Research on Voiceprint Recognition of Electrical Faults With Lower False Alarm Rate

To reduce the false alarm rate of voiceprint recognition of electrical faults under noise interference, a two-stage algorithm was proposed in this paper. Energy storage motors of circuit breakers were taken as the research object. The sound of normal motors and motors with different kinds of abnormalities was captured. After preprocessing like framing and windowing, the Mel frequency cepstral coefficient (MFCC) were extracted as feature vectors. Then, what was different from normal was that we first fed the feature vectors into a novelty detection algorithm, one-class SVM. Only samples that were familiar to one-class support vector machine (SVM) would be delivered to the second algorithm, C-SVM. The second algorithm classified sound signals into different categories. The experiment results show that the two-stage combination can effectively reduce the false alarm rate in noisy environment and the recall rate of faults only slightly decreased when there is little noise. Specifically speaking, it reduced the false alarm rate by 63.47% when the signal-to-noise ratio (SNR) is –20dB and the recall of faults only dropped by 0.33% when SNR is 20dB.