Author : Ruchita Bari and M. R. Bachute

The aim of this paper is to remove short term noise. VAD has attracted significant research efforts in the last two decades. In this paper, I develop a novel VAD algorithm based on spectral Clustering methods. I proposed a VAD Technique which is a managed learning algorithm. This algorithm divides the input signal into two part clusters. (i.e., speech presence and speech absence frames). I use labeled data in order to correct the parameters of the kernel used in spectral clustering method for computing the comparison matrix. The parameters obtained in the teaching phase together with the eigenvectors of the normalize Laplacian of the parallel matrix and Gaussian mixture model (GMM) are utilize to compute the likelihood ratio needed for voice activity detection. Simulation results demonstrates the improvement of the proposed method compared to conventional arithmetic model-based VAD algorithms in existence of transient noise.

Keywords: Gaussian mixture model, spectral clustering, transient noise, log like hood algorithm, voice activity detection

Article published in International Journal of Current Engineering and Technology, Vol.6, No.6 (Dec-2016)