A New Keyword Spotting Approach for Spontaneous Mandarin Speech

For many practical applications of keyword spotting, input signal is a spontaneous conversation. Generally speaking, keyword spotting system will degrade significantly because of mismatch between acoustic model and speech. To solve this problem, this paper presents a two-pass based keyword spotting strategy. Different from one-pass based system, decoding process is done in the whole acoustic space, not especially for keywords space. 1-best phoneme sequence is achieved during the decoding process firstly. Then keywords are extracted by a keyword list and confusion matrix in the second pass. At last, the phoneme lattice based confidence measure is adopted to reject false alarms. Experiments show that two-pass based strategy performances a significant improvement on the spontaneous telephone conversations