To show the effectiveness of the Ergodic HMM as a language model, speaker dependent continuous speech recognition experiments were carried out. In these experiments, a standard speech recognition algorithm employed a continuous mixture HMM and beam search, with the Ergodic HMM providing the language information[6].