Is the listener using fixed reference targets, when perceiving either read or spontaneous speech, or is ne/she attuning his/her perception mechanism to the specific speaking style, applying some kind of style normalization. To answer this question we spliced function words from three different speaking styles into two types of lexically identical short sentences, either from free conversation, or from the same text read aloud. To avoid F0-influences the sentences were made unvoiced by LPC-resynthesis. It turned out that in a number of cases function words segmented from more carefully pronounced speaking styles were heard as accented when spliced into more reduced speaking styles, merely because of their longer duration and/or more spectral contrast than the original words.