Article

Received 11.03.2022

Revised 17.07.2022

Accepted 22.08.2022

Retrieved from Vol. 26, No. 3, 2022

Pages 95 -103

Gavrylenko, О., Dmytruk, J., & Teteruk, А. (2022). Mathematical models and artificial intelligence tools usage for automation of writing musical compositions. The National Transport University Bulletin, 26(3), 95-103. https://doi.org/10.33744/2308-6645-2022-3-53-095-103

Mathematical models and artificial intelligence tools usage for automation of writing musical compositions

О. Gavrylenko J. Dmytruk А. Teteruk

The task of automating the process of music creation gained popularity among researchers after the emergence of new formats for the presentation of music data and the development of software for music analysis, which made it possible to transform audio materials into sets of structured data. The aim of this research is solve the problem of the duration of the generated composition and increase the level of uniqueness of the output fragment. To solve the problem, a combination of approaches using neural networks and Markov chains is proposed. Neural network that takes the numerical representation of the reference audio file as training data prepares audio at the first stage. At the second stage, the Markov chain approach is applied to the neural network generation result. The LSTM neural network was chosen for the solution. The article briefly describes the network architecture and the approach to building up the structure of Markov chain. Solution output proves that random selection of intervals when applying the Markov chain approach to the fragment generated by the neural network allows to increase the uniqueness of the generated melody. The issue of imposing restrictions on the choice of intervals that will allow maintaining the tempo and harmony of the original fragment remains open.

 

LSTM; Markov chain; neural network; music generation
  1. Shulei Ji, Jing Luo, Xinyu Yang. A Comprehensive Survey on Deep Music generation: Multi-level Representations, Algorithms, Evaluations, and Future Directions. 2020 [Electronic resource] – https://arxiv.org/pdf/2011.06801.pdf.
  2. Personalized Popular Music Generation Using Imitation and Structure / Shuqi Dai et al. 2021 [Electronic resource] – https://arxiv.org/pdf/2105.04709.pdf.
  3. Symbolic Music Generation with Transformer-GANs / Aashiq Muhamed et al. 2021 [Electronic resource] – https://ojs.aaai.org/index.php/AAAI/article/view/16117.
  4. Hasim Sak, Andrew Senior, Francoise Beaufays. Long Short-Term Memory Recurrent Neural Network Architectures for Large Scale Acoustic Modeling [Electronic resource] – https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43905.pdf.
  5. F. A. Gers, J. Schmidhuber, and F. Cummins, “Learning to forget: Continual prediction with LSTM,” Neural Computation, vol. 12, no. 10, pp. 2451–2471, 2000.
  6. F. A. Gers, N. N. Schraudolph, and J. Schmidhuber, “Learning precise timing with LSTM recurrent networks,” Journal of Machine Learning Research, vol. 3, pp. 115–143, Mar. 2003.
  7. Kashyrina O.Iu. Analiz yakosti roboty systemy heneruvannia muzyky na osnovi LSTM / Kashyrina O.Iu., Havrylenko O.V.. // Materialy Pershoi Vseukrainskoi naukovo-praktychnoi konferentsii molodykh vchenykh ta studentiv «Inzheneriia prohramnoho zabezpechennia i peredovi informatsiini tekhnolohii» (SoftTech-2021). – 2021. – pp. 28–31. (In Ukrainian)
  8. Ilana Shapiro, Mark Huber. Markov Chains for Computer Music Generation. Journal of Humanistic Mathematics. 2021. Vol. 11, no. 2 [Electronic resource] – https://scholarship.claremont.edu/cgi/viewcontent.cgi?article=1848&context=jhm.

https://doi.org/10.33744/2308-6645-2022-3-53-095-103