Critical mechanical faults in wind turbine systems lead to considerable downtime and repair costs. Improving the detection and diagnosis of such faults thus brings about significant cost reductions for operations and maintenance (O&M) and electricity production. One of the most common defects in drivetrains are rolling element bearing faults. Detecting the faults in their incipient phase can prevent a more catastrophic breakdown and save a company time and money. This paper focuses on separating the bearing fault signals from masking signals coming from drivetrain elements like gears or shafts. The separation is based on the assumption that signal components of gears or shafts are deterministic and appear as clear peaks in the frequency spectrum, whereas bearing signals are stochastic due to random jitter on their fundamental period and can be classified as cyclostationary. A technique that recently gained more attention for separating these two types of signals is the cepstral editing procedure and it is investigated further in this paper as an automated procedure. The performance of the developed methods is validated on experimental data from the National Renewable Energy Laboratory (NREL) in the context of the wind turbine gearbox condition monitoring round robin study.