intechnguВестник НГУ. Серия: Информационные технологииVestnik NSU. Series: Information Technologies1818-79002410-0420НГУ10.25205/1818-7900-2022-20-3-38-50intechngu-206Research ArticleСтатьиМетод для восстановления аудиосигнала с помощью свёрточных нейронных сетейMethod for Restoring Audio Signal Using Convolutional Neural Networkshttps://orcid.org/0000-0001-8650-3749ДементьеваК. И.DementyevaK. I.

 Дементьева Кристина Игоревна, ассистент кафедры прикладной математики и кибернетики, аспирант

Новосибирск 

 Kristina I. Dementyeva, Assistant of the Department of Applied Mathematics and Cybernetics, Postgraduate studentNovosibirsk 

k.morozova@cyber.sibsutis.ruhttps://orcid.org/0000-0002-1224-4278РакитскийА. А.RakitskyA. A.

 Антон Андреевич Ракитский, кандидат техническиъ наук, доцент кафедры прикладной математики и кибернетики, научный сотрудник,  старший научный сотрудник

Новосибирск 

 Anton A. Rakitskiy, Candidate of Technical Sciences, Associate Professor of the Department of Applied Mathematics and Cybernetics, Researcher, Senior Researcher

Novosibirsk

rakitsky.anton@gmail.comНовосибирский государственный университетРоссияNovosibirsk State UniversityRussian FederationСибирский государственный университет телекоммуникаций и информатики; Новосибирский государственный университетРоссияSiberian State University of Telecommunications and Information Sciences; Novosibirsk State UniversityRussian Federation2022120220232033850Copyright © Дементьева К.И., Ракитский А.А., 20232023Дементьева К.И., Ракитский А.А.Dementyeva K.I., Rakitsky A.A.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://intechngu.elpub.ru/jour/article/view/206

В работе представлено исследование возможности восстановления искажённого аудио-сигнала. На основе полученных ранее результатов использования методов глубокого машинного обучения разработана концепция нейронной сети, предназначенная для коррекции искаженного звукового сигнала. На базе первоначально полученной разработано несколько новых архитектур нейронных сетей, ориентированных на восстановление аудиосигнала. В статье приведены описания разработанных архитектур с теоретическим обоснованием возможности их применения. Представленные архитектуры были протестированы для решения задачи восстановления партии конкретного инструмента в музыкальной композиции, из которой она была удалена.

This paper describes a research of the restoring distorted audio signal possibility. Based on the previously obtained results of using deep machine learning methods, the concept of a neural network to correct a distorted audio signal has been developed. On the basis of the originally obtained results, several new neural network architectures were developed, focused on the audio signal restoring. The paper contains descriptions of the developed architectures with a theoretical substantiation of the possibility of their application. The presented architectures were tested to solve the problem of restoring the part of a specifc instrument in a musical composition where it was removed. The results of testing the developed architectures of neural networks are presented in several forms.

аудиосигналвосстановление сигналаметоды машинного обучениянейронные сетирегрессияaudio signalsignal recoverymachine learning methodsneural networksregressionReferencesMokrý O., Rajmic P. Audio inpainting: Revisited and reweighted //IEEE/ACM Transactions on Audio, Speech, and Language Processing. 2020. Vol. 28. Pp. 2906–2918.Mokrý O., Rajmic P. Audio inpainting: Revisited and reweighted //IEEE/ACM Transactions on Audio, Speech, and Language Processing. 2020. Vol. 28. Pp. 2906–2918.Heo H. S., So B. M., Yang I. H., Yoon S. H., Yu, H. J. Automated recovery of damaged audio fles using deep neural networks //Digital Investigation. 2019. Vol. 30. Pp. 117–126.Heo H. S., So B. M., Yang I. H., Yoon S. H., Yu H. J. Automated recovery of damaged audio fles using deep neural networks // Digital Investigation. 2019. Vol. 30. Pp. 117–126.Kwon B., Gong M., Huh J. and Lee S. Identifcation and Restoration of LZ77 Compressed Data Using a Machine Learning Approach. 2018 Asia-Pacifc Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), Honolulu, HI, USA, 2018, pp. 1787–1790. DOI: 10.23919/APSIPA.2018.8659755Kwon B., Gong M., Huh J., Lee S. Identifcation and Restoration of LZ77 Compressed Data Using a Machine Learning Approach. 2018 Asia-Pacifc Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), Honolulu, HI, USA, 2018, pp. 1787–1790. DOI: 10.23919/APSIPA.2018.8659755Mousavi A., Patel A. B., Baraniuk R. G. A deep learning approach to structured signal recovery. 2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton), Monticello, IL, 2015, pp. 1336–1343. DOI: 10.1109/ALLERTON.2015.7447163Mousavi A., Patel A. B., Baraniuk R. G. A deep learning approach to structured signal recovery. 2015 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton), Monticello, IL, 2015, pp. 1336–1343. DOI: 10.1109/ALLERTON.2015.7447163.Zhang Z., Rao B. D. Sparse Signal Recovery with Temporally Correlated Source Vectors Using Sparse Bayesian Learning. IEEE Journal of Selected Topics in Signal Processing. 2011. Vol. 5, no. 5. Pp. 912–926. DOI: 10.1109/JSTSP.2011.2159773Zhang Z., Rao B. D. Sparse Signal Recovery with Temporally Correlated Source Vectors Using Sparse Bayesian Learning. IEEE Journal of Selected Topics in Signal Processing, 2011. Vol. 5, no. 5, pp. 912–926. DOI: 10.1109/JSTSP.2011.2159773.Bellasi D., Maechler P., Burg A., Felber N., Kaeslin H., Studer C. Live demonstration: Real-time audio restoration using sparse signal recovery. 2013 IEEE International Symposium on Circuits and Systems (ISCAS), Beijing, 2013, pp. 659–659. DOI: 10.1109/ISCAS.2013.6571929.Bellasi D., Maechler P., Burg A., Felber N., Kaeslin H., Studer C. Live demonstration: Realtime audio restoration using sparse signal recovery. 2013 IEEE International Symposium on Circuits and Systems (ISCAS), Beijing, 2013, pp. 659–659. DOI: 10.1109/ISCAS.2013.6571929Adler A. et al. Audio inpainting //IEEE Transactions on Audio, Speech, and Language Processing. 2011. Vol. 20, no. 3. Pp. 922–932.Adler A. et al. Audio inpainting //IEEE Transactions on Audio, Speech, and Language Processing, 2011. Vol. 20, no. 3, pp. 922-932.Doshi S. Various Optimization Algorithms For Training Neural Network. Towards Data Science. 2019 Jan. 13.Doshi S. Various Optimization Algorithms For Training Neural Network. Towards Data Science. 2019 Jan.13.Diederik P. Kingma and Jimmy Lei Ba. Adam : A method for stochastic optimization. 2014. arXiv:1412.6980v9Diederik P. Kingma and Jimmy Lei Ba. Adam: A method for stochastic optimization. 2014. arXiv:1412.6980v9

The authors declare that there are no conflicts of interest present.