Construction of the system of recognition of voice signals

Author(s):  Hein Min Zaw, no, no, Kursk state University, Kursk, Russia, heinminnzaw13@gmail.com

V.M. Dovgal, Dr., Kursk State University, Kursk, Russia

V.A. Kudinov, Dr., Prof., Kursk State Agricultural Academy named after prof. I.I. Ivanova, Kursk, Russia, kudinovva@yandex.ru

Issue:  Volume 46, № 2

Rubric:  Infocommunication technologies

Annotation:  The paper describes approaches to the development of a speech recognition system, which is presented as a complex set of different processing units, for which a set of parameters is established that provide the most accurate description of a speech signal in an unfavorable jamming environment. Two main approaches are considered when analyzing in the frequency domain - analyzing the spectrum using a bandpass filter unit and analyzing coding with linear prediction (LPC analysis). It was determined that in order to ensure that the parameters are informative with acceptable computational costs for their processing, the spectrum obtained using the discrete Fourier transform is divided into frequency bands and the average signal energy value is calculated in each band. For the parametric description of a speech signal, linear prediction coefficients and reflection coefficients are used. Applying the inverse Fourier transform to the logarithm of the linear prediction spectrum allows us to obtain cepstral coefficients, the advantage of which is their uncorrelatedness and the possibility of reducing the influence of the transmission channel on the parameters of the speech signal. For a practical example, the logarithmic spectrum of the signal obtained by the usual Fourier transform and using LPC coefficients is considered. The main advantages of the LPC-method in comparison with the analysis with the help of a filter block are noted.

Keywords:  speech signal, recognition system, Fourier transform, auxiliary coefficients, LPC coefficients, z-characteristic.

Full text (PDF):  Download

Downloads count:  254