Josep Lluis Arcos and Joan Serra together with two collaborators from the Max Plank Institute in Germany participated to the annual MIREX international evaluation framework for Music Information Retrieval algorithms, where participants submit their algorithms as binary executables and they run such algorithms on different public and pivate datasets.

It serves as the main evaluation reference for the community (in the same vein of TREC for text retrieval, TRECVID for video, or CLEF for image retrieval).

The IIIA participated in the Structure Segmentation task, where one has to "cut" songs into coherent sections in an unsupervised way (sections as intro, chorus, verse, etc.). The results have been posted today, and this team won on all of the 4 evaluated datasets. A brief summary of the results follows:

MIREX09 Dataset:
- Ours: F_A = 0.652 / F_B3 = 0.644
- Next competitor: F_A = 0.571 / F_B3 = 0.591

RWC-Quaero Dataset:
- Ours: F_B3 = 0.765
- Next competitor: F_B3 = 0.660

RWC-AIST Dataset:
- Ours: F_A = 0.688 / F_B3 = 0.710
- Next competitor: F_A = 0.604 / F_B3 = 0.643

SALAMI Dataset:
- Ours: F_A = 0.580 / F_B3 = 0.492
- Next competitor: F_A = 0.572 / F_B3 = 0.490