Power exponential densities for the training and classification of acoustic feature vectors in speech recognition

We consider a parametric family of multivariate density functions formed by mixture models from univariate functions of the type exp(-\x\(alpha)) for modeling acoustic feature vectors used in automatic recognition of speech. The parameter alpha is used to measure the non-Gaussian nature of the data. Previous work has focused on estimating the mean and the variance of the d ata for a fixed alpha. Here we attempt to estimate the alpha from the data using a maximum likelihood criterion. Among other things, we show that ther e is a balance between a and the number of data points N that must be satis fied for efficient estimation. Numerical experiments are performed on multi dimensional vectors obtained from speech data.