This article considered the guarantee of confidentiality task of audio information transmitted over a packet communication channel. Domain analysis showed, that in most cases cryptographical method used for guarantee of confidentiality of audio information, however, along with them, matrix methods of protective coding used, with implementation on FPGA or DSP processors. These methods, using random matrices, impose additional instrumental errors during decoding. The purpose of the study: is to assess the quality of protective coding ‒ masking of audio data by quasi-orthogonal structured matrices. The proposed method of masking digital audio information by quasi-orthogonal matrices differs from the known ones in predictable results and simplity of implementation. An alternative approach to the classical use of metrics is proposed for evaluating the result of masking. It is based on the analysis of the spectral component of the signal. It is shown that masking with a square Mersenne ‒ Walsh matrix brings digital audio information, presented in the form of a matrix, to a form close in spectrum to white noise. This reliably protects it in the communication channel from unauthorized access. The obtained results showed that the proposed method is promising and suggests the study of the influence of the choice of orthogonal matrix structures and their sizes on the results of masking.