Research on Condition Monitoring Method of Power Equipment Based on Voiceprint Signal

In this paper, a voiceprint fusion feature generation method based on Mel Frequency Cepstral Coefficients (MFCC) and Gammatone Frequency Cepstral Coefficients (GFCC) is proposed to accurately recognize the operating states of different electric power equipment in high noise environments. First, the voiceprint signals of six power equipment operating states are directionally collected and preprocessed; second, the voiceprint signal features are extracted based on the MFCC and GFCC feature extraction methods; Third, the feature dimensions are reduced according to Principal Component Analysis (PCA) and the downscaled processed features are fused to obtain fusion features; finally, the white Gaussian noises of 5 and 10 dB are added to the voiceprint signal dataset, and the support vector machine (SVM) is used to generate the fusion features. Finally, 5dB and 10dB white Gaussian noise are added to the voiceprint signal dataset, and support vector machine (SVM) is used to accurately recognize the operating state of power equipment to verify the accuracy and robustness of the fused features. The experimental results show that the recognition rate of equipment voiceprints based on fused features is significantly improved compared with the MFCC and GFCC feature extraction methods before and after dimensionality reduction.