Title :
A framework for analysis of music similarity measures
Author :
Jensen, Jesper Hojvang ; Christensen, Mads Grasboll ; Jensen, Soren Holdt
Author_Institution :
Dept. of Electron. Syst., Aalborg Univ., Aalborg, Denmark
Abstract :
To analyze specific properties of music similarity measures that the commonly used genre classification evaluation procedure does not reveal, we introduce a MIDI based test framework for music similarity measures. We introduce the framework by example and thus outline an experiment to analyze the dependency of a music similarity measure on the instrumentation of a song compared to the melody, and to analyze its sensitivity to transpositions. Using the outlined experiment, we analyze music similarity measures from three software packages, namely Marsyas, MA toolbox and Intelligent Sound Processing toolbox. The tested timbral similarity measures perform instrument recognition relatively well, although they are sensitive to transpositions and differences between sound fonts. The beat/rhythm/melody similarity measures are not always able to recognize the same melody played with different instruments.
Keywords :
music; pattern classification; MIDI; beat/rhythm/melody similarity measures; classification evaluation procedure; intelligent sound processing tool-box; music similarity measures; test framework; Accuracy; Histograms; Instruments; Mel frequency cepstral coefficient; Multiple signal classification; Timbre;
Conference_Titel :
Signal Processing Conference, 2007 15th European
Conference_Location :
Poznan
Print_ISBN :
978-839-2134-04-6