Hi. I am trying to write a program to verify the speaker. I'm confusing with the viterbi algorithm in Hidden Markov Model(HMM). As I know, the viterbi algorithm is used to find the best path of the sequence. I would like to know how this algorithm can help when I want to verify whether the speaker is imposter or not. [the log likelihood computed is too small. It is around -300 for the true speaker.] I would like to know whether I miss some important steps while verifying the speaker.I found the transition matrix,observation matrix and the initial state sequence for the training data set and the testing data set. Then, I found the best path and the probability,p(T) by using Viterbi algorithm. In Viterbi algorithm, the initial state distribution & transition matrix that I used are from training data set while the observation sequence is from the testing data set. Finally,I found log likelihood=log(p(T)).
speaker verification with HMM
Started by ●February 26, 2007