Some Issues in Investigating Speech Styles

This paper considers some of the issues surrounding the collection of a database contrasting the two styles of Casual and "Careful" speech.One of the aims of the VOX project , of which this work is part, is to study speech style variation for use in Speech Synthesis, and the discussions in this paper will be made with reference to the data collection technique that has been developed for the project at LIMSI in Paris.Recording is well underway at LIMSI as part of the French e contribution to VOX, and at Sheffield we are implementing a variation of this technique, suitable for English.but with results that can be compared to the French.The term.Speech Style is taken to refer to certain types of intra-speakcr variation.Here we must distinguish between the variation due to the emotive state of the speaker, that due to linguistic repertoires associated with social and regional variation and that due to repertoiru associated with situational context.It is only the third of these types of variation which we address here.That is. the different ways in which someone will pronounce the same words.for example, sometimes saying Idld ju/ and'at other times ldja/ or Idlju/ for did you .For instance, a casual style would be distinguished from the phonemic reductions associated with the informant being in a depressed state or very excited and a careful style would be distinguished from a formal register, in that formal speech belongs to a linguistic repertoire l which might have both words and syntax that differ from a non-formal register.Thus. the term 'style' is intended here to capture only variation due to different tasks, settings and audiences, and the way in which these interact with the speaker s perception of the audience requirements.Thus a speaker, while using a particular dialect, may enunciate it in a different manner according to the perceived difficulties of the communicational setting or the attributed importance of accuracy in conveying the message.Amongst researchers working on speech style there may not be complete agreement over the way in which it has been defined in this paper.The term does not have astable techniml definition as yet 2 and in this work we are adopting.what might be seen as. a rather restrictive view.It may be more precise than other uses but this is not without problems: for data collection purposes, the definition must relate also to what it is possible to record and measure.emotive factors can not really be excluded so easily from different recording conditions.De nitions of Style, and the terms used to refer to different styles, are also dependent upon the psychological theory underlying explanations of why these variations occur, and some of the existing theories will be presented in this paper.The paper will also look at practical attempts to study style contrasts and present the method for style data collection being applied at Sheffield.There is, then.much scope for discussion and it is hoped that this paper will make a useful contribution.