In this paper, a recogniton algorithm based on the characteristics of Chinese is proposed for the real-time recognition of Chinese connected digit using an associative method. A Chinese character is a CV monosyllable. A Chinese digit string is easy to be segmented into individual digits which carrys little information. The present DTW method only takes account of entire difference between two digits regardless of local difference whose exsistance sometimes is enough for us to distinguish the two digits from each other. We have found l.the recognition performance will improve when recognition results from two seperated parts of one digit are associated. 2. the recognition performance will improve further when recognition results from different distance measures are associated.

Using above recognition method, we have conducted several experiments on four-to-five-digit strings.A 16-ch BPF is used as spectral analyzer. For the speech spoken at normal speed C about 200 to 250 digits/min. ), the recognition accuracy is greater than 97% with a response time less than 0.5 second.