音素HMMを用いたキーワードスポッティングに関する諸検討

This paper describes the investigation on the parameters in the key word spotting by use of the Phoneme HMM. The system constituted the word HMM connecting the phoneme HMMs which are composed using the word data base uttered by 5 males 5 females. The extranction of the initial and ending points of the word by adding the filler Model to the word HMM, the input parameters to model, and the effect of the use of the autovariance model amd the covariance model are discussed in this paper. The LPC cepstrum, delta-cepstrum and delta-power are investigated as the input parameters. The number of false alarm can be decreased by the investigation, but the detection rate can not be increased. And the speaker adaptaion algotihm is finally adopted.