Conditional Random Fields offer some advantages over traditional models for sequence labeling. These conditional models have mainly been introduced up to now in the information retrieval context for information extraction or POS-tagging tasks. This paper investigates the use of these models for signal processing and segmentation. In this context, the input we consider is a signal that is represented as a sequence of real-valued feature vectors and the training is performed using only partially labeled data. We propose a few models for dealing with such signals and provide experimental results on the data from the eye movement challenge.