Vocal effort levels and underlying acoustic phonetic characteristics

This thesis presents empirical research which investigates diverse vocal effort levelsin anechoic conditions from a large number of British English speakers. Five vocaleffort labels were stipulated: hushed, normal, raised, loud and shout. New resultsshow similar averages to previous work, but lower standard deviations which areattributed to the more descriptive vocal effort labels used in this study.Building on this data, a phonetic investigation into vocal effort was carried out,taking its inspiration from the natural segmental variation in speech which occurs atdifferent vocal efforts. Speech was split into voiced and unvoiced components whichallowed for a more phonetically motivated profile of vocal effort. In a bid to mirroranalysis with the speech communication chain of talker to listener, listening tests wereconducted where subjects rated the perceived vocal effort level of isolated words.Listeners also rated perceived clarity of articulation, taking the assumption thatdifferent degrees of clarity are an inherent feature of different speech levels. Multipleregression analysis revealed that listeners exploit the voiced-unvoiced distinction intheir perception of vocal effort and clarity of articulation. The resulting validationequation from the multiple regression model showed that vocal effort levels can bereliably predicted