Abstract: Abnormal sounds will be generated when partial discharge faults occur in switchgear, and voiceprint recognition technology can realize live detection of switchgear. This paper proposes a voiceprint recognition method for switchgear partial discharge based on MFCC and MobileNetv3-ECA network. Firstly, a bone conduction sensor is adopted to collect voiceprint data of switchgear during operation. The data is framed and windowed to calculate MFCC coefficients, which are further converted into two-dimensional RGB images. Secondly, taking the MobileNetv3 network as the backbone, the ECA attention mechanism is used to replace the original SE module, so as to construct the MobileNetv3-ECA network for feature extraction and classification of MFCC voiceprint images, thereby realizing the identification of switchgear partial discharge faults. Classification experiments are carried out on five types of voiceprint data, including normal condition, tip discharge, intermittent discharge, floating potential discharge and surface discharge. The results show that the proposed method achieves 100% accuracy and 100% recall rate. Compared with MobileNetv3, ResNet, AlexNetV3 and other existing voiceprint recognition methods, the MobileNetv3-ECA network has faster convergence speed and higher classification accuracy.