Syllable-Phoneme based Continuous Speech Recognition

Abstract

Syllable-phoneme based continuous speech recognition research is presented in this article. This investigation is continuation of previous experiments. Achieved results are set as a point for further investigations. Evaluation of distinct acoustic models is considered and an evaluation criterion is proposed. According to proposed criterion recommendations were made for syllable-phoneme unit and model set formation. Few such sets were formed and tested, evaluating their performance in total speech recognition accuracy and recognition accuracy of distinct models. At the same time we tested syllable set formation pattern according to quality of syllable that could be additional to selection according to its repetition counts. Achieved results suggest using frequency criteria. Bibl. 11 (in English; summaries in English, Russian and Lithuanian).