An approach of verification of key words the weighted likelihood ratio(LR) fusion based on the revised rivalmodel
is proposed to be applied to the utterance. Error function is proposed in order to realize the parameter optimization
based on the SVM. In the experiment stage, two baseline systems are setup. In the first experiment, the different weight
methods based on the rival model are distinguished. In the second experiment, three different models are compared. The
experiment result shows that, the LR computed by the weighted likelihood ratio fusion makes a better performance on
the keyword spotting, which decreases 26.53% and 7.78% comparing to the normal rival-model and classical anti-model
respectively.