Computer modeling and estimation of linguistic stress patterns

The concept of linguistic stress is vital in research in speech perception, speech production and computer speech recognition. The research described here has produced a new and reliable way, using computer analysis, to estimate the linguistic stress levels on individual syllables in complex utterances. An interactive "Computer Automated Stress Extraction System" (CASES) was developed to facilitate the research. This system can rapidly compute and display estimated syllable linguistic stress patterns, and all of the usual acoustic parameters from digitized speech. The following results were obtained: a) The computer estimated stress agreed well with perceived data. The stress measure adopted used a combination of five acoustic cues, and this yielded stable estimates of stress magnitudes, in simple utterances, even under variations of syllable locations and with speaker differences. With complex utterances, it was shown that in addition to the acoustic cues, listeners relied on previous linguistic knowledge in the perception of a general stress pattern. b) A new model of stress production was devised which suggested that the stress pattern of an utterance centered around a few special emphasis syllables with the stress pattern on the other syllables determined by anticipatory and recency effects.