Abstract

A synthetic voice system which can convert typed text to speech calculates the intonation presented by the input text. The system utilizes a pitch (F0) module to calculate an F0 value for the beginning and middle of each phoneme. The following procedure is used. The F0 value for all the stressed syllables are calculated along with all F0 values for the syllables preceding a silence. The calculated F0 values for the syllables are placed on their associated phonemes. The valleys between the stressed syllables are approximated. When the last syllable of a declarative sentence is stressed and in WH question and exclamatory sentences, the FO fall is controlled to be gradual at first and then sharper toward the last utterance. When that last syllable of the declarative sentence is not stressed, the fall is sharper at first and then more gradual toward the last utterance. In "yes/no" questions, there is a final rise after the last stressed syllable of the sentence. The last stressed syllable is assigned a low FO value which is approximately equal to the average FO values of the speaker. To prevent an unnatural sounding, sharp FO rise in these questions when the last accented syllable occurs on the last syllable of the sentence, the final FO rise is lower than that of the "yes/no" question when the last accented syllable does not occur on the last stressed syllable of the sentence.

Other References

• "Synthesis by Rule of English Intonation Patterns," by Mark D. Anderson et al, from proceedings of IEEE International Conference (1984), pp. 2.8.1-2.8.4
• "The structure and processing of fundamental frequency contours" by Kim E. A. Silverman, submitted for the degree of Doctor of Philosophy, University of Cambridge, Apr., 1987, pp. 5.26-5.49
• "Language Sound Structure" by Mark Aronoff et al, from the Massachusetts Institute of Technology, (1984)
• IEEE Computer (Aug. 1990), vol. 23, No. 8 "Text-to-Speech Conversion Technology" by Michael O'Malley, pp. 17-2