Evaluating a Melody Extraction Engine

  • Authors:
  • Thomas Sødring;Alan F. Smeaton

  • Affiliations:
  • -;-

  • Venue:
  • Proceedings of the 24th BCS-IRSG European Colloquium on IR Research: Advances in Information Retrieval
  • Year:
  • 2002

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper introduces the CEOLAIRE Music Information Retrieval System; a system which stores, indexes and provides content-based retrieval on a collection of over 7,000 music files. What makes CEOLAIRE different from most other music information retrieval systems is that it indexes actual raw compressed or uncompressed audio rather than just indexing MIDI, which is effectively instructions for generating musical notes. The paper includes an overview of the CEOLAIRE system and includes an evaluation of the effectiveness of its melody extraction engine, the crucial part of CEOLAIRE which recognises the notes and melody being played. Our results show that for the type of melody matching used in CEOLAIRE'S retrieval engine, the performance of our melody recognition is quite acceptable.