The invention relates to a lexical stress predicting method intended to be used in the prosody analysis step of a natural language processing stage within a system of speech synthesis based on a Romanian language text, with application in developing intelligent human-machine communication interfaces. According to the invention the method uses a statistic strategy based on character-level n-grams employing a modified "Katz back-off" smoothing algorithm. The claimed method requires low linguistic knowledge, is based on a medium-sized training corpus and is very accurate in the prediction of lexical stress for the Romanian language.
展开▼