Fault Detection of In-Service Bridge Expansion Joint Based on Voiceprint Recognition

Table 8

Performance comparison of VPR models.

Model	Accuracy (%)	maAuC	Inference time with GPU (ms)	Inference time with CPU (ms)
ConFormer VPR (w/MFA)	99.58	0.975	16.9	127.4
ConFormer VPR (w/o MFA)	98.55	0.965	16.8	115.1
Transformer	98.79	0.929	10.1	70.8
ECAPA-TDNN	99.70	0.900	11.2	32.8
ResNet18	97.15	0.771	4.5	196.9
ResNet34	98.67	0.758	7.2	139.6