Development of a Dictionary for Speech Synthesis

In order to synthesize any given English text, it is necessary to convert the text to a phonemic representation. In the procedure discribed, function words, prefixes, suffixes, “functional roots” (as “pel” in “impel, propel, repel”), and exceptions form a base dictionary. A large corpus of words is then processed, shortest words first, by attempting to decompose each word into prefixes, roots, and suffixes which are in the base lexicon. If the decomposition fails, or is judged to be incorrect, then the word may be added to the lexicon. There is a tendency, however, for the short high-frequency words of English to be more irregular from the letter-to-sound point of view than words of lower frequency. Hence content words are examined for possible conversion directly by a letter (s)-to-sound algorithm. In this way, the conversion of text to phonemes uses a dictionary that takes advantage of the morphemic structure of English words, and yet does not include those words that can be converted directly without the need for lexical entries.