Feature extraction and classification are the main body of acoustic target recognition. Firstly, amplitude extraction based on harmonic set, MFCC and wavelet packet decomposition are selected as feature selection methods to extract the features of four kinds of measured sound signals respectively. Then, the feature vectors extracted by MFCC method with the best clustering effect are combined with GMM classifier, BP neural network, OVO-SVMs and multi-layer support vector machine as classifiers, and MATLAB is used for simulation recognition and classification. Finally, the recognition results are compared and analyzed. The results show that BP neural network has the best recognition effect, but its robustness is poor, the experimental results are unstable, and the selection of training and test data has a great impact on the recognition rate. The multi classification algorithm based on support vector machine has a better stable recognition rate.
Acoustic Target Recognition Based on MFCC and SVM
Lect. Notes Electrical Eng.
International Conference on Man-Machine-Environment System Engineering ; 2022 ; Beijing, China October 21, 2022 - October 23, 2022
2022-08-21
6 pages
Article/Chapter (Book)
Electronic Resource
English
Acoustic target recognition , MFCC , Support vector machine , BP neural network Engineering , Manufacturing, Machines, Tools, Processes , Engineering Economics, Organization, Logistics, Marketing , Industrial and Organizational Psychology , Artificial Intelligence , Aerospace Technology and Astronautics
MFCC-CNN-based mine hoist fault acoustic analysis method
European Patent Office | 2021
|Speech Emotion Recognition Based on 1D CNN and MFCC
IEEE | 2023
|European Patent Office | 2019
|