Koshel Y. Models and methods of speech synthesis based on neural systems of ordinary differential equations

Українська версія

Thesis for the degree of Doctor of Philosophy (PhD)

State registration number

0825U001338

Thesis Registration Form

0825U001338.pdf

Applicant for

Specialization

  • 113 - Прикладна математика

Specialized Academic Board

PhD 8597

Oles Honchar Dnipro National University

Essay

The dissertation is devoted to the study of modern methods of speech synthesis based on the latest approaches to building neural network models. First of all, a new class of neural network models is considered - neural ordinary differential equations, which allows modeling dynamic processes in the form of systems of differential equations, the right-hand sides of which are defined as a neural network. The main advantage of this class of models in the context of the chosen subject area is that it allows using the well-known tools to study the qualitative characteristics of the constructed models. During the work on the dissertation, a new model was developed based on neural ordinary differential equations, which allows approximating dynamic processes given by one-dimensional time series, as well as adapting to qualitative changes in these processes to preserve the quality of the approximation. Based on the new model, a system for speech synthesis from its distinctive characteristics is built. For this purpose, a new software product has been implemented that calculates estimates of the fundamental frequency, spectral characteristics, and loudness for a streaming signal.

Read more

Research papers

1. Belozyorov V. Ye., Koshеl Ye. V., Zaytsev V.G. The possibility of the robust chaos emergence in the Lorenz-like nonautonomous system - Journal of Optimization, Differential Equations and Their Applications. DOI 10.15421/141902, ISSN (print) 2617–0108, ISSN (on-line) 2663–6824. - V. 27. - №2.- December 2019. - P. 44-59. — Режим доступу: https://model-dnu.dp.ua/index.php/SM/article/download/138/121

2. Belozyorov V. Ye., Dantsev D. V., Koshel Y. V. On the Equivalence of Real Dynamic Process and Its Models. - Journal of Optimization, Differential Equations and Their Applications. DOI 10.15421/142109, ISSN (print) 2617-0108, ISSN (on-line) 2663–6824. - V. 29. - №2.- December 2021. - P. 76-91. - Режим доступу: https://model-dnu.dp.ua/index.php/SM/article/download/162/144

3. Belozyorov V. Ye., Koshel Y. V. Univariate time series analysis with hyperneural ODE. - Journal of Optimization, Differential Equations and Their Applications. DOI 10.15421/142311, ISSN (print) 2617-0108, ISSN (on-line) 2663-6824. - V. 31. - №2.- December 2023. - P. 50-66. - Режим доступу: https://model-dnu.dp.ua/index.php/SM/article/download/187/168

4. Belozyorov V. Ye., Koshel Y. V. On systems of neural ODEs with generalized power activation functions. - Journal of Optimization, Differential Equations and Their Applications. DOI 10.15421/141901, ISSN (print) 2617–0108, ISSN (on-line) 2663–6824. - V. 32. - №2.- December 2024. - P. 56-91. - Режим доступу: https://model-dnu.dp.ua/index.php/SM/article/download/201/182

5. Koshel Y. V. Adaptation of the world framework for frame-by-frame real-time speech analysis. - System technologies. DOI10.34185/1562-9945-5-148-2023-03, ISSN (print) 1562-9945, ISSN (on-line) 2707-7977. - V. 5. - № 148. - 2023. - P. 21-36. - Режим доступу: https://journals.nmetau.edu.ua/index.php/st/article/view/1543/848

6. Koshel Y. V. Neural network-assisted continuous embedding of univariate data streams for time series analysis. - System technologies. DOI 10.34185/1562-9945-2-151-2024-08, ISSN (print) 1562-9945, ISSN (on-line) 2707-7977. - V. 2. - № 151. - 2024. - P. 92-101. - Режим доступу: https://journals.nmetau.edu.ua/index.php/st/article/view/1712/1012

7. Koshel E. Local Jacobian estimation for delay embedded time series data. Тези доповіді на 18-й міжнародній науково-практичній конференції “Математичне та програмне забезпечення інтелектуальних систем” (MPZIS-2020). - 18-20 листопада 2020 р. Дніпро, с. 152-153. - Режим доступу: http://mpzis.dnu.dp.ua/wp-content/uploads/2020/12/%D0%9C%D0%9F%D0%97%D0%86%D0%A1-20201.pdf

7. Koshel E. Local Jacobian estimation for delay embedded time series data. Тези доповіді на 18-й міжнародній науково-практичній конференції “Математичне та програмне забезпечення інтелектуальних систем” (MPZIS-2020). - 18-20 листопада 2020 р. Дніпро, с. 152-153. - Режим доступу: http://mpzis.dnu.dp.ua/wp-content/uploads/2020/12/%D0%9C%D0%9F%D0%97%D0%86%D0%A1-20201.pdf

8. Koshel Е. Lyapunov spectrum estimation for delay embedded time series. - Тези доповіді на 19-й міжнародній науково-практичній конференції “Математичне та програмне забезпечення інтелектуальих систем” (MPZIS-2021). - 17-19 листопада 2021 р. Дніпро, с. 110-111. - Режим доступу: https://www.dnu.dp.ua/docs/ndc/2021/28_%D0%9C%D0%9F%D0%97%D0%86%D0%A1-2021.pdf#page=110

9. Koshel Y. V., Belozyorov V. Ye. Modelling Dynamical Processes Described By Time Series Data With Neural Differential Equations. - Тези доповіді на 20-й міжнародній науково-практичній конференції “Математичне та програмне забезпечення інтелектуальих систем” (MPZIS-2022). - 23-25 листопада 2022 р. Дніпро, с. 111-112. — Режим доступу: http://mpzis.dnu.dp.ua/wp-content/uploads/2022/12/MPZIS-2022-1.pdf

10. Koshel Eu. Hyper neural ODE for univariate time series analysis, - Тези доповіді на 21-й міжнародній науково-практичній конференції “Математичне та програмне забезпечення інтелектуальних систем” (MPZIS-2023). - 22-24 листопада 2023, Дніпро, с. 21-22. - Режим доступу: http://mpzis.dnu.dp.ua/wp-content/uploads/2023/11/mpzis-2023.pdf

11. Koshel Y. V, Belozyorov V. Ye. Generation of stable systems of ordinary differential equations for univariate time series modeling. - Тези доповіді на ХХVII МІЖНАРОДНІЙ КОНФЕРЕНЦІЇ З АВТОМАТИЧНОГО КЕРУВАННЯ «АВТОМАТИКА 2024» - 20-22 листопада 2024 р., Дніпро, C. 36 - 37. - Режим доступу: https://automatika2024.dp.ua/files/%D0%90%D0%B2%D1%82%D0%BE%D0%BC%D0%B0%D1%82%D0%B8%D0%BA%D0%B0-2024%20(%D1%82%D0%B5%D0%B7%D0%B8%20%D0%B4%D0%BE%D0%BF%D0%BE%D0%B2%D1%96%D0%B4%D0%B5%D0%B9).pdf

12. Belozyorov V. Ye, Koshel Y. V. On use of power activation functions with a degree less than 1 in neural networks. - Тези доповіді на 22-й міжнародній науково-практичній конференції «Математичне та програмне забезпечення інтелектуальних систем» (MPZIS-2024). - 20-22 листопада 2024 р., Дніпро, C. 7 - 9. - Режим доступу: http://mpzis.dnu.dp.ua/wp-content/uploads/2024/11/%D0%9C%D0%9F%D0%97%D0%86%D0%A1-2024-1.pdf

13. Koshel Eu. Improvement for peaks and zero-crossings detection for time-domain pitch estimation algorithms - Proceedings of 1st international scientific and practical conference «CURRENT CHALLENGES OF SCIENCE AND EDUCATION», - September 18-20 2023, Berlin, - P. 116-118. - Режим доступу: https://sci-conf.com.ua/wp-content/uploads/2023/09/CURRENT-CHALLENGES-OF-SCIENCE-AND-EDUCATION-18-20.09.23.pdf

Read more

Files

autoreferat-Анотації.pdf

Дисертація Кошель Є-В-.pdf

Similar theses

0825U001354

Serhii Malyshev

Bifurcations and stability of nonlinear vibrations of beam structures with fatigue cracks

0825U001318

Oksana Melashchenko

Mathematical model and solution method for solving the optimization layout problem of geometric objects considering stretch and shrink transformations.

0825U001311

Valeriia Karieva

The Problem of Finding an Upper Optimality Estimate for Liver Regeneration Strategies Using Adaptive Dynamic Programming Methods

0825U000766

Oleksandr V. Mykhail

On deformation peculiarities of piezo passive and piezo active composite bodies with cracks and corner points.

0825U000723

Anatolii A. Dmytruk

Mathematical modeling of drying processes in multicomponent capillary-porous bodies considering phase changes