关键词检出算法  

Keyword spotting algorithm

作　　者：戴海生[1] 朱小燕[2] 罗予频[1] 杨士元[1] 

机构地区：[1]清华大学自动化系,北京100084 [2]清华大学计算机科学与技术系智能技术与系统国家重点实验室,北京100084

出　　处：《清华大学学报（自然科学版）》2006年第1期130-132,共3页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(60272019)

摘　　要：在关键词检出系统中由于替换错误而导致的识别性能下降时,为了在较低的误警率下提高关键词检出系统的检出率,提出了关键词恢复算法,通过对被拒绝的语音信号再次分析从而恢复关键词。实验表明:引入关键词恢复算法,在误警率为10%时,检出率从78.1%提高到85.3%;检出率的上限从85.0%提高到92.0%。采用关键词恢复的检出算法极大地改善了系统性能,不仅在较低误警率时检出率有很大的提高,而且关键词检出率的上限也得到很大的提高。Keyword spotting system detection performance degrades when substitution errors occur during the keyword recognition. Therefore, a keyword recovery algorithm was developed to reduce the detection degradation to a low false alarm rate to recover keywords after rejecting acoustic noise, Tests show that the detection performance improves from 78.1% to 85.3% for a false alarm rate of 10% with the upper range of the detection performance improved from 85.0% to 92.0%. Thus, the system performance is greatly enhanced by the keyword recovery algorithm for both the average and maximum detection performance.

关 键 词：声音识别 语音确认 关键词检出 关键词恢复 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]