Statistical modeling of speech signals based on generalized gamma distribution

In this letter, we propose a new statistical model, two-sided generalized gamma distribution (GGammaD) for an efficient parametric characterization of speech spectra. GGammaD forms a generalized class of parametric distributions, including the Gaussian, Laplacian, and Gamma probability density functions (pdfs) as special cases. We also propose a computationally inexpensive online maximum likelihood (ML) parameter estimation algorithm for GGammaD. Likelihoods, coefficients of variation (CVs), and Kolmogorov-Smirnov (KS) tests show that GGammaD can model the distribution of the real speech signal more accurately than the conventional Gaussian, Laplacian, Gamma, or generalized Gaussian distribution (GGD).