The computation of the squared envelope spectrum (SES) for vibration data collected from a rotating machine is a common tool used to diagnose a defective bearing. To enhance the diagnosis, pre-processing methods have to be implemented before the evaluation of the SES in order to suppress vibrations from sources other than the bearings. Conventionally, suppression of the spurious components is achieved by exploiting either their separation in the frequency domain or the contrast between deterministic signals and the random vibration components from a defective bearing. Recently, the phase editing (PE) method was used to exploit another characteristic of the vibration signal and improve diagnosis in stationary speed conditions. PE caters to, and exploits, the scenario in which vibrations from a defective bearing have a small amplitude compared to vibrations from other components, effectively thresholding the amplitudes of the spectral components of a signal. In this paper, the PE method is applied analytically and experimentally to a broader class of machinery, encompassing machines with varying speed conditions. It is demonstrated that the separation of the bearing component and masking components is equally effective in the non-stationary case.