The vibration and noise generated during the operation of the transformer are directly related to its operating state and internal defects. The analysis of its voiceprint characteristics is helpful to further understand the operating conditions of the equipment,and ensure the safety and stability of the power system. Based on the analysis of voiceprint features,a transformer defect diagnosis method based on deep neural network and ensemble learning model is proposed. Taking the time-domain and frequency-domain features of transformer voiceprint data as multi-channel input,an integrated learning model is constructed based on a deep neural network model,and the effective recognition of transformer voiceprint features is realized. An ensemble learning model improves the accuracy of transformer defect diagnosis. Based on the transformer voiceprint sample library constructed in this paper,the recognition accuracy rate of the method for single transformer defects is 99.2%,and the recognition accuracy rate for transformer mixed defects is 99.7%. The research results show that the method can effectively identify the operating state of the transformer,and can provide technical reference for the operation and maintenance of the transformer.