Furthermore it is very important that we use language model, syntax and accidence model in middle or big glossary continuous speech recognition.