Reducing expressive variation in speech with synchronous speech
Fred A. Cummins · The Journal of the Acoustical Society of America · 2001
Any two utterances of the same word (or other unit) will differ. The variation arises from the summation of both linguistic and nonlinguistic factors. A major problem in studying prosody has been deciding what are the linguistic, and what are the nonlinguistic factors which collectively influence a measured quantity such as duration or pitch excursion. A substantial reduction in variability can be obtained by having two or more speakers read a text in synchrony. This task demands that they eliminate most of the nonessential prosodic variability in their speech. Speakers have little difficulty with this task, and can produce highly synchronized speech with little effort and little or no previous practice. A parallel exists with musical performance: ensemble playing is much less variable than the performance of a soloist. Synchronous speech so obtained is potentially useful for distinguishing between linguistic and nonlinguistic sources of variability. It thus may be a useful tool for the experimental phonetician, where speech relatively untainted by individual expressiveness is required. It may also be of use for applications in which a minimization of expressive variation is required. Its suitability for collecting speech to use in a database for concatenative synthesis is currently being assessed.