An improved model for unsupervised voice activity detection

The antique way to express our self is speech and nowadays speech is being used in many applications especially in machine communication. As the application of speech is increasing at rapid rate, therefore various techniques are evolving to separate out the speech signals from audio signal which is mixture of noise and speech. The method to distinguish voice and noise is known as voice activity detection. This method is gaining huge popularity as it removes background noise and acceptable approach in the area of speech coding, audio surveillance and monitoring. In this manuscript, hybrid model of unsupervised classifier is investigated. The proposed approach is tested at different levels of noise signal and overlap window size. To validate the proposed approach, a comparison with existing artificial neural network and support vector machine (SVM) is presented. The outcomes of the proposed method are observed better than the existing methods with the accuracy of 99.73% along with better SNR of 25.61 dB. Also proposed model LFV-KANN efficiently handles increase in noise power by hybridisation of two classifiers: ANN and K-means clustering.