부산대학교 도서관

This article establishes a deep learning recognition model based on prototype networks for voiceprint recognition of hydroelectric units. Firstly, obtain the sound time-frequency signal of the hydroelectric unit through a spectrogram. Subsequently, the Transform will map the collected time-frequency signal into the embedding space. Finally, a prototype network is used to calculate the distance between each sample and all class prototypes in the embedded space, and all samples are recognized and classified based on the size of the distance. This can transform the complex classification process into calculating the distance between samples embedded in the space and the prototype. Compared to other voiceprint recognition methods, the proposed deep learning recognition model has the advantages of high computational efficiency and high recognition accuracy, which can effectively solve the problem of fewer samples for voiceprint faults in hydroelectric units.