Key labels for GTZAN data set

The GTZAN data set is probably one of the most prominent data sets used in research related to Music Information Retrieval and Audio Content Analysis. The data set contains 1000 excerpts from songs, sorted into 10 genres.

While this set is old and it is clear that it has its disadvantages [1], it is a well-known, widely-used, and, last but not least, easily available set. This is why I chose this data set to annotate the musical keys of the excerpts. The complete annotation can be found here.


I used my tuning fork to identify the key of each song. I did not label excerpts that

  • included modulations, or
  • were particularly difficult to identify.


Genre Major Keys Minor Keys # annotated
blues 39598
country 94599
disco 435598


Besides the obvious application of directly evaluating key detection systems, we used this data set also for an indirect evaluation of a detection of tonal components in a spectrum [2].


[1] Sturm, Bob: An Analysis of the GTZAN Music Genre Dataset, Proceedings of the 2nd International ACM Workshop on Music Information Retrieval with User-Centered and Multimodal Strategies (MIRUM), Nara, Japan, November, 2012

[2] Kraft, Sebastian; Lerch, Alexander, Zoelzer, Udo: The Tonalness Spectrum: Feature-Based Estimation of Tonal Components, Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx), Maynooth, Ireland, September 2-5, 2013


3 thoughts on “Key labels for GTZAN data set

  1. Hi Alexander,

    Li and Chan have also annotated the keys of GTZAN. Their annotations are here:

    T. Li and A. Chan, “Genre classification and the invariance of MFCC features to key and tempo,” in Proc. Int. Conf. MultiMedia Modeling, (Taipei, China), Jan. 2011.

    • Thanks for the link. I did know about the annotations, but I did not know they were publicly available. I added the link to my data set list on the links page.


Leave a Reply

Your email address will not be published. Required fields are marked *