Abstract
Keyword spotting (KWS) is an active research issue in recent years. Among existing spotting methods based on pattern matching are DTW method, HMM method and neural network (NN) method, and these method have achieved good result to some certain extent. This paper proposes a new method of keyword spotting based on theory of speech feature space trace time normalization. Experiments show that this method has achieved performance close to that of manual spotting, having some practicability.