Acoustic signal feature extraction method based on manifold learning and its application

Ming Ma ； Zhihong Liu ； Yuanliang Su

《International Core Journal of Engineering》 7卷12期 (2021/12) Pp. 446-454

https://doi.org/10.6919/ICJE.202112_7(12).0061

摘要

A acoustic signal feature extraction method based on manifold learning was proposed and its application was explored. The purpose of this study was to extract accurately the acoustic features of low signal-to-noise ratio noise sources under noisy background, and also to improve the accuracy of sound source recognition. The short-time Fourier transform was used to locally linearize the signal, and singular value decomposition was used to eliminate the noise component, improving the effectiveness of the source signal in the acoustic signal. The adaptive local linear embedding was presented and used to construct the manifold learning high-dimensional matrix, and the adaptive algorithm was proposed to optimize the selection of adjacent points. Finally, the implementation process and simulation results of this method are given, and the method is successfully applied to the sound frequency feature extraction of compressor in oxygen plant under strong background noise. Furthermore, the validity of method has been tested by results of sound source classification feature recognition and the comparison of other manifold methods.

關鍵字

Singular value decomposition ； adaptive local linear embedding ； manifold learning ； acoustic signal feature extraction ； sound source recognition

參考文獻

W Xu, X Lifang, Dan Y , et al. [IEEE 2008 International Conference on Biomedical Engineering And Informatics (BMEI) - Sanya, China (2008.05.27-2008.05.30)] 2008 International Conference on BioMedical Engineering and Informatics - Speech Visualization based on Locally Linear Embedding (LLE) for the Hearing Impaired[J]. 2008:502-505.

Abdel-Galil T K, El-Hag A H, Gaouda A M, et al. Denoising of partial discharge signal using eigendecomposition technique[J]. IEEE Transactions on Dielectrics & Electrical Insulation, 2008, 15(6): 1657-1662.

Qin Yuhua ,Kai Lijun Duan, Baoding Wu, Xu. Similarity measure method based on spectra subspace and locally linear embedding algorithm[J]. Infrared Physics and Technology, 2019,100.

D Y Meng, Z B Xu. Manifold structure reconstruction method based on Isomap[J]. Chinese Journal of Computers, 2010, 33(3):545-555.

Diange Yang, Bing Li, Ziteng Wang, et al. Dynamic wave superposition method for moving sound sources[J]. Acta Physica Sinica, 2012, 61(5): 295–304.