Convolution-based classification of audio and symbolic representations of music

Abstract

We present a novel convolution-based method for classification of audio and symbolic representations of music, which we apply to classification of music by style. Pieces of music are first sampled to pitch–time representations (spectrograms or piano-rolls) and then convolved with a Gaussian filter, before being classified by a support vector machine or by k-nearest neighbours in an ensemble of classifiers. On the well-studied task of discriminating between string quartet movements by Haydn and Mozart, we obtain accuracies that equal the state of the art on two data-sets. However, in multi-class composer identification, methods specialised for classifying symbolic representations of music are more effective. We also performed experiments on symbolic representations, synthetic audio and two different recordings of The Well-Tempered Clavier by J. S. Bach to study the method’s capacity to distinguish preludes from fugues. Our experimental results show that our approach performs similarly on symbolic representations, synthetic audio and audio recordings, setting our method apart from most previous studies that have been designed for use with either audio or symbolic data, but not both.

Keywords:

• Classification algorithms
• composer classification
• genre classification
• convolution
• filtering
• audio music classification
• symbolic music classification

Acknowledgements

The authors would like to thank Peter van Kranenburg and Ruben Hillewaere for sharing with us the Haydn and Mozart string quartet data-sets; and the anonymous reviewers for their insightful comments.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

Gissel Velarde, Aalborg University, Department of Architecture, Design and Media Technology, Rendsburggade 14, Building: 6-314, 9000 Aalborg, Denmark. Email: [email protected].

1 Results published by MIREX 2016 (http://music-ir.org/mirex/wiki/2016:MIREX2016_Results).

2 For the horn sound we use the SYNTHTYPE function of the Matlab MIDI Toolbox (Eerola & Toiviainen, 2003). The horn sound was used as it was the best choice of the two available sounds in the toolbox that we used for rendering (the alternative was Shepard tones). For the string sound we used fluidsynth (www.fluidsynth.org) with FluidR3 GM sound font.

3 Toolbox accessed from http://www.cs.tut.fi/sgn/arg/CQT/ on 28 August 2015.

4 http://www.music-cog.ohio-state.edu/Humdrum/representations/kern.html

5 http://www.musedata.org/encodings/bach/bg/keybd/. Accessed on 23 February 2015.

6 http://www.music-cog.ohio-state.edu/Humdrum/representations/kern.html

Additional information

Funding

The work for this paper carried out by G. Velarde, D. Meredith, C. Cancino Chacón and M. Grachten was done as part of the EC-funded collaborative project, ‘Learning to Create’ (Lrn2Cre8). The project Lrn2Cre8 acknowledges the financial support of the Future and Emerging Technologies (FET) programme within the Seventh Framework Programme for Research of the European Commission, under FET [grant number 610859]. G. Velarde is also supported by a PhD fellowship from the Department of Architecture, Design and Media Technology, Aalborg University. C. Cancino-Chacónb and M. Grachten have also been funded by the European Research Council (ERC) under the EU’s Horizon 2020 Framework Programme (ERC Grant Agreement number 670035, project CON ESPRESSIONE).