Approximation of fractional of energy for sound signal fragment by mixture of radial-basis functions

Author(s):  S.V. Umanets, Belgorod branch of Public Stock Company «Rostelekom», Belgorod, Russia, 286084@bsu.edu.ru

Issue:  Volume 45, № 4

Rubric:  Computer simulation history

Annotation:  The algorithm of digital speech signal processing is considered. The segment of the speech signal in the frequency domain is represented as fractions of energy. The fractions of energy are considered as a function from of frequency interval numbers as an argument. The problem of choosing an equivalent for function of the fractions of energy in the form of a weighted sum of radial basis functions is formulated. To estimate the measure of proximity between the initial function and the selected mixture, a functional in the form of a sum of squared differences is composed. The weight coefficients are subject to an additional restriction, taking into account the physical feasibility. Parameters of radial basis functions and weight coefficients were selected by the method of conjugate gradients. It was found that the accuracy of the approximation strongly depends on the initial values of the selected parameters. An algorithm for specifying the initial approximation, which increased the accuracy, is proposed

Keywords:  fraction of energy, the method of conjugate gradients, a mixture of radial-basis functions, a mixture of Gaussian functions

Full text (PDF):  Download

Downloads count:  387