Head, jaw and tongue movements contribute to speech artifacts in magnetoencephalography (MEG). Their sources lay close to MEG sensors, therefore, the spatio-temporal signal space separation method (tSSS), specifically suppressing nearby artifacts, can be used for speech artifact suppression. After data reconstruction by signal space separation (referred as SSS), tSSS identifies artifacts by their correlated temporal behavior inside and outside the sensor helmet. The artifacts to be eliminated are thresholded by the quantitative level of this correlation determined by correlation limit (CL). Unnecessarily high CL value may result in suboptimal interference suppression. We evaluated the performance of tSSS with different CLs on MEG data containing speech artifacts. MEG was recorded with 204 planar gradiometers and 102 magnetometers in two subjects counting aloud. The recorded data were processed by tSSS using CLs 0.98, 0.8 and 0.6, and traces were compared. The speech artifact was increasingly suppressed with decreasing CL, but sufficient suppression was achieved at different CL in each subject. Alpha rhythm was not suppressed with CL 0.98 or 0.8; some amplitude reduction with CL 0.6 occurred in one subject. The tSSS is a robust tool suppressing MEG artifacts. It can be fine tuned for challenging artifacts which, after insufficient rejection might resemble brain signals.