Key points are not available for this paper at this time.
Twenty-five English noun-verb pairs, differing principally in their stress patterns, were recorded by 16 native speakers of American English. The fundamental frequency, relative amplitude, duration, and integral of the amplitude with respect to time of the stressed and unstressed syllables were measured and related to the aurally perceptible stress patterns. The relative value of these factors as cues to such patterns varied with different speakers, but it was found that certain “trading effects” offset a lack of differentiation in one acoustic dimension by changes, coherent with the perceptual stress pattern, in another dimension. A simple binary automatic stress recognition program, based on the redundancies inherent in the acoustic correlates and trading effects, was then devised. The stress judgments made with this program agreed with the perceived patterns 99.2%
Philip Lieberman (Fri,) studied this question.