Statistical recognition methods application for isolated word recognition
Evgeny I. Bovbel, Igor E. Kheidorov · 2002
This paper is devoted to the problem of effective isolated word recognition using a priory acoustical information about the vocabulary. It is proposed to implement the HMM method to estimate the acoustic phoneme probabilities and use them as inputs for the acoustic-phonetic hidden Markov model specially designed for the given phonetic context. The experiment shows that such approach can be successfully used for isolated word recognition.