Computer recognition of linguistic stress patterns in connected speech

This paper presents an automatic method which estimates the magnitude of syllable stress in continuous speech using a composite of three acoustic parameters: fundamental frequency, intensity, and duration. Results show that fundamental frequency is the most prominent cue of stress, followed by intensity and vowel duration. Estimations of stress magnitudes based on these three cues are shown to yield stable values in spite of differences in the vowels used, in speakers, and in syllable positions.