Safari Books Online is a digital library providing on-demand subscription access to thousands of learning resources.
The applicability of various types of CSS systems to assistive devices depends on three broadly defined characteristics: intelligibility; naturalness; and technical considerations. Obviously, intelligibility of the CSS system is crucial to its usefulness in an SGD. Naturalness, broadly defined, is a multifaceted attribute that includes both the extent to which the voice resembles a human talker (perhaps a specific human talker), and also the extent to which the CSS system is able to impart human-like expressiveness to the synthetic speech. Expressiveness would mean, at a minimum, an ability to render the prosodic features of an utterance to reflect the different meanings talkers might want to convey with exactly the same phonetic content. Consider, for example, the different meanings one might intend by "Yes." or "Yes?" or "Yes!" and the associated prosody for each. Expressiveness can also mean an ability to impart more global emotional qualities such as happiness, sadness, or anger to synthetic speech. Finally, the CSS system must be implemented within the technical constraints of the SGD in which it is to be used. In the following, we consider each of these factors as they apply to present generation CSS technology generally, and the ModelTalker system specifically.
Users of SGDs depend heavily upon the quality of their synthetic speech. Foremost, of course, is the concern that the synthetic speech be intelligible. For many years, the DECtalk systems were regarded as the most intelligible systems on the commercial market. For example, Perfect Paul— the most intelligible DECTalk voice—provides sentence-level intelligibility of 86-95% words correct in meaningful sentences, and single-word intelligibility of about 87% correct for isolated words in an open response set (Greene, Manous, & Pisoni, 1984; Logan, Greene, & Pisoni, 1989). Other voices built into DECTalk (there are 9 in all) provide slightly lower intelligibility (Logan, et al., 1989). Such results led to DECtalk being the CSS system of choice in many SGDs.