Speech recognition refers to biometric methods of personal identification along with fingerprint identification, hand configuration, shape and features of the face, retina, or gene structure. One of the main incentives for the study of speech recognition processes is the desire for more reliable recognition of the personality, since the uniqueness of the biometric characteristics provides a higher reliability of identification. The purpose of the article . Analysis of the difference between the verification and identification of the speech signal, which determines the different formulation of the problem and different methods for its solution. In addition, an analysis of the physical characteristics of the formation of the voice, determining the individual feature of the tonality of the subject. Subject recognition systems are divided into two types by voice: verification and speaker identification. During verification, the received voice sample is compared with the standard and its identity with that standard is established. It is necessary to make a decision out of two possible: there is a subject to those for whom it claims to be, or not. The task of identifying the subject is significantly different from the task of verification, it is necessary to identify from the set of N subjects who, according to their voice characteristics, will coincide with the characteristics of previously stored standards. The speech signal is a complex, multifaceted phenomenon, the parameters of which depend on a number of features of each person. The choice of the analyzed parameters, which can be used to recognize the subject, should be based on the objectives, methods and objectives of recognition. In modern voice recognition systems, people tend to use as much high-level information as possible, which is closely related to low-level information. The probability of correct recognition of the speaker increases in proportion to which signs are analyzed, the number of signs and their correlation. If during verification a solution can be formed using automatic systems, then today there are no clear characteristics when identifying by which it is possible to identify similarities between the disguised voice and the reference, since there are no algorithms for reducing the disguised voice to the “normal” pattern. The combination of subjective and objective methods can improve the accuracy of recognition, but the probability of correct identification of the subject in the case of masking the voice is still low.