The relationship between context redundancy and key-word intelligibility was examined in sentences having both high and low redundancy. The study partially replicates work by Philip Lieberman (1963) in which he concluded that the degree of stress and carefulness of articulation of a word varies inversely with its redundancy. More words and more extreme redundancies in word pairs were used in the current study. These word pairs were placed in similar positions in two sets of sentences: sentence pairs that one might find in text, and adages together with sentences that might be spoken. With the text-type sentence pairs, there was an intelligibility advantage for the words in lower-redundancy contexts. For the adage and spoken-sentence pairs, there was no intelligibility advantage for words in either context. It is suggested that the adage is not a good representative of high-redundancy contexts. The result that intelligibility and redundancy are inversely related in some instances (text-type sentences) indicates that information control by a speaker is going on even while reading test sentences. That is, utterance planning for the speech-production mechanism involves interpreting semantic information which may be expressed in fine adjustments in motor control.
To mark the completion of a ten-year effort to develop a high performance text-to-speech algorithm, we have established a benchmark system called “MITalk.” Components of the computer-simulated bench mark include: (1) conversion of abbreviations and special text symbols, (2) a lexicon consisting of about 11 000 morphs with pronunciation and parts of speech, (3) morpheme analysis, (4) letter-to-sound rules, (5) syntactic analysis, (6) rules for stress assignment, boundary placement and phonological recoding, (7) fundamental frequency and segmental duration prediction, (8) phonetic-to-parametric conversion, and (9) digital formant synthesis. The MITalk-79 system is being extensively documented and its performance is being evaluated. The presentation will summarize aspects of system organization and performance. (A more complete description will be given in a one-week course to be offered June 25–29, 1979.) The oral presentation will include a five-minute demonstration of synthetic speech generated from English text with absolutely no human intervention. While currently simulated on a large digital computer, MITalk-79 is amenable to practical IC technology. Implementation issues will be briefly discussed. [We gratefully acknowledge the synthesis-by-rule programs and advice provided by Dennis Klatt.]
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.