Large vocabulary continuous speech recognition system for Polish

Downloads

Authors

  • K. Marasek

Abstract

This paper describes the development of the LVCSR (Large Vocabulary Continuous Speech Recognition) system for Polish. All system components have been built from scratch: grapheme-to-phoneme converter, language models and acoustic models. Test results for twenty thousands word vocabulary continuous speech recognition (read sentences) are given. The system can be used as a basis for application oriented continuous speech recognition.