The speech input system of Chinese Character for office automation needs a large vocabulary spoken word recognition. For a word recognition system the recognition time is proportional to the vocabulary size in it. So it is difficult to achieve real time operation of the system. We propose a new word recognition system architecture using severel word recognizers which have full recognition function and template memory separately. The system's vocabulary is the sum of the recognizer's vocabulary's and the recognition time is near to a recognizer's. The central processor fetches the numbers and similarity parameters and decides which is correct input word. If the correct input word is behind the first, a select key will be typed.

We have organized several such system vocabulary in which is over 2,000 - 3,000 words and they are on trial in several organization now.