Advertisement

deepGTTM-III: Multi-task Learning with Grouping and Metrical Structures

  • Masatoshi HamanakaEmail author
  • Keiji Hirata
  • Satoshi Tojo
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 11265)

Abstract

This paper describes an analyzer that simultaneously learns grouping and metrical structures on the basis of the generative theory of tonal music (GTTM) by using a deep learning technique. GTTM is composed of four modules that are in series. GTTM has a feedback loop in which the former module uses the result of the latter module. However, as each module has been independent in previous GTTM analyzers, they did not form a feedback loop. For example, deepGTTM-I and deepGTTM-II independently learn grouping and metrical structures by using a deep learning technique. In light of this, we present deepGTTM-III, which is a new analyzer that includes the concept of feedback that enables simultaneous learning of grouping and metrical structures by integrating both deepGTTM-I and deepGTTM-II networks. The experimental results revealed that deepGTTM-III outperformed deepGTTM-I and had similar performance to deepGTTM-II.

Keywords

A generative theory of tonal music (GTTM) Grouping structure Metrical structure Deep learning 

Notes

Acknowledgments

This work was supported by JSPS KAKENHI Grant Numbers 17H01847, 25700036, 16H01744, and 23500145.

References

  1. 1.
    Lerdahl, F., Jackendoff, R.: A Generative Theory of Tonal Music. MIT Press, Cambridge (1985)Google Scholar
  2. 2.
    Hirata, K., Hiraga, R.: Ha-Hi-Hun plays Chopin’s Etude. In: Working Notes of IJCAI-03 Workshop on Methods for Automatic Music Performance and Their Applications in a Public Rendering Contest (2003)Google Scholar
  3. 3.
    Hirata, K., Matsuda, S., Kaji, K., Nagao, K.: Annotated music for retrieval, reproduction, and sharing. In: Proceedings of the 2004 International Computer Music Conference (ICMC 2004), pp. 584–587 (2004)Google Scholar
  4. 4.
    Hirata, K., Matsuda, S.: Interactive music summarization based on GTTM. In: Proceedings of the 2002 International Society for Music Information Retrieval Conference (ISMIR 2002), pp. 86–93 (2002)Google Scholar
  5. 5.
    Hamanaka, M., Hirata, K., Tojo, S.: Melody morphing method based on GTTM. In: Proceedings of the 2008 International Computer Music Conference (ICMC 2008), pp. 155–158 (2008)Google Scholar
  6. 6.
    Hamanaka, M., Hirata, K., Tojo, S.: Melody extrapolation in GTTM approach. In: Proceedings of the 2009 International Computer Music Conference (ICMC 2009), pp. 89–92 (2009)Google Scholar
  7. 7.
    Hamanaka, M., Hirata, K., Tojo, S.: Implementing ‘a generative theory of tonal music’. J. New Music Res. 35(4), 249–277 (2006)CrossRefGoogle Scholar
  8. 8.
    Hamanaka, M., Hirata, K., Tojo, S.: FATTA: full automatic time-span tree analyzer. In: Proceedings of the 2007 International Computer Music Conference (ICMC 2007), pp. 153–156 (2007)Google Scholar
  9. 9.
    Miura, Y., Hamanaka, M., Hirata, K., Tojo, S.: Decision tree to detect GTTM group boundaries. In: Proceedings of the 2009 International Computer Music Conference (ICMC 2009), pp. 125–128 (2009)Google Scholar
  10. 10.
    Kanamori, K., Hamanaka, M.: Method to detect GTTM local grouping boundaries based on clustering and statistical learning. In: Proceedings of the 2014 International Computer Music Conference (ICMC 2014), pp. 125–128 (2014)Google Scholar
  11. 11.
    Hamanaka, M., Hirata, K., Tojo, S.: \(sigma\)GTTM III: learning-based time-span tree generator based on PCFG. In: Proceedings of the 11th International Symposium on Computer Music Multidisciplinary Research (CMMR 2015), pp. 303–317 (2015)Google Scholar
  12. 12.
    Hamanaka, M., Hirata, K., Tojo, S.: Musical structural analysis database based on GTTM. In: Proceedings of the 2014 International Society for Music Information Retrieval Conference (ISMIR 2014), pp. 325–330 (2014)Google Scholar
  13. 13.
    Hamanaka, M., Hirata, K., Tojo, S.: deepGTTM-I: local boundary analyzer based on a deep learning technique. In: Proceedings of the 12th International Symposium on Computer Music Multidisciplinary Research (CMMR 2016), pp. 8–20 (2016)Google Scholar
  14. 14.
    Hamanaka, M., Hirata, K., Tojo, S.: deepGTTM-II: automatic generation of metrical structure based on deep learning technique. In: Proceedings of 13th Sound and Music Computing Conference (SMC 2016), pp. 203–210 (2016)Google Scholar
  15. 15.
    Choi, K., Fazekas, G., Sandler, M.: Automatic tagging using deep convolutional neural networks. In: Proceedings of the 2016 International Society for Music Information Retrieval Conference (ISMIR 2016), pp. 805–811 (2016)Google Scholar
  16. 16.
    Zhou, X., Lerch, A.: Chord detection using deep learning. In: Proceedings of the 2015 International Society for Music Information Retrieval Conference (ISMIR 2015), pp. 52–58 (2015)Google Scholar
  17. 17.
    Deng, J., Kwok, Y.: Hybrid Gaussian-HMM-deep learning approach for automatic chord estimation with very large vocabulary. In: Proceedings of the 2016 International Society for Music Information Retrieval Conference (ISMIR 2016), pp. 812–818 (2016)Google Scholar
  18. 18.
    Oord, A., Sander, D., Benjamin, S.: Deep content-based music recommendation. In: Proceedings of the Advances in Neural Information Processing Systems 26 (NIPS 2013), pp. 2643–2651 (2013)Google Scholar
  19. 19.
    Sigtia, S., Benetos, E., Dixon, S.: An end-to-end neural network for polyphonic piano music transcription. IEEE/ACM Trans. Audio Speech Lang. Process. (TASLP) 24(5), 927–939 (2016)CrossRefGoogle Scholar
  20. 20.
    Hinton, G.E., Osindero, S., Teh, Y.W.: A fast learning algorithm for deep belief nets. Neural Comp. 18, 1527–1554 (2006)MathSciNetCrossRefGoogle Scholar
  21. 21.
    MakeMusic Inc., “Finale” (2018). http://www.finalemusic.com/

Copyright information

© Springer Nature Switzerland AG 2018

Authors and Affiliations

  • Masatoshi Hamanaka
    • 1
    Email author
  • Keiji Hirata
    • 2
  • Satoshi Tojo
    • 3
  1. 1.RIKENTokyoJapan
  2. 2.Future University HakodateHakodateJapan
  3. 3.JAISTNomiJapan

Personalised recommendations