Kotvytskyi I. Distorted speech and music signals quality evaluating

The thesis is devoted to the objectification and automation of evaluation of the quality of speech and music signals distorted in communication channels. Particular attention is paid to the analysis of the objective measures of the quality of speech and music signals. They are segmental signal-to-noise ratio SSNR, log-spectral distortion LSD, bar-spectral distortion BSD, perceptual assessment of the speech quality PESQ and perceptual evaluation of the music quality PEAQ. The matching maps between subjective DMOS quality measure and objective SSNR, LSD, BSD and PEAQ measures for situations of linear and non-linear signal distortion has been proposed. The dependences of subjective quality indicators on clipping of speech and musical signals on the value of their clipping provide the necessary basis for further calibration of objective measures of the clipping degree. At the same time, objective measures of the clipping value of speech and musical signals has been proposed. The limits of the use of objective speech quality assessments instead of subjective noisy speech intelligibility have been clarified, which can greatly simplify the procedure for objectively assessing speech intelligibility. The technology of automation of subjective assessment of speech intelligibility by the articulation method has been improved and its efficiency was tested by means of the developed prototype of the automated system of articulation testing. The results of comparison, by objective and subjective assessment of sound quality, of signals processing algorithms in the multi-microphone arrays are refined.