Rotary self-motile micro- or nanoparticles are usually present in a mixture of microswimmers with different linear and angular velocities. We show that using a combination of a flow field and a soft periodic potential in the overdamped regime of dynamics, we can separate these rotary active particles into well-defined dynamical subclasses. Stochastic diffusion enhances the separation process with new types of drift absent in the noise-free case. A practical candidate for the implementation of this system is an acoustic field, and the observed phenomena can be useful for laboratory microscales manipulations.