A VOICE DETECTING METHOD USING A SUPPORT VECTOR MACHINE

摘要

An audio detecting method using a support vector machine is provided to use a likelihood ratio per a frequency, which is used in an existing audio detecting method based on a statistical model of audio, as a feature vector of an SVM(Support Vector Machine), thereby drastically increasing performance of the existing audio detecting method based on a statistical model of audio. A feature vector about a likelihood ratio is extracted from a voice database for training(310). An SVM(Support Vector Machine) model is generated by using the feature vector about the extracted likelihood ratio(360). A likelihood ratio is obtained from an input audio signal(330). A value of a decision function is calculated by applying the likelihood ratio to the SVM model(340). Whether audio exists is determined by comparing the value of the decision function with a threshold value.