Production and Perception of Fast Speech

This thesis reports on a series of experiments investigating how speakers produce and listeners perceive fast speech. The main research question is how the perception of naturally produced fast speech compares to the perception of artificially time-compressed speech. Research has shown that listeners can understand speech at much faster rates than they can produce themselves. The current study attempts to answer for this discrepancy and addresses the following questions: Why is speech intelligibility relatively unaffected by time compression? How do segmental intelligibility, prosodic patterns and other sources of information contribute? Does the intelligibility of synthetic speech suffer more from time compression than that of natural speech, and if so, why? Several intelligibility experiments were set up to answer these questions. Whereas artificial time compression of speech is normally conducted in a linear way, production studies on normal and fast-rate speech have shown that speakers compress some parts more than others. When speakers speed up, unstressed syllables are shortened more, relatively, than stressed syllables. Thus, the prosodic pattern of fast-rate speech is even more pronounced than that at a normal speech rate. This raises the question whether this natural non-linear way of speeding up might reflect a communicative strategy in order to save the stressed syllables, which are the most informative ones. Speakers are claimed to tailor their speech to the needs of the listener. Furthermore, prosodic patterns are known to be an important source of information under adverse listening conditions. Therefore, this study investigates whether modelling the temporal pattern of artificially time-compressed speech in accordance with the temporal pattern of natural fast speech improves intelligibility and ease of processing over linear compression. Secondly, it is investigated whether listeners find artificially time-compressed speech more difficult to process than naturally produced fast speech. It turns out that both the changed temporal pattern of naturally produced fast speech, and its increased slurring, or reduced articulation, make naturally produced fast speech more difficult to process than artificially time-compressed speech. This means that both the temporal and the segmental changes that speakers apply when speeding up their speech rate do not make perception easier for the listener, but are due to speakers s inability to speed up otherwise. The findings are considered in relation to current models of speech production and perception. This study is of interest to phoneticians, phonologists, and psycholinguists, as well as researchers working in the domain of speech technolog