure. The filter is based upon a model for the evolution of a partial behavior across a wide range of pitch from four brass instruments. Statistics are computed independently for the partial attributes of frequency and log-power first differences. We present observed power spectral density shapes, total powers, and histograms, as well as least-squares approximations to these. We demonstrate that a Kalman filter tracker using this partial model is capable of tracking partials in music. We discuss how the filter structure naturally provides quality-of-fit information about the data for use in further processing and how this information can be used to perform partial track initiation and termination within a common framework. We propose that a model-based approach to partial tracking is preferable to existing approaches which generally use heuristic rules or birth/death notions over a small time neighborhood. The advantages include better performance in the presence of cluttered data and simplified tracking over missed observations.

Paper Details

Date Published: 2 October 1998
PDF: 12 pages
Proc. SPIE 3461, Advanced Signal Processing Algorithms, Architectures, and Implementations VIII, (2 October 1998); doi: 10.1117/12.325677
Show Author Affiliations
Andrew Sterian, Univ. of Michigan (United States)
Gregory H. Wakefield, Univ. of Michigan (United States)


Published in SPIE Proceedings Vol. 3461:
Advanced Signal Processing Algorithms, Architectures, and Implementations VIII
Franklin T. Luk, Editor(s)

© SPIE. Terms of Use
Back to Top