METHOD FOR AUTOMATICALLY ASSESSING THE QUALITY OF SPEECH SIGNALS WITH LOW-RATE CODING Russian patent published in 2021 - IPC G10L15/20 G10L19/08 

FIELD: computer technology.

SUBSTANCE: invention relates to computer technology for processing audio data. The effect is achieved by comparing the image (m, C) of the input DS with the LRVE, which is formed in accordance with the known j-th protocol, and a single reference image (m_{j ref}, С_{j ref}) of the j-th class, j=1,2,…,J obtained on the basis of the training sample with the maximum value of the speech signal quality estimate e_jmах. The divergence ν_j is used as a measure of distinguishing between them. Here m, m_{j ref} are the vectors of the mathematical expectation of the analyzed DS and the j-th reference image, respectively, C, Cj et are the covariance matrices of the analyzed DS and the j-th reference image. Based on the functional dependence е_j=ƒ(ν_j), formed at the training stage and described analytically by a power polynomial, with a known value of the divergence v_j between the image (m, С) and one reference image (m_{j ref}, С_{j ref}) of the j-th class, calculating the value of assessing the quality of the investigated DS with LRVE, formed according to the j-th protocol, without conversion to the PCM format.

EFFECT: ensuring automatic assessment of the quality of LRVE signals without converting the investigated digital stream (DS) to the pulse-code modulation (PCM) format, which ensures the establishment of a functional (analytical) relationship between the divergence values ​​and the selected measures of the speech signal quality.

1 cl, 7 dwg