Speaker variation in English prosodic boundary

This paper analyses the rate of inter-speaker consistency in the way multiple speakers render prosodic events when they read the same scripts. Prosodically labeled data of five speakers from the Boston Radio Speech Corpus (BURSC) are used to measure the degree of speaker variation in rendering prosodic boundaries. The results indicate that the average rate of consistency on the presence or absence of boundary tones 89.71%. For the rate of consistency for the levels of boundary tones, an average consistency of 79.25% is achieved when three levels of category (i.e., ip, IP and no boundary) are compared in pairs of speakers. The rate is lowered to 76.74% when a comparison of prosodic strength is made after both speakers in a pair agreed that there are phrasal tones on aligned words. When a pair of speakers both have a prosodic boundary on a given word, the agreement rate on the type of phrasal tones is 50.95%. The rate of speakers' consistency in the presence of boundary tones is comparable to the rate of inter-transcriber reliability. The comparable rates with regard to locating prosodic boundaries in utterances by speakers and transcribers may be interpreted that the production and perception of prosodic phrasing are closely related to each other. The high rate of speakers' consistency is interpreted to be affected by syntactic structures, in spite of the lack of isomorphic relations between prosodic phrasing and syntactic phrasing.