Large heavy-duty bearings have special working conditions. Under low speed conditions， the impact duration is pro? longed， the system response amplitude is reduced， and the fault information is easier to be covered by noise. Acoustic emission technology has been widely used in the field of structural health monitoring and equipment condition detection because of its sensi? tivity to weak damage. The spatial localization method in acoustic emission technology can be used to accurately locate faults of large bearing with low speed and heavy load. The localization effect depends on the accurate arrival time of signals. The identifica? tion and accurate separation of each acoustic emission event is a major challenge at present. Gate recurrent unit network （GRU） can consider the internal in sequence data and extract temporal correlation features， which has certain advantages in signal process? ing. Akaike information criterion （AIC） can effectively identify two different stochastic processes. In this paper， an acoustic emis? sion signal time of arrival picking method based on GRU and AIC is proposed. The results based on the lead and test data show that the proposed method has great potential in determining the large， heavy-duty， low-speed bearings acoustic emission signal ar? rival time by comparing with the traditional AIC， threshold discrimination and short term averaging/long term averaging.