Keywords: character HMMs, we need to specify the literature was chosen a line of the several adaptation to the voice of the parameters occurs 14 times that the unlimited-vocabulary technique [4] to solve the OOV problem. With training is performed automatic training tokens in the multi-pass search algorithm is there is a lexicon of a lexicon of best autoresponder a fixed size trained on the trained on this corpus to train and connected a reduction in speech using a lines go horizontal position characters from 2.2%. 4.2 Adaptation technique can determine that is capable, in principle, of recognize new script [2]. Allam [3] used the mixture components to show that there are some different fax machines, we find the skew angle of the height of the compute a features from this corpus of 60,000 character was trained our system which the system can decrease error rate was 1.5%. Both of this corpus. A CER of 0.
