Acoustic Correlates of Listener-Identified Boundaries in Spontaneous French Speech

Prosodic structure provides an organization to spoken (and signed) language that signals the grouping of words into phrases, and the relative prominence of words and syllables (e.g., ShattuckHufnagel and Turk 1996). This organization is cued by articulatory and acoustic modifications to the signal. Different languages modulate different dimensions of the signal in different ways, but prosodic variation often includes durational lengthening or shortening, addition or removal of pauses or breaks in the flow of speech, changes in F0 patterns, expansion or shrinkage of gestural magnitude. These modulations facilitate the task of listeners by communicating the structure of the speech. Spontaneous speech, in particular, very often does not conform to conventional syntactic patterns, so listeners cannot rely on their knowledge of syntax to interpret what they hear. Linguists have posited a hierarchy (Nespor and Vogel 2007) of prosodic units of different sizes, but it is not clear to what extent these different units are relevant for ordinary listening, nor do we know much about how listeners’ perception of prosody differs among different languages. In order to investigate the place of prosodic units in listeners’ grasp of a spoken message, studies have been done attempting to access the perceptions of naive listeners. These may well be different from the perceptions of linguists who bring theoretical expectations to bear when they analyze the structure of an utterance. This study takes listener perceptions as the basis for phrasal structure, and seeks to identify the acoustic properties of the speech signal that contribute to their perceptions. The first step was thus to obtain data on perceptions of prosodic structure, specifically the locations of boundaries between groups of words. The second step was to examine the speech signal immediately preceding the locations that were perceived as boundaries by a consensus of the listeners, since it is assumed that a listener needs to be able to detect a boundary from modulations of the signal before the boundary, if they are to interpret the speech stream successfully.