Results of experiments are shown in Figure 3 . For comparison, ``no grammar'' results and word bigram model results are also shown.
In sentence speech recognition experiments, we obtained 39.5% for the 8-state ergodic HMM for text-open data. The the other hand, the word bigram grammar was 36.8%. These results mean that the Ergodic HMM outperformed the word bigram grammar for text-open data. In addition, the Ergodic HMM showed a superior generalization ability to the word bigram grammar, as evidenced by the relatively small difference between the text-open and text-closed performance.
