A new method for approximate melody matching

Author

Wu, Ya-Dong ; Li, Yang ; Liu, Bao-Long

Author_Institution

Dept. of Comput. Sci. & Eng., Shanghai Jiao Tong Univ., China

Volume

5

fYear

2003

fDate

2-5 Nov. 2003

Firstpage

2687

Abstract

This paper represents a new method for approximate melody matching, called Linear Alignment Matching (LAM). It takes a geometrical approach and compares melodies by matching their pitch-time contours. Error toleration mechanism is added to make the method robust against humming errors, and the melody similarity is computed from the viewpoint of both rhythm and pitch. A Query-By-Humming system is built based on LAM to test its performance 62 humming clips are searched in a music database comprising 3864 songs, and LAM\´s top-3 success rate is 90.3%, winning the traditional approximate string matching algorithm by 11%. This experiment result shows the validity of the "Linear Alignment Matching" method and the feasibility of using it as the query engine in large digital music libraries.

Keywords

audio databases; content-based retrieval; digital libraries; music; string matching; very large databases; LAM; approximate melody matching; content-based music retrieval; digital music libraries; error toleration mechanism; humming clips; humming errors; linear alignment matching; music database; pitch-time contours; query engine; query-by-humming system; rhythm; string matching algorithm; Computer errors; Computer science; Content based retrieval; Engines; Feature extraction; Matched filters; Music information retrieval; Rhythm; Robustness; Spatial databases;

fLanguage

English

Publisher

ieee

Conference_Titel

Machine Learning and Cybernetics, 2003 International Conference on

Print_ISBN

0-7803-8131-9

Type

conf

DOI

10.1109/ICMLC.2003.1259994

Filename

1259994