Voiceprint monitoring is of great importance for evaluating the operational status of power equipment such as transformers. However, the labor cost for analyzing voiceprint data is high, and differences in experience levels can influence the stability of the analysis results. To improve computational efficiency and reliability of voiceprint recognition for power equipment, this paper proposes a voiceprint recognition method based on attention autoencoder feature fusion and pulse neural network. This article constructs a voiceprint feature fusion and fault classification identification method based on embedded attention mechanism autoencoder and pulse neural network. Firstly, the voiceprint temporal waveform is converted into various spectral data through wavelet transform, Mel frequency cepstral coefficients, etc. Then, a multi branch convolutional autoencoder is used to perform convolution calculations on the voiceprint frequency domain signal at different scales, achieving multi-level feature fusion enhancement. Finally, a pulse neural network is used to intelligently identify the enhanced voiceprint feature signal, achieving power equipment fault identification based on voiceprint. Compared with other commonly used algorithms, this method has significant advantages in accuracy, recall, and F1 multiple indicators.