Analog circuit is an essential part of the integrated circuit. One of the current research hotspots in integrated circuit testing is the detection of faults occurring in analog circuits and the accurate identification of fault types based on deep learning techniques. To address the difficulties in fault detection of analog integrated circuits, the advanced achievements of artificial intelligence in the field of image recognition and speech classification is referenced and an analog circuit fault detection idea based on a deep learning algorithm of self-attention mechanism is proposed, which can be used to detect faults in Sallen-Key low-pass filter circuits. The output signal is sampled into an audio signal and fed into an audio classification model based on a self-attentive transform network for training, testing, and optimization. The results show that fault detection based on the self-attentive mechanism audio classification has an average accuracy of 93. 1% and a maximum accuracy of 98. 1% . Nine different fault types can be detected. The model converges fast and can detect faults in analog circuits, which thoroughly verifies the feasibility of the proposed idea.