VOIP compressed-domain automatic speaker recognition based on probabilistic stochastic histogram

Compressed-domain automatic speaker recognition is based on the analysis of the compressed parameters of speech coders. This paper presents compressed-domain speaker recognition approach based on the Probabilistic Stochastic Histogram algorithm. We propose a framework based on Vector Quantization Probabilistic Stochastic Histogram(VQPSH) algorithm and perform speaker recognition on the feature vector which is directly extracted from G.729, G.723.1 6.3k, G723.1 5.3k compressed bit stream. We also propose a speaker recognition algorithm based on Gaussian Mixture Model Probabilistic Stochastic Histogram (GMMPSH). The experimental results show the Probabilistic Stochastic Histogram is superior to classical GMM using the same feature vector.