Freed, Adrian adrian@cnmat.berkeley.edu

Center for New Music and Audio Technologies
1750 Arch Street
Berkeley, CA 94709
USA
phone (510) 643-9990, fax (510) 642-7918

ICMC 1997 Poster Proposal

Database of Challenging Musical Sounds for Evaluation and Refinement of Pitch Estimators

Keywords: Pitch estimation, Standardized databases, Fundamental frequency estimation

Content Area: Audio Analysis and Resynthesis

Speech researchers have made the most thorough study of the performance of pitch estimation algorithms. A key method they use is to evaluate algorithm performance with standardized databases of speech that have been "hand" analyzed. Such a database does not exist for musical signals.

As a result, pitch estimation papers in the computer music community describe algorithms evaluated using short sound examples often chosen to show new work in a good light. It is therefore impossible to predict performance of published algorithms in real musical situations, and difficult for researchers to identify the important areas of fruitful new work. To address these problems we are assembling a publicly available database of sound files and accurate analyses.

Sounds in this database can be grouped into two important and not well represented categories: complete musical phrases, and challenging examples. Complete musical phrases are used to evaluate the impact of estimation errors in common musical contexts. Challenging examples are used to identify particular points of weakness from which an algorithm may suffer. Included are sounds with: pitch synchronous and additive noise, room ambiance, cross-talk from adjacent strings, ambiguous octaves, inharmonicity, missing fundamentals, glissandi, vibrato and trills.

In this poster, we will present an overview of the contents of the sound database, explain how new sounds can be contributed to it, and describe how it has been used to refine new pitch estimation algorithms.