The Chinese speech understanding system BJD-I with dialogue capability used for railway reservation and query service consists of three parts: （1）Speech Acoustic Processor （2）Linguistic Processor

（3）Voice Response Processor.In the

speech acoustic processor

two algorithms are proposed.One is for voice detection and syllable separation of continuous speech by maximum formant（MFDS）.The another is for Fast Speed Dynamic Programming matching of patterns （FSDP）.The experiments show that the correct rate of syllable separating for continuonus speech by MFDS is above 98%.By using FSDP algorithm

the recognition accuracy is as same as the optimized dynamic programming presented by Sako

but the pattern matching speed is above twice faster.In the linguistic processor

a grammar model RSNG （Regular Semantic Network Grammar） for syntax analysis and a knowledge representation method called tree-structure knowledge representation are set up.The performance of BJD-I Chinese speech understanding system has been tested by a man speaker and the experiment resultes show that the correct rate of semantic entry recognition is 80% and the 98% of service was well completed.