This is the number of percentage points lost according to accuracy 1 when
computed on the 465 song excerpts after distortions of the signal:
downsampling/resampling,
gsm encoding/decoding, filtering, volume change, addition of
reverberation and white noise (with a signal-to-noise ratio of 20 dB).
- Anonymous -0.22
- Dixon_auco 0.22
- Klapuri 0.43
- Scheirer 1.29
- Tzan_histsumbands 3.48
- Uhle 4.52
- Dixon_trac 5.08
- Tzan_medsumbands 6.24
- Tzan_medmultibands 6.45
- Dixon_indu 12.47
- Alonso_auco 15.91
- Alonso_sppr 30.32
NB: Slight changes after fixing a few ground-truth errors (on the `song' dataset) on November 16th 2004.
Fabien Gouyon
2004-11-16
