Robust Keyword Search Using Subword Lattice

Abstract. Keyword search methods based on subword lattice can exclude the problem of Out of Vocabulary Words and improve quality of keyword search. This paper describes the keyword search method based on subword lattice and a-posteriori probability. In the paper it is proposed to use wavelet-based speech feature vector in order to improve the noise robustness of keyword search. The experiments show good keyword search results in noise conditions for the developed algorithm