基于语音检索的侦听系统的实现与优化

Implementation and Optimization of Interception System Based on Speech Retrieval

摘要: 本文以基于关键词识别的语音检索技术为基础，结合自适应的相关性过滤模块，构建了优化的侦听系统。对基于语音检索的侦听系统从两方面进行了改进和优化：一是以上下文相关的声韵母作为关键词识别的声学建模基元，二是添加相关性判决模块对与主题无关的语音进行再过滤。实验数据证实，相关性判决模块能以召回率的小部分牺牲换取系统查准率的大幅度提升。

Abstract: Based on the technology of speech retrieval and combined with the adaptive correlation filter module, an optimized interception system is constructed. The interception system based on speech retrieval is improved and optimized from two aspects. One is using context related initial- final as the acoustic model of keyword recognition. The other is putting forward a correlation judgment module to re-filter the speech documents that have nothing to do with the subject. Confirmed by the experimental data, the great enhancement of system precision can be obtained by small sacrifice of recall rate while using correlation judgment module.