| |
Abstract:
Studies on the acoustic correlates of prosody have emphasized
its dependence on F0 variation (e.g. Van Lancker & Sidtis
1992). We examine how spectral manipulations affect the
identification of sentential prosody. Three affective and three
linguistic conditions were tested: happy/angry/sad and
statement/continuation/question. A female native speaker of
American English produced the stimuli. The recorded signals were
manipulated to create synthesized, filtered and re-entrant
conditions. The filtered conditions included a low bandpass range
(80-300Hz), corresponding to the range of F0, and a high bandpass
range (1830-2240Hz). In the re-entrant condition, the spectral
signal was convolved with another recorded steady state vowel. The
re-entrant condition retains the overall spectral envelope while
the spectral fine-structure and the formant structure are modified.
Contrary to expectation, psychophysical testing (n=36) showed that
performance is remarkably resilient to spectral manipulation.
Interestingly, only the re-entrant condition significantly reduced
performance. We hypothesize that the acoustic integrity of syllabic
information is critical to prosodic analysis. Typical syllables
have duration, onset, and phonemic cues. Our data suggest that
phonemic information is not necessary (high band pass) and duration
is not sufficient (re-entrant) to effectively cue sentential
prosody.
|
