Abstract
Deep Frequency Modulation (FM) synthesis is the method of generating approximate or new waveforms by the network inspired by the conventional FM synthesis. The features of the method include that the activation functions of the network are all vibrating ones with distinct parameters and every activation function (oscillator unit) shares an identical time t. The network learns a training waveform given in the temporal interval designated by time t and generates an approximating waveform in the interval. As the first step of the feasibility study, we examine the basic performances and potential of the deep FM synthesis in small-sized experiments. We have confirmed that the optimization techniques developed for the conventional neural networks is applicable to the deep FM synthesis in small-sized experiments.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Chowning, J.M.: The synthesis of complex audio spectra by means of frequency modulation. J. Audio Eng. Soc. 7(21), 526–534 (1973)
Gashler, M.S., Ashmore, S.C.: Training deep fourier neural networks to fit time- series data (2014). arXiv preprint arXiv:1405.2262v1 (2014)
Godfrey, L.B.: Parameterizing and aggregating activation functions in deep neural networks. Ph.D. thesis, University of Arkansas, May 2018
Goodfellow, I., Bengio, Y., Courville, A.: Deep Learning. The MIT Press, Cambridge (2016)
Mingo, L., Aslanyan, L., Castellanos, J., Daz, M., Riazanov, V.: Fourier neural networks: an approach with sinusoidal activation functions. Int. J. Inf. Theories Appl. 11, 52–55 (2004)
Roads, C.: The Computer Music Tutorial. The MIT Press, Cambridge (1996)
Wikipedia: Yamaha DX7. https://en.wikipedia.org/wiki/Yamaha_DX7. Accessed 24 Jan 2019
Acknowledgments
This work has been supported by JSPS Kakenhi 16H01744. The authors would like to thank to Prof. Ichiro Fujinaga of McGill University, Mr. Adrien Ycart of Queen Mary University, and Mr. Masafuji Takahashi of Future University Hakodate for fruitful discussions and valuable suggestions.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2021 Springer Nature Switzerland AG
About this paper
Cite this paper
Hirata, K., Hamanaka, M., Tojo, S. (2021). Feasibility Study of Deep Frequency Modulation Synthesis. In: Kronland-Martinet, R., Ystad, S., Aramaki, M. (eds) Perception, Representations, Image, Sound, Music. CMMR 2019. Lecture Notes in Computer Science(), vol 12631. Springer, Cham. https://doi.org/10.1007/978-3-030-70210-6_15
Download citation
DOI: https://doi.org/10.1007/978-3-030-70210-6_15
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-70209-0
Online ISBN: 978-3-030-70210-6
eBook Packages: Computer ScienceComputer Science (R0)