Improving the Intelligibility of Noisy Speech based on Auditory Detection with Noise Clearing

An improved method for increasing the intelligibility of noisy speech is proposed. A constructive algorithm for its implementation with an operational assessment of speech intelligibility in real time is implemented. The advantage of the method, in comparison with others, is a more complete account of speech transformation in the periphery of hearing and the use of a 25% overlap of adjacent segments. For this purpose, in order to effectively suppress noise in the observed signal, a bank of auditory filters and quadratic detection with variable coefficients, small for the noise components of the signal and large for its powerful components, have been introduced. The high efficiency of noise purification provided by the developed algorithm has been confirmed. It is shown that at a high noise level or a small signal-to-noise ratio, for example, equal to −10 dB, this algorithm provides an increase in speech intelligibility compared to the initial one by 32.6%.