ISCA & IEEE Workshop on Spontaneous Speech Processing and Recognition

April 13-16, 2003
Tokyo Institute of Technology, Tokyo, Japan

The Phonetic Patterning of Spontaneous American English Discourse

Steven Greenberg, Hannah Carvey, Leah Hitchcock, Shuangyu Chang

International Computer Science Institute, Berkeley, CA, USA

Statistical analysis of a manually annotated, 45-minute subset of the SWITCHBOARD corpus indicates that pronunciation variation observed in spontaneous American English discourse is highly structured at the level of the syllable, particularly when prosodic stress accent (i.e., syllable prominence) is taken into account. The pattern of segmental substitutions and deletions observed are largely associated with different constituents within the syllable (nuclei and codas, respectively); their frequency of occurrence is inversely proportional to stress-accent magnitude. The phonetic identity of vocalic nuclei is also related to stress accent, as is the probability of segmental deletion in the coda. Such data imply that "information" governs much of the phonetic patterning of spoken language characteristic of the real world.

Full Paper

Bibliographic reference.  Greenberg, Steven / Carvey, Hannah / Hitchcock, Leah / Chang, Shuangyu (2003): "The phonetic patterning of spontaneous American English discourse", in SSPR-2003, paper MAO5.