The temporal pattern of human communication is inhomogeneous and bursty, as reflected by the heavy tail distribution of the inter-event times. For the origin of this behavior two main mechanisms have been suggested: a) Externally driven inhomogeneities due to the circadian and weekly activity patterns and b) intrinsic correlation based inhomogeneity rooted deeply in the task handling strategies of humans. Here we address this question by providing systematic de-seasoning methods to remove the circadian and weekly patterns from the time series of communication events. We find that the heavy tails of the inter-event time distributions are robust with respect to this procedure indicating that burstiness is mostly caused by the latter mechanism b). Moreover, we find that our de-seasoning procedure improves the scaling behavior of the distribution.