Speech parameters vector based on arithmetic fourier transform

This paper is devoted to the developing of speech character vector based on the arithmetic Fourier transform for speech recognition tasks. The calculation complexity and accuracy of three spectrum methods- fast Fourier transform (FFT), linear prediction (LP) method and arithmetic Fourier transform (AFT) were estimated. It was found that AFT is very useful to calculate speech mel-parameters and provides a simple parallel VLSI realization. If to apply the cos transformation for the AFT mel-spectral parameters we achieve robust parameters for speech signal variations. As a result of the researches the speech parameter vector based on AFT coefficients was designed.