Academic Journal of Computing & Information Science, 2023, 6(13); doi: 10.25236/AJCIS.2023.061316.
Yandi Luo, Ming Fang
Shanghai Aerospace Electronic Technology Institute, Shanghai, China
This paper focuses on the single-channel speech separation problem and uses deep neural network technology to deeply analyze the loss function, feature extraction and quality evaluation indexes. By improving the existing separation algorithms, a new method combining the jointly constrained loss function and the integrated optimizer is proposed. The study first examines the background and significance of research in the field of speech separation and outlines the current state of research. Then, the basics of speech separation technology and the underlying principles of deep neural networks are introduced in detail. The newly proposed system structure and quality evaluation metrics are used to compare the separation effects of different methods. In addition, the study improves the performance of the algorithm and enhances the ability of the model to avoid local optimums and improve the training efficiency through the proposed joint constrained loss function and integration optimizer.
single-channel speech separation; loss function; integration optimizer; deep neural network
Yandi Luo, Ming Fang. Research on Speech Separation Method Based on Deep Neural Network. Academic Journal of Computing & Information Science (2023), Vol. 6, Issue 13: 107-114. https://doi.org/10.25236/AJCIS.2023.061316.
[1] Feng Qibin. Research on single-channel hybrid speech separation algorithm based on computational auditory scene analysis[D]. Taiyuan University of Technology, 2019.
[2] Zhenlei Li, Fei Yang, Na Li et al. Response characteristics of acoustic emission Meier inverse spectral coefficients for loaded rupture of prefabricated cracked sandstone[J]. Journal of China University of Mining and Technology,2023,52(04):713-726.DOI:10.13247/j.cnki.jcumt.20220667
[3] Qinghua Wang ,Jiyun Sun, Jianhua Hu et al. Prediction of adjustment parameters of slant mill piercing machine based on deep neural network[J/OL]. Forging Technology, 2023,(11):73-78+103 [2023-11-30] https://doi.org/10.13330/j.issn.1000-3940.2023.11.012.
[4] Tian Le, Cao Langcai. DMU improvement algorithm for interactive dynamic influence diagram based on lookahead [J]. Systems Engineering and Electronics, 2014, 36(06):1201-1206.
[5] LIN Yujian, WEI Yunlong, CHEN Qiqi et al. A sigmoid function optimization method for embedded computing platform [J]. Small Microcomputer Systems, 2021, 42(10):2053-2058.
[6] SONG Daiyue, LI Kaizhuang, CHEN Qianqian. Optimization of regularization coefficients for ISAR high-resolution imaging algorithm[J]. Electronic Information Countermeasures Technology, 2023, 38(06): 68-75.
[7] YE Xiaowen, ZHANG Yin'e, ZHOU Qi. Image restoration method for generative adversarial networks based on improved reconstruction loss function [J/OL]. Journal of Gannan Normal University,1-6[2023-11-30]https://doi.org/10.13698/j.cnki.cn36-1346/c.2023.06.018.