ISCA Archive Interspeech 2005
ISCA Archive Interspeech 2005

Improvement of rejection performance of keyword spotting using anti-keywords derived from large vocabulary considering acoustical similarity to keywords

Makoto Yamada, Tsuneo Kato, Masaki Naito, Hisashi Kawai

This paper proposes an efficient anti-keyword derivation method to improve the rejection performance of keyword spotting. In this method, each anti-keyword is derived from the large vocabulary lexicon considering acoustic similarity to keywords, making use of the confusion matrix. Experimental results show that a 3% improvement of the rejection rate is obtained compared to conventional methods that do not have our anti-keyword models.