Abstract: A framework was proposed which attempts to incorporate landmarks into segment based Mandarin speech recognition system．In the method,landmarks provide boundary information and phonetic class information,and the information is used to direct the decoding process．To prove the validity of this method,two kinds of landmarks which can be detected reliably were used to direct the decoding process of a segment model(SM)based Mandarin LVCSR system．Experiments conducted on “863-test” set show that decoding time can be saved about 12.92% without obviously decreasing the recognition accuracy．Thus,potential of the method is demonstrated.

Key words: Speech recognition,Stochastic segment modeling,Decoding,Landmark