Non-reference estimation of intelligibility of noisy speech signals

Abstract

A non-reference estimation method of the values of the intelligibility index of noisy speech signals is proposed. It is based on the intelligibility measure SNR (signal-to-noise rate) loss and Scalart speech noise reduction method. The reliability of the obtained non-reference estimates is compared by comparison with the values obtained using the reference method.

The accuracy of the proposed non-reference measure of intelligibility is investigated in the case of the effect of various types of noise on speech signals. The obtained results attest to the relatively high accuracy of the proposed method of non-reference estimation (the average relative error is 1.05–3.55%).