摘要： 给出一种适用于在线垃圾模型的基于动态排位信息的关键词确认方法，利用识别过程中声学得分的排位信息进行关键词确认，能在不降低检出率的同时有效降低系统的误警率，效果优于同类方法。该方法不依赖于具体的关键词表，计算简单，能够应用于实际工程中。

关键词: 语音识别, 关键词检出, 在线垃圾模型, 关键词确认

Abstract: This paper proposes a novel keyword verification algorithm based on dynamic ranking information for on-line garbage model keyword spotting systems, using ranking information of acoustics scores in the recognition process. It reduces false alarm rate effectively without lowering detection rate and gets better verification performance than others. The method does not depend on specific keywords vocabulary, is easy to calculate and can be applied to practical projects.

Key words: speech recognition, Keyword Spotting(KWS), on-line garbage model, keyword verification

中图分类号: 

• TP301.6