ntvНаучно-технический вестник информационных технологий, механики и оптикиScientific and Technical Journal of Information Technologies, Mechanics and Optics2226-14942500-0373Университет ИТМО10.17586/2226-1494-2024-24-5-758-769ntv-142Research ArticleКОМПЬЮТЕРНЫЕ СИСТЕМЫ, ИНФОРМАЦИОННЫЕCOMPUTER SCIENCEСовместное распознавание акустических сцен и аудиособытий с помощью многозадачного обучения компактных моделейLow-complexity multi task learning for joint acoustic scenes classification and sound events detectionhttps://orcid.org/0000-0002-3929-7484СурковМ. К.SurkovM. K.

Сурков Максим Константинович - аспирант

Санкт-Петербург, 197101

Maxim K. Surkov - PhD Student

Saint Petersburg, 197101

surkovmax007@mail.ruУниверситет ИТМОРоссияITMO UniversityRussian Federation202413122024245758769Copyright © Сурков М.К., 20242024Сурков М.К.Surkov M.K.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://ntv.elpub.ru/jour/article/view/142

Введение. Задача распознавания метаинформации заключается в выявлении и извлечении данных различной природы (речь, шумы, акустическая сцена, акустические события, аномальные звуки) из входного аудиосигнала. Существуют подходы, способные обеспечить высокую точность распознавания метаинформации различной природы в аудиозаписях. Данные модели часто опираются на глубокие нейронные сети с числом обучаемых параметров более сотни миллионов. Как следствие, такие модели невозможно использовать в реальных коммерческих системах, так как они ограничены в вычислительных ресурсах. Это влияет на работу умных устройств, таких как мобильные телефоны, умные часы, колонки, системы «умный дом». Обычно к умным устройствам предъявляются серьезные требования по энергоэффективности, что влияет на применение тех или иных компонентов в составе таких продуктов. Тактовые частоты процессоров, объемы оперативной и дисковой памяти в таких устройствах сильно ограничены и не способны работать с нейросетевыми моделями с большим числом обучаемых параметров. Подобные ограничения требуют поиска возможных решений, которые бы позволили применять технологии распознавания метаинформации в коммерческих устройствах. Возможным решением могут стать так называемые компактные нейросетевые модели, которые за счет архитектуры и многозадачных алгоритмов обучения способны распознавать метаинформацию в аудиозаписях и используют ограниченное число обучаемых параметров. Коммерческий интерес к данной задаче согласуется и с заинтересованностью научного сообщества. Так, в рамках международного конкурса под названием «Detection and Classification of Acoustic Scenes and Events» организаторами были сформулированы специальные подзадачи — распознавание акустической сцены при использовании низкоресурсных систем («Low- Complexity Acoustic Scene Classification») и детекции аудиособытий («Sound Event Detection with Weak Labels and Synthetic Soundscapes»). Важными исследовательскими вопросами являются как создание оптимальной архитектуры компактной нейронной сети, так и алгоритмов их обучения для получения низкоресурсной высокоточной системы распознавания акустических сцен и аудиособытий.Метод. Исследование выполнено на основе корпуса данных задач Challenge «Low-Complexity Acoustic Scene Classification» и «Sound Event Detection with Weak Labels and Synthetic Soundscapes». Предложена архитектура многозадачной нейронной сети, состоящая из общего кодировщика и двух независимых декодировщиков для каждой из двух задач. Рассмотрены классические алгоритмы многозадачного обучения SoftMTL и HardMTL, а также разработаны их модификации CrossMTL, которые опираются на идею переиспользования данных от одной задачи при обучении декодировщика решать вторую задачу, и FreezeMTL, в процессе которого обученные веса общего кодировшика замораживаются после обучения на первой задаче и используются для оптимизации второго декодировщика.Основные результаты. Показано, что применение модификации CrossMTL дает возможность существенно увеличить точность классификации акустических сцен и детекции аудиособытий по сравнению с классическими подходами SoftMTL и HardMTL. Алгоритм FreezeMTL позволяет получить модель, демонстрирующую точность классификации сцен в 42,44 % и детекции событий в 45,86 %, что сравнимо с показателями базовых решений задач 2023 года.Обсуждение. Предложена компактная нейронная сеть, состоящая из 633,5 тыс. обучаемых параметров, требующая 43,2 млн арифметических операций для обработки аудио длиной в одну секунду. Модель использует на 7,8 % меньше обучаемых параметров и на 40 % меньше арифметических операций по сравнению с наивным применением двух независимых моделей. Разработанную модель можно применить в умных устройствах за счет уменьшения числа обучаемых параметров и арифметических операций, необходимых для ее применения.

The task of automatic metainformation recognition from audio sources is to detect and extract data of various natures (speech, noises, acoustic scenes, acoustic events, anomalies) from a given audio input signal. This area is well developed and known to the scientific community and has various approaches with high quality. But, the vast majority of such methods are based on large neural networks with a huge number of weights to be trained. Subsequently, it is impractical to use them in environments with severely limited computing resources. The smart device industry is currently growing rapidly: smartphones, smart watches, voice assistants, TV, smart home. Such products have limitations in both processor and memory. At that moment, the State-of-the-Art way to cope with these conditions is to use so-called low-complexity models. Moreover, in recent years, the interest of the scientific community in the above-mentioned problem has been growing (DCASE Workshop). One of the most crucial subtasks in the global meta information recognition problem is the task of Automatic Scene Classification and the task of Sound Event Detection. The most important scientific questions are the development of both the optimal low-complexity neural network architecture and learning algorithms to obtain a low-resource, high-quality system for classifying acoustic scenes and detecting sound events. In this paper the datasets from DCASE Challenge “Low-Complexity Acoustic Scene Classification” and “Sound Event Detection with Weak Labels and Synthetic Soundscapes” were used. A multitask neural network architecture was proposed consisting of a common encoder and two independent decoders for each of the two tasks. The classical algorithms of multitask learning SoftMTL and HardMTL were considered, and their modifications were developed: CrossMTL, which is based on the idea of reusing data from one task when training the decoder to solve the second task, and FreezeMTL, in which the trained weights of the common encoder are frozen after training on the first task and used to optimize the second decoder. As a result of the experiments, it was shown that the use of the CrossMTL modification can significantly increase the accuracy of the classification of acoustic scenes and event detection in compare with classical approaches SoftMTL and HardMTL. The FreezeMTL algorithm made it possible to obtain a model that provides 42.44 % accuracy in scene classification and 45.86 % accuracy in event detection, which is comparable to the results of the baseline solutions of 2023. In this paper, a low-complexity neural network consisting of 633.5 K trainable parameters was proposed, requiring 43.2 M MACs to process one second audio. This approach uses 7.8 % fewer trainable parameters and 40 % fewer MACs compared to the naive application of two independent models. The developed model can be used in smart devices due to a small number of trainable parameters, as well as a small number of MACs required for its application.

распознавание акустической сценыдетекция аудиособытийкомпактные моделимногозадачные нейронные сетимногозадачное обучениераспознавание метаинформацииумные устройстванейронные сетиacoustic scene classificationsound event detectioncompact modelsmultitask neural networksmultitask learningmeta-information recognitionsmart devicesneural networksReferencesKriman S., Beliaev S., Ginsburg B., Huang J., Kuchaiev O., Lavrukhin V., Leary R., Li J., Zhang Y. Quartznet: Deep automatic speech recognition with 1D time-channel separable convolutions // Proc. of the ICASSP 2020 — 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). 2020. P. 6124–6128. https://doi.org/10.1109/icassp40776.2020.9053889Kriman S., Beliaev S., Ginsburg B., Huang J., Kuchaiev O., Lavrukhin V., Leary R., Li J., Zhang Y. Quartznet: Deep automatic speech recognition with 1D time-channel separable convolutions. Proc. of the ICASSP 2020 — 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2020, pp. 6124–6128. https://doi.org/10.1109/icassp40776.2020.9053889Lakhotia K., Kharitonov E., Hsu W.-N., Adi Y., Polyak A., Bolte B., Nguyen T.-A., Copet J., Baevski A., Mohamed A., Dupoux E. On generative spoken language modeling from raw audio // Transactions of the Association for Computational Linguistics. 2021. V. 9. P. 1336–1354.Lakhotia K., Kharitonov E., Hsu W.-N., Adi Y., Polyak A., Bolte B., Nguyen T.-A., Copet J., Baevski A., Mohamed A., Dupoux E. On generative spoken language modeling from raw audio. Transactions of the Association for Computational Linguistics, 2021, vol. 9, pp. 1336–1354.Gulati A., Qin J., Chiu C.-C., Parmar N., Zhang Y., Yu J., Han W., Wang S., Zhang Z., Wu Y., Pang R. Conformer: Convolutionaugmented transformer for speech recognition // Proc. of the Annual Conference of the International Speech Communication Association, INTERSPEECH. 2020. P. 5036–5040. https://doi.org/10.21437/interspeech.2020-3015Gulati A., Qin J., Chiu C.-C., Parmar N., Zhang Y., Yu J., Han W., Wang S., Zhang Z., Wu Y., Pang R. Conformer: Convolutionaugmented transformer for speech recognition. Proc. of the Annual Conference of the International Speech Communication Association, INTERSPEECH, 2020, pp. 5036–5040. https://doi.org/10.21437/interspeech.2020-3015Hsu W.N., Tsai B., Bolte Y.-H.H., Salakhutdinov R., Mohamed A. HuBERT: How much can a bad teacher benefit ASR pre-training? // Proc. of the ICASSP 2021 — 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). 2021. P. 6533–6537. https://doi.org/10.1109/icassp39728.2021.9414460Hsu W.N., Tsai B., Bolte Y.-H.H., Salakhutdinov R., Mohamed A. HuBERT: How much can a bad teacher benefit ASR pre-training?. Proc. of the ICASSP 2021 — 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021, pp. 6533–6537. https://doi.org/10.1109/icassp39728.2021.9414460Radford A., Kim J.W., Xu T., Brockman G., McLeavey C., Sutskever I. Robust speech recognition via large-scale weak supervision // Proceedings of Machine Learning Research, PMLR. 2023. V. 202. P. 28492–28518.Radford A., Kim J.W., Xu T., Brockman G., McLeavey C., Sutskever I. Robust speech recognition via large-scale weak supervision. Proceedings of Machine Learning Research, 2023, vol. 202, pp. 28492–28518.Gong Y., Khurana S., Karlinsky L., Glass J. Whisper-at: Noise-robust automatic speech recognizers are also strong general audio event taggers // Proc. of the Annual Conference of the International Speech Communication Association, INTERSPEECH. 2023. P. 2798–2802. https://doi.org/10.21437/interspeech.2023-2193Gong Y., Khurana S., Karlinsky L., Glass J. Whisper-at: Noise-robust automatic speech recognizers are also strong general audio event taggers. Proc. of the Annual Conference of the International Speech Communication Association, INTERSPEECH, 2023, pp. 2798–2802. https://doi.org/10.21437/interspeech.2023-2193Panayotov V., Chen G., Povey D., Khudanpur S. Librispeech: an asr corpus based on public domain audio books // Proc. of the IEEE International Conference On Acoustics, Speech and Signal Processing (ICASSP). 2015. P. 5206–5210. https://doi.org/10.1109/icassp.2015.7178964Panayotov V., Chen G., Povey D., Khudanpur S. Librispeech: an asr corpus based on public domain audio books. Proc. of the IEEE International Conference On Acoustics, Speech and Signal Processing (ICASSP), 2015, pp. 5206–5210. https://doi.org/10.1109/icassp.2015.7178964Moritz N., Wichern G., Hori T., Le Roux J. All-in-One transformer: Unifying speech recognition, audio tagging, and event detection // Proc. of the Annual Conference of the International Speech Communication Association, INTERSPEECH. 2020. P. 3112–3116. https://doi.org/10.21437/interspeech.2020-2757Moritz N., Wichern G., Hori T., Le Roux L. All-in-One transformer: Unifying speech recognition, audio tagging, and event detection. Proc. of the Annual Conference of the International Speech Communication Association, INTERSPEECH, 2020, pp. 3112–3116. https://doi.org/10.21437/interspeech.2020-2757Karita S., Soplin N.E.Y., Watanabe S., Delcroix M., Ogawa A., Nakatani T. Improving transformer-based end-to-end speech recognition with connectionist temporal classification and language model integration // Proc. of the Annual Conference of the International Speech Communication Association, INTERSPEECH. 2019. P. 1408–1412. https://doi.org/10.21437/interspeech.2019-1938Karita S., Soplin N.E.Y., Watanabe S., Delcroix M., Ogawa A., Nakatani T. Improving transformer-based end-to-end speech recognition with connectionist temporal classification and language model integration. Proc. of the Annual Conference of the International Speech Communication Association, INTERSPEECH, 2019, pp. 1408–1412. https://doi.org/10.21437/interspeech.2019-1938Chen S., Wu Y., Wang C., Liu S., Tompkins D., Chen Z., Che W., Yu X., Wei F. Beats: Audio pre-training with acoustic tokenizers // Proceedings of Machine Learning Research. 2023. V. 202, P. 4672–4712.Chen S., Wu Y., Wang C., Liu S., Tompkins D., Chen Z., Che W., Yu X., Wei F. Beats: Audio pre-training with acoustic tokenizers. Proceedings of Machine Learning Research, 2023, vol. 202, pp. 4672–4712.Gemmeke J.F., Ellis D.P.W., Freedman D., Jansen A., Lawrence W., Moore R.C., Plakal M., Ritter M. Audio set: An ontology and humanlabeled dataset for audio events // Proc. of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). 2017. P. 776–780. https://doi.org/10.1109/icassp.2017.7952261Gemmeke J.F., Ellis D.P.W., Freedman D., Jansen A., Lawrence W., Moore R.C., Plakal M., Ritter M. Audio set: An ontology and humanlabeled dataset for audio events. Proc. of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2017, pp. 776–780. https://doi.org/10.1109/icassp.2017.7952261Dosovitskiy A., Beyer L., Kolesnikov A., Weissenborn D., Zhai X., Unterthiner T., Dehghani M., Minderer M., Heigold G., Gelly S., Uszkoreit J., Houlsby N. An image is worth 16x16 words: Transformers for image recognition at scale // Proc. of the ICLR 2021 — 9th International Conference on Learning Representations. 2021.Dosovitskiy A., Beyer L., Kolesnikov A., Weissenborn D., Zhai X., Unterthiner T., Dehghani M., Minderer M., Heigold G., Gelly S., Uszkoreit J., Houlsby N. et al. An image is worth 16×16 words: Transformers for image recognition at scale. Proc. of the ICLR 2021 — 9th International Conference on Learning Representations, 2021.Drossos K., Lipping S., Virtanen T. Clotho: An audio captioning dataset // Proc. of the ICASSP 2020 — 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). 2020. P. 736–740. https://doi.org/10.1109/icassp40776.2020.9052990Drossos K., Lipping S., Virtanen T. Clotho: An audio captioning dataset. Proc. of the ICASSP 2020 — 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2020, pp. 736–740. https://doi.org/10.1109/icassp40776.2020.9052990Poria S., Hazarika D., Majumder N., Naik G., Cambria E., Mihalcea R. MELD: A multimodal multi-party dataset for emotion recognition in conversations // Proc. of the 57th Annual Meeting of the Association for Computational Linguistics. 2019. P. 527–536. https://doi.org/10.18653/v1/p19-1050Poria S., Hazarika D., Majumder N., Naik G., Cambria E., Mihalcea R. MELD: A multimodal multi-party dataset for emotion recognition in conversations. Proc. of the 57th Annual Meeting of the Association for Computational Linguistics, 2019, pp. 527–536. https://doi.org/10.18653/v1/p19-1050Lipping S., Sudarsanam P., Drossos K., Virtanen T. Clotho-AQA: A crowdsourced dataset for audio question answering // Proc. of the 30th European Signal Processing Conference (EUSIPCO). 2022. P. 1140–1144. https://doi.org/10.23919/eusipco55093.2022.9909680Lipping S., Sudarsanam P., Drossos K., Virtanen T. Clotho-AQA: A crowdsourced dataset for audio question answering. Proc. of the 30th European Signal Processing Conference (EUSIPCO), 2022, pp. 1140–1144. https://doi.org/10.23919/eusipco55093.2022.9909680Engel J., Resnick C., Roberts A., Dieleman S., Norouzi M., Eck D., Simonyan K. Neural audio synthesis of musical notes with wavenet autoencoders // Proceedings Conference on Machine Learning, PMLR. 2017. V. 70. P. 1068–1077.Engel J., Resnick C., Roberts A., Dieleman S., Norouzi M., Eck D., Simonyan K. Neural audio synthesis of musical notes with wavenet autoencoders. International Conference on Machine Learning. PMLR, 2017, vol. 70, pp. 1068–1077.Chu Y., Xu J., Zhou X., Yang Q., Zhang S., Yan Z., Zhou C., Zhou J. Qwen-audio: Advancing universal audio understanding via unified large-scale audio-language models // arXiv. 2023. arXiv:2311.07919. https://doi.org/10.48550/arXiv.2311.07919Chu Y., Xu J., Zhou X., Yang Q., Zhang S., Yan Z., Zhou C., Zhou J. Qwen-audio: Advancing universal audio understanding via unified large-scale audio-language models. arXiv, 2023, arXiv:2311.07919. https://doi.org/10.48550/arXiv.2311.07919Bai J., Bai S., Chu Y., Cui Z. Qwen technical report // arXiv. 2023. arXiv:2309.16609. https://doi.org/10.48550/arXiv.2309.16609Bai J., Bai S., Chu Y., Cui Z. Qwen technical report. arXiv, 2023, arXiv:2309.16609. https://doi.org/10.48550/arXiv.2309.16609Schmid F., Morocutti T., Masoudian S., Koutini K., Widmer G. CPJKU submission to dcase23: Efficient acoustic scene classification with cp-mobile: Technical Report / Detection and Classification of Acoustic Scenes and Events (DCASE). 2023. 5 p.Schmid F., Morocutti T., Masoudian S., Koutini K., Widmer G. CPJKU submission to dcase23: Efficient acoustic scene classification with cp-mobile: Technical Report. Detection and Classification of Acoustic Scenes and Events (DCASE), 2023. 5 p.Salamon J., MacConnell D., Cartwright M., Li P., Bello J.P. Scaper: A library for soundscape synthesis and augmentation // Proc. of the IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA). 2017. P. 344–348. https://doi.org/10.1109/waspaa.2017.8170052Salamon J., MacConnell D., Cartwright M., Li P., Bello J.P. Scaper: A library for soundscape synthesis and augmentation. Proc. of the IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA), 2017, pp. 344–348. https://doi.org/10.1109/waspaa.2017.8170052Tarvainen A., Valpola H. Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results // Advances in Neural Information Processing Systems. 2017. V. 30. P. 1196–1205.Tarvainen A., Valpola H. Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. Advances in Neural Information Processing Systems, 2017, vol. 30, pp. 1196–1205.Zhang Z., Luo P., Loy C.C., Tang X. Facial landmark detection by deep multi-task learning // Lecture Notes in Computer Science. 2014. V. 8694. P. 94–108. https://doi.org/10.1007/978-3-319-10599-4_7Zhang Z., Luo P., Loy C.C., Tang X. Facial landmark detection by deep multi-task learning. Lecture Notes in Computer Science, 2014, vol. 8694, pp. 94–108. https://doi.org/10.1007/978-3-319-10599-4_7Dai J., He K., Sun J. Instance-aware semantic segmentation via multitask network cascades // Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2016. P. 3150–3158. https:// doi.org/10.1109/cvpr.2016.343Dai J., He K., Sun J. Instance-aware semantic segmentation via multitask network cascades. Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 3150–3158. https://doi.org/10.1109/cvpr.2016.343Zhao X., Li H., Shen X., Liang X., Wu Y. A modulation module for multi-task learning with applications in image retrieval // Lecture Notes in Computer Science. 2018. V. 11205. P. 415–432. https://doi.org/10.1007/978-3-030-01246-5_25Zhao X., Li H., Shen X., Liang X., Wu Y. A modulation module for multi-task learning with applications in image retrieval. Lecture Notes in Computer Science, 2018, vol. 11205, pp. 415–432. https://doi.org/10.1007/978-3-030-01246-5_25Liu S., Johns E., Davison A.J. End-to-end multi-task learning with attention // Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2019. P. 1871–1880. https://doi.org/10.1109/cvpr.2019.00197Liu S., Johns E., Davison A.J. End-to-end multi-task learning with attention. Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 1871–1880. https://doi.org/10.1109/cvpr.2019.00197Ma J., Zhao Z., Yi X., Chen J., Hong L., Chi E.H. Modeling task relationships in multi-task learning with multi-gate mixture-ofexperts // Proc. of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2018. P. 1930–1939. https://doi.org/10.1145/3219819.3220007Ma J., Zhao Z., Yi X., Chen J., Hong L., Chi E.H. Modeling task relationships in multi-task learning with multi-gate mixture-ofexperts. Proc. of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2018, pp. 1930–1939. https:// doi.org/10.1145/3219819.3220007Misra I., Shrivastava A., Gupta A., Hebert M. Cross-stitch networks for multi-task learning // Proc. of the IEEE Conference on Computer Vision and Pattern Recognition. 2016. P. 3994–4003. https://doi.org/10.1109/cvpr.2016.433Misra I., Shrivastava A., Gupta A., Hebert M. Cross-stitch networks for multi-task learning. Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 3994–4003. https://doi.org/10.1109/cvpr.2016.433Ruder S., Bingel J., Augenstein I., Søgaard A. Latent multi-task architecture learning // Proceedings of the AAAI Conference on Artificial Intelligence. 2019. V. 33. N 01. P. 4822–4829. https://doi.org/10.1609/aaai.v33i01.33014822Ruder S., Bingel J., Augenstein I., Søgaard A. Latent multi-task architecture learning. Proceedings of the AAAI Conference on Artificial Intelligence, 2019, vol. 33, no. 01, pp. 4822–4829. https:// doi.org/10.1609/aaai.v33i01.33014822Krause D.A., Mesaros A. Binaural signal representations for joint sound event detection and acoustic scene classification // Proc. of the 30th European Signal Processing Conference (EUSIPCO). 2022. P. 399–403. https://doi.org/10.23919/eusipco55093.2022.9909581Krause D.A., Mesaros A. Binaural signal representations for joint sound event detection and acoustic scene classification. Proc. of the 30th European Signal Processing Conference (EUSIPCO), 2022, pp. 399–403. https://doi.org/10.23919/eusipco55093.2022.9909581Khandelwal T., Das R.K. A multi-task learning framework for sound event detection using high-level acoustic characteristics of sounds // Proc. of the Annual Conference of the International Speech Communication Association, INTERSPEECH. 2023. P. 1214–1218. https://doi.org/10.21437/interspeech.2023-909Khandelwal T., Das R.K. A multi-task learning framework for sound event detection using high-level acoustic characteristics of sounds. Proc. of the Annual Conference of the International Speech Communication Association, INTERSPEECH, 2023, pp. 1214–1218. https://doi.org/10.21437/interspeech.2023-909French R.M. Catastrophic forgetting in connectionist networks // Trends in Cognitive Sciences. 1999. V. 3. N 4. P. 128–135. https://doi.org/10.1016/s1364-6613(99)01294-2French R.M. Catastrophic forgetting in connectionist networks. Trends in Cognitive Sciences, 1999, vol. 3, no. 4, pp. 128–135. https:// doi.org/10.1016/s1364-6613(99)01294-2McCloskey M., Cohen N.J. Catastrophic interference in connectionist networks: The sequential learning problem // Psychology of Learning and Motivation. 1989. V. 24. P. 109–165. https://doi.org/10.1016/s0079-7421(08)60536-8McCloskey M., Cohen N.J. Catastrophic interference in connectionist networks: The sequential learning problem. Psychology of Learning and Motivation, 1989, vol. 24, pp. 109–165. https://doi.org/10.1016/s0079-7421(08)60536-8Kirkpatrick J., Pascanu R., Rabinowitz N., Veness J., Desjardins G., Rusu A.A., Milan K., Quan J., Ramalho T., Grabska-Barwinska A., Hassabis D., Clopath C., Kumaran D., Hadsell R. Overcoming catastrophic forgetting in neural networks // Proceedings of the national academy of sciences. 2017. V. 114. N 13. P. 3521–3526. https://doi.org/10.1073/pnas.1611835114Kirkpatrick J., Pascanu R., Rabinowitz N., Veness J., Desjardins G., Rusu A.A., Milan K., Quan J., Ramalho T., Grabska-Barwinska A., Hassabis D., Clopath C., Kumaran D., Hadsell R. Overcoming catastrophic forgetting in neural networks. Proceedings of the national academy of sciences, 2017, vol. 114, no. 13, pp. 3521–3526. https://doi.org/10.1073/pnas.1611835114Kim J.W., Lee G.W., Kim H.K., Seo Y.S., Song I.H. Semi-supervised learning-based sound event detection using frequency-channel-wise selective kernel for DCASE challenge 2022 Task 4: Technical Report / Detection and Classification of Acoustic Scenes and Events (DCASE), 2022. 4 p.Kim J.W., Lee G.W., Kim H.K., Seo Y.S., Song I.H. Semi-supervised learning-based sound event detection using frequency-channel-wise selective kernel for DCASE challenge 2022 Task 4: Technical Report. Detection and Classification of Acoustic Scenes and Events (DCASE), 2022, 4 p.

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