Cyclic Wiener filtering method (CWFM) has been widely used in engine noise sources separation but it is extremely time-consuming. In order to improve the efficiency, this paper proposes a fast separation method of engine noise sources based on the short-time segmentation method. In this method, signals from the acceleration condition are discretized into a large number of short-time signal segments to replace the steady-state conditions to separate engine noise sources at all speeds. The prerequisites for the short-time stationary assumption of the signal under acceleration condition are studied through experiments. The cylinder pressure-dependent noise (CPDN) separated by the fast separation method has been proven to have high precision through an experiment of accelerated condition under a load near zero. This method is used for noise sources separation of a diesel engine. It only needs to repeat the uniform acceleration several times for each load and the engine noise in all operating conditions can be separated into CPDN, load-dependent noise (LDN) and mechanical noise (MN). The result initially indicates the optimization direction of the engine noise. In addition, CWFM can be used to further separate the combustion noise in the operating conditions where the CPDN is prominent. The engine noise sources can be separated efficiently and accurately by maximizing the advantages of these two methods.