1st Prize in the MIREX Symbolic Melodic Similarity Contest.
To appear in on-line proceedings:
http://www.music-ir.org/evaluation/mirex-results/sym-melody/index.html
We present a method for melody retrieval using the Implication/
Realization (I/R) model, a model of melodic structure.
As a preprocessing step to retrieval, all melodies to
be compared are analysed using our I/R parser. The similarity
ranking of melodies from data base melodies with
respect to the query is determined by computing the edit
distance between the I/R analysis of the query and those
of the data base melodies. The parameters in the distance
measure were optimized by a genetic algorithm, evaluating
on melodic similarity ground truth. In the MIREX ’05
contest for symbolic melodic similarity, our method was
compared to various other methods using the RISM A/II
database, and performed best. The results of the contest
are presented and discussed at the end of the paper.
This
research has been partially supported by the Spanish Ministry
of Science and Technology under the project TIC
2003-07776-C2-02 “CBR-ProMusic: Content-based Music
Processing using CBR” and EU-FEDER funds.
Peer reviewed