To obtain high recognition precision even when the load of a network is small and voice made up with a large vocabulary is inputted.

A voice recognition means 18 of a terminal 1 analyzes voice signals from a voice input means 13 and extracts a syllable train in which a syllable is made into a basic unit. A communication means 16 transmits the syllable train to a host 2 through a network 3. A text converting means 29 of the host 2 inputs the syllable train received by a communication means 26, obtains a word train to matche with the syllable train by using a difference model and a language model and outputs the word train as a text. The means 26 transmits back the text to the terminal 1 and a display means 11 of the terminal 1 displays the text.