[Zurück]

T. Lidy, A. Rauber:
"MIREX 2006: Computing Statistical Spectrum Descriptors for Audio Music Similarity and Retrieval";
Poster: MIREX - Music Information Retrieval Evaluation eXchange, Victoria, Kanada; 08.10.2006 - 12.10.2006; in: "MIREX 2006", (2006), 5 S.

This paper describes our submission to the MIREX 2006 Audio Music Similarity and Retrieval task.
The task was to submit an audio feature extraction algorithm, to compute music similarity measures and to return a distance matrix from an audio collection consisting of 5000 pieces, which was subsequently evaluated through human listening tests as well as objective statistics.
We submitted a new implementation of the Statistical Spectrum Descriptor (SSD) audio feature extractor and computed the distance matrix directly from feature space.
Results from the human evaluation show that our approach is among the top 5 algorithms which furthermore show no statistically significant performance differences. The evaluation of a number of objective statistics ranked our algorithm 3rd in most of the cases. Our submission was one of the two fastest in terms of total runtime, having the shortest distance computation time.

The approach has also been evaluated on Audio Cover Song Identification, where it was the best-performing "Audio Music Similarity and Retrieval" submission, outperformed, however, by 4 submissions which were specifically designed for the cover identification task.

http://aleph.ub.tuwien.ac.at/F?base=tuw01&func=find-c&ccl_term=AC06586800

http://publik.tuwien.ac.at/files/pub-inf_3611.pdf