A project financed by the government of St. Petersburg and concerned with elaboration of the voice interface for providing automatic input of Russian speech is described. Speech is represented at the morpheme level, which substantially decreases the vocabulary size. Morpheme databases are developed and used for collecting the statistics of morpheme coordination from text corpuses. The degree of coordination between root morphemes has principal importance during recognition. This processing provides invariance under grammatical deviations and increases the speed of recognition of Russian and other languages that have complex word-formation mechanisms.
展开▼