One of the principal interests of researchers in automatic speech processing at present is the adequate representation of the great variability in the speech signal. Some search for common behaviour among many speakers, "invariants" in the signal, yet others search for a chart of all that varies in order to better understand the limits of variability. In a first approximation, the study of speaking styles should be of little interest to the first group. It would seem to have little effect on the nature of the linguistic message. Yet, when applying speech recognition systems to increasingly natural speech and realistic applications, it has been found that a change in speaking style affects recognition results. Elsewhere, in working with the hard-of-hearing, it was found that processors on hearing aids could be greatly improved if the information used when one speaks clearly is modelled. And yet elsewhere, developers of text-to-speech synthesis systems recognise that their systems could benefit in intelligibility and naturalness from models of clear speech. The characterisation and subsequent modelling of this style phenomenon therefore have become of great interest. As for the second group, taking a view that we must understand all sources of variability in order to correctly process speech [LR84] [LR90], style is one of the many dimensions to be explored to the limits of speaker space. The term "speaking styles" has been used and defined in many ways [LI92], almost as numerous as the authors who have dealt with the subject. It can be said that many suprasegmental changes that could not be attributed to a more well-defined area, such as emotion [MU93] have been thrown on the heap which has until present been called style, by the speech community [EM92]. We shall return to the definition of speaking style below. First it is necessary to see what has been accomplished in communities which have a longer experience in studying style, sociolinguistics and psycholinguistics.