In this paper, we give the general principles behind an automatic system, developed at IRIT and capable of labelling speech signal for phonetic events. First, we present a non-linear smoothing method that was run on both amplitude and ZCR. Next, we give results obtained on EUROM-0 English, French and Swedish corpora. The results demonstrate that the labelling operation becomes completely independent from either language, corpus or speaker. Moreover, this operation requires no manual adaptation or training whatsoever. Keywords: Automatic Segmentation, Multilingual Speech Analysis, Phonetic Events, Non-Linear Filtering