This paper presents the development of the speech recognition systems based on Hidden Markov Models. HMM-based isolated words and continuous speech recognizers are presented. The network of connected HMM has been applied to improve accuracy of continuous speech recognition. The HMM network is build on a priori knowledge, for instance: phonetic description, phonological rules, syntax. The recognizers has been studied in an noisy environment. We have proposed two methods of creation Hidden Markov Models for noisy speech recognition. Both can be used in noise auto-adapting systems.

Keywords: HMM, Noisy Speech Recognition