提出一种以词片为识别基元的特定位置后验概率网格方法,解决了语音检索中的集外词处理问题,实现开集语音文档检索。通过引入前端查询扩展技术生成多个候选查询以补偿语音识别过程带来的误差及用户输入误差。在检索过程中引入编辑距离权重来实现模糊查询,提高系统的查全率并增强系统鲁棒性。实验表明该方法在集外词处理和系统鲁棒性等方面优于传统方法。 A new posterior probability grid method based on word-specific recognition unit is proposed, which solves the problem of processing extraneous words in speech retrieval and achieves the retrieval of open speech documents. By introducing front-end query expansion technology to generate multiple candidate queries to compensate for the error caused by the speech recognition process and user input errors. In the retrieval process, the author introduced the edit distance weight to achieve fuzzy query, improve system recall rate and enhance system robustness. Experiments show that this method is superior to the traditional methods in terms of word processing and system robustness.