The present work introduces an analysis framework for the de-
tection of metastable signal segments in multivariate time series. It
is shown that in case of linear data these segments represent tran-
sient generalized synchronization, while metastable segments in circu-
lar data reflect transient mutual phase synchronization. We propose
a single segmentation approach for both types of data considering the
space-time structure of the data. Applications to both event-related
potentials and single evoked potentials obtained from an auditory odd-
ball experiment reveal the lack of the component P300 in an experi-
mental condition, indicates attention effects in component N100 and
shows dramatic latency jitters in single trials. A comparison of the
proposed method to a conventional index of mutual phase synchro-
nization demonstrates the superiority of considering space-time data
structures.