Abstract: The action acoustic signal of high voltage circuit breaker contains the mechanical state information of its mechanical structure. Taking the CT26 spring operating mechanism of LW30-252 SF₆ high voltage circuit breaker as the research object, this paper built a fault simulation platform to simulate five typical latent faults of high voltage circuit breaker, including oil leakage of oil buffer, fatigue of closing spring, wear of transmission shaft pin, jamming of main shaft and loosening of anchor bolt. With the sound of circuit breaker action as the detection signal, the Mel Frequency Cepstral Coefficient, Gammatone Filter Cepstral Coefficient and Power-Normalized Cepstral Coefficient of acoustic signal were extracted to construct the Mixed Cepstral Coefficient, which was input into convolution neural network for fault identification. The method was verified on the measured latent fault voiceprint data set of circuit breaker. The results show that this method can realize the voiceprint diagnosis of five kinds of latent mechanical faults of circuit breaker.