The research issues that must be dealt with in the progress of automatic Chinese speech recognition from laboratory demonstration into commerical use were described briefly. The characteristics and rules of phonetic ceding for Chinese speech were examined

putting emphasis on:（1） More explicilty using language model at the initial/final level should be helpful for continuous Chinese speech recognition;（2）The combination of distinctive guiding features and descriptive uniform features should speed searching

reduce mismatching and improve discriminating allophones. The challenges for improving robust speech recognition at the acoustic processing stage was also discussed. A step by step training and adaptation method was presented for large vocabulary continuous Chinese speech recognition.