2005:Audio Genre Classification Results

Goal: To classify polyphonic music audio (in PCM format) into genre categories.

Dataset: Two sets of data were used: Magnatune and USPOP. The Magnatune dataset has a hierarchical genre taxonomy, while the USPOP categories are at a single level. The audio sampling rates used were either 44.1 KHz or 22.05 KHz (mono). More data information is in the following table:

Dataset	Size (@ 44.1 KHz)	Number of Training Files	Number of Testing Files
Magnatune	34.3 GB	1005	510
USPOP	28.4 GB	940	474

OVERALL
Rank	Participant	Mean of Magnatune Hierarchical Classification Accuracy and USPOP Raw Classification Accuracy
1	Bergstra, Casagrande & Eck (2)	82.34%
2	Bergstra, Casagrande & Eck (1)	81.77%
3	Mandel & Ellis	78.81%
4	West, K.	75.29%
5	Lidy & Rauber (SSD+RH)	75.27%
6	Pampalk, E.	75.14%
7	Lidy & Rauber (RP+SSD)	74.78%
8	Lidy & Rauber (RP+SSD+RH)	74.58%
9	Scaringella, N.	73.11%
10	Ahrendt, P.	71.55%
11	Burred, J.	62.63%
12	Soares, V.	60.98%
13	Tzanetakis, G.	60.72%

Rank Participant Mean of Magnatune Hierarchical Classification Accuracy and USPOP Raw Classification Accuracy 1 Bergstra, Casagrande & Eck (2) 82.34% 2 Bergstra, Casagrande & Eck (1) 81.77% 3 Mandel & Ellis 78.81% 4 West, K. 75.29% 5 Lidy & Rauber (SSD+RH) 75.27% 6 Pampalk, E. 75.14% 7 Lidy & Rauber (RP+SSD) 74.78% 8 Lidy & Rauber (RP+SSD+RH) 74.58% 9 Scaringella, N. 73.11% 10 Ahrendt, P. 71.55% 11 Burred, J. 62.63%