Masking of speech messages based on modern computer technologies.

Masking of speech messages based on modern computer technologies..

MASKING SPEECH MESSAGES BASED ON MODERN COMPUTER TECHNOLOGIES

Source: magazine «Special Equipment»

In recent years, the protection of confidential voice messages has received increasing attention. This is evidenced by numerous publications in specialized journals, including virtually every issue of the journal “Special Equipment” [1–15]. On the one hand, this is due to the high polyinformation content of voice messages. On the other hand, the diversity of information threats in relation to acoustic (speech) information and the specifics of their development and implementation scenarios [1, 2, 5, 13], which is reflected in a wide variety of modern methods and means of protecting voice messages from unauthorized access.

There are two main types of neutralization of acoustic (speech) information leakage from communication channels [1, 2, 5, 8]:

• means of physical protection of voice messages, including jammers, blockers, filters, and means of searching for audio information leakage channels;
• means of semantic protection of speech information in speech communication channels.

It is known that the means of preventing leakage of audio information from the first group have a number of weak points and restrictions on their use in a particular practical situation, depending on the type of communication line, its terminal equipment, technical qualifications of personnel and other factors [1, 2, 13].

Until now, the semantic protection of speech messages by means of cryptographic methods has been considered by specialists as the only possibility of guaranteed or highly reliable protection of various speech communication channels regardless of the conditions of negotiations, technical characteristics of the communication equipment and other factors. The corresponding speech protection devices, in which cryptographic algorithms are applied to sections and/or parameters of the speech signal during the speech conversion process (for example, the cryptographic conversion algorithm according to GOST 28147-89) are called analog scramblers and digital speech conversion systems based on codecs and vocoders with subsequent encryption [2, 8, 12, 14]. A characteristic feature of the operation of scramblers is the division of the original speech signal (IS) into separate sections on a frequency-time grid with their subsequent mixing, summation and transmission in the communication channel in analog form. The peculiarity of the operation of digital speech encryption devices is cryptographic transformations over digital data of the wave form or parametric description of the RS with subsequent transmission in the communication channel in digital form [2, 8, 12, 14].

The main goal of developing speech communication systems is to identify, transmit and preserve those speech characteristics that are most important for the listener's perception. Communication security during the transmission of speech messages, and, above all, the direction of semantic protection of the RS, is based on the use of a large number of different methods and means of converting the RS. They change the characteristics of speech in such a way that it becomes unintelligible or unrecognizable to the eavesdropper who intercepted the processed speech message, or the very fact of its transmission is hidden.

Recently, both developers and consumers of speech message semantic protection tools have shown an increasingly stable tendency to use new computer technologies for ensuring voice communication security (SVCS) without using classical cryptographic methods. In this regard, computer technologies for masking voice messages are becoming increasingly attractive; one of the types of generalized classification of these technologies is shown in Fig. 1. With lower financial costs for development (primarily software), distribution and acquisition of such technologies, they can become a kind of buffer between cryptographic systems and systems for physical protection of voice messages in communication channels [5]. In addition, with the help of such technologies, it is possible to solve a number of other equally important SVCS tasks than technical closure of the RS in order to protect it from unauthorized access by introducing inaudibleness. For example, it is possible to carry out covert transmission of voice messages in various communication channels and change your voice to achieve its unrecognizability while maintaining natural sound.

Fig. 1. General classification of methods for masking speech messages

It is clear that the impetus for the widespread use of these technologies of masking sounds and speech was received in connection with the rapid development in recent years of technical means of multimedia and new approaches to the description and processing of speech signals. One of them is the approach to the construction of special software and hardware of OBRS, combining the idea of ​​​​translating an audio (speech) signal into the form of corresponding graphic images and back from an image into sound or speech without loss of information content and/or intelligibility with the capabilities of known and promising methods of digital image processing. The main core of this approach is the development and application of methods for identifying and reconstructing the parameters of narrow-band Hilbert signals present in these images. Such a parametric description of a complex original acoustic (speech) signal allows either to completely recreate its sound, or to restore and voice a «new» sound signal based on the changed and specified properties in such a parametric description [5, 7].

Research has shown that the data required to calculate the parameters of elementary narrow-band signals that make up the original sound or speech can be contained in dynamic spectral scans of this acoustic signal (AS), namely, in images of correctly calculated amplitude sonograms and/or spectrograms. Such images can be obtained during dynamic spectral analysis-synthesis of sounds and speech (DSAS), sliding over the original signal with a selected analysis window with a transition from the samples weighted by it to their frequency image based on the adopted orthogonal basis. One example of such procedures can be short-term Fourier analysis-synthesis of audio signals. Although, in some applications of OBRS for conducting DSAS it is possible to use not only harmonic, but also other bases, for example, Wavelet functions, short-term Fourier analysis-synthesis of audio signals and speech is traditionally used more often.

The parameters of narrow-band Hilbert elementary sound signals that make up the sound of the original sound or speech are manifested in the images of dynamic spectrograms in the form of a set of contours (lines) of brightness differences or tracks (chains) of local and global extremes of color saturation in levels of one color. Using special software, it is possible to identify frequencies, amplitudes, and phases of elementary sounds of a complex acoustic (speech) signal using such contours (tracks), which are visible on the frequency-time grid of dynamic spectrograms (see Fig. 2, top panel), and then reconstruct, modify, destroy, or recreate them to solve a specific OBRS problem using various known methods and tools for digital image processing.

Above – a natural female voice with highlighted trajectories of narrow-band speech components;
In the center is an artificial female voice synthesized according to a given sample;
Below is an artificial male voice synthesized according to a given sample.

Fig. 2. Sonograms of the phrase “Thank you for the coffee. What time is it?”

Thus, a powerful arsenal of tools provided by well-known graphic editors such as “Adobe Photoshop”, “Corel Draw”, “Photo Editor” and others can be applied to the selected section in the center of the upper panel of Fig. 2 of the graphic image of the PC. After the necessary processing of this section of the spectrogram image in the selected graphic editor, it can be inserted back to its and another place for subsequent synthesis and listening to the new acoustic or speech signal modified in this way.

Note that in the sonogram (spectrogram) images of natural and artificial voices (Fig. 2 and subsequent ones), constructed using the specialized software (SPO) «Lazur», the time parameter is plotted along the abscissa axis, and the frequency parameter is plotted along the ordinate axis, starting from the lower left corner of the image. The maximum power of the signal under study in the node of the frequency-time grid is indicated in black, the minimum in white, and intermediate values ​​in shades of gray.

Let us consider in more detail the various classes of masking speech messages, implemented using the proposed approach to processing RS through processing their graphic images.

Artificial Voice Systems

If traditional voice changers did not pay much attention to the sound quality (naturalness and authenticity) of artificial speech, now the situation is changing. Thus, there are reports of software products that search by voice sample. Often, during investigations, in order not to be recognized, operatives have to pretend to be someone else. All this leads to the emergence of the task of high-quality voice change during the implementation of measures for the comprehensive protection of speech information.

This is quite a difficult task, since each person's voice is individual and recognizable. Moreover, auditory perception is so perfect that it allows one to recognize the most subtle nuances of a speech signal. Human hearing quite accurately determines the signs of artificiality and naturalness of speech. Therefore, in order to solve the problem of creating a computer system of artificial voice while maintaining the naturalness of sound, both based on the speaker's voice and on a given voice sample, it is necessary to dwell in more detail on the concept of speech and its main features.

Speech is usually understood as a human-generated audio message that can be objectively recorded, measured, stored, processed and, importantly, reproduced using devices and algorithms. That is, a speech message can be represented as a speech signal, which in turn can be used to reproduce speech in reverse. That is, an equivalence sign can be placed between audio speech and its representation as a speech signal, including in digital form contained in computer files.

It is known that speech is a complex process of communication between people, including both information about the individual voice of the speaker and information about phonetic quality. Therefore, it is important to ensure the correct choice and justification of the system of features that will determine the principle of speech construction. The main features responsible for the individual coloring of speech can be divided into two groups: those associated with the physiological mechanisms of speech production and those associated with the methods of bringing it into action (articulatory activity) [3].

The first group of features is based on a well-known model of the vocal tract, consisting of a transfer function of a resonance system and a generator of excitation signal pulses. The transfer function almost completely characterizes the individual geometric shape of the cavities of the speech apparatus. The main parameters here are the characteristics of the four formant regions (average frequency, frequency range, energy), the spectrum envelope, formant trajectories and derivatives of these parameters. The frequency of excitation pulses is directly dependent on the vibrations of the vocal cords, which, in turn, depend on the length, thickness and tension of the latter. The main parameters here are the frequency (period) of the fundamental tone, the tone/noise parameter, sonority, the rise of the fundamental tone and derivatives of these parameters [3].

To calculate the parameters related to the physiological characteristics of the vocal tract, the methods of spectral-temporal analysis are most often used. Such methods of speech signal analysis are adequate to the natural mechanism of speech perception. Such methods are often based on classical Fourier analysis or parametric autoregressive analysis (linear prediction as a special case). The parameters of the first group are quite easy to extract from images of narrow-band dynamic sonograms, based on the approach proposed above.

The second group of parameters also includes intonation characteristics of the speech flow, such as intensity, speech intonation, stress system, rhythmic pattern of the speech phrase.

Among the speech signal parameters that determine the individuality of a person's voice, it is necessary to highlight the integral speech parameters that cannot be attributed to any of the groups considered, but they are strongly correlated with them and are formed under the influence of the anatomical features of the speech-forming tract and the articulatory activity of a person. That is, the analysis of the integral parameters makes it possible to determine the features of individual pronunciation for speech segments of different phonetic content [3].

It can be assumed that by changing the given characteristics, based on the proposed approach to speech processing, through the processing of graphic images of the RS, it is possible to find ways to solve the task at hand — a qualitative change in the voice by changing or generating certain parameters of the speech signal.

The most common devices on the market of technical means of protecting speech information are those designed to change the voice during telephone conversations. As a rule, they have a step range of voice change: child's, female, male. Thus, the DTVC II device (South Korea) has a two-position mode switch (female/male), a four-position switch for the degree of voice change (low/high), an audio range amplifier, an operational switch for the changer (while the conversation is not interrupted). The feedback system allows you to listen to the changed voice in real time.

An analysis of other existing inexpensive devices designed to change the voice during telephone conversations showed that most often they change the frequency range of the speech signal, less often the timbre of the voice to low or high. The «new» voices obtained in this way do not have the necessary naturalness and naturalness of sound, and in some cases have a metallic tint, «coldness» or «hoarseness». On the other hand, due to the technical implementation of the devices, the number of degrees of voice change is limited. In addition, after a short time interval in the process of telephone conversations using such devices, it becomes clear to the subscriber that the interlocutor has deliberately changed his voice. Higher-quality voice changers have another significant drawback — high cost.

In our opinion, it is possible to create high-quality voice changers based on standard office equipment, such as a computer, by implementing the proposed approach through processing images of its graphic images. Specialized software for such a computer system should modify both the harmonic structure of the speech signal, which usually contains individual features of the speaker, and the Pirogov phonetic function, which is responsible for the semantic content of the speech message. Such procedures can already be carried out on images of dynamic sonograms with subsequent synthesis of a new artificial speech signal based on the modified graphic image. A combination of such effects, with correctly performed calculations, will probably allow achieving the desired result. Some difficulties may arise when modifying pause sections. Therefore, the problem of reliable determination of tonal and noise sections in the speech stream requires its solution.

It is clear that only due to software implementation on standard technical means such a voice changer will not only be much cheaper than existing analogues, but will also provide a higher quality, truly natural sound of the artificial speech signal. Software implementation will allow for smoother voice changes from male to female, from child to adult.

An example of the capabilities of a high-quality voice changer according to a given sample is the Voice Mouse program, developed in the Moscow State University Technical Park. Particularly good results are achieved with its help when translating text into speech, voiced by a female voice. On male voices, the naturalness of the sound is much worse. This can be seen by comparing the sonogram images in the central and lower panels (Fig. 2) with the image on the upper panel.

However, it is still too early to talk about the operation of such computer systems of artificial voice in real time. The computational complexity of the algorithms does not yet allow implementing such a mode on computers with a processor class below Pentium-III. However, the experiments conducted have shown the prospects and practical significance of the chosen direction of research. Some examples of changing the rate of speech and voice based on the proposed approach are given in [7].

Technical speech closure

By technical speech closure we mean speech masking technologies related to methods and means of semantic protection of speech information and aimed at ensuring the illegibility of the protected speech message. Their implementation in practice can be expressed in mixing speech with noise and interference and/or in modifying the RS according to parameters calculated from its descriptions according to a previously known transformation law (closure-restoration).

A common form of technical speech closure is mixingthe original RS with interference in order to transmit a new, unintelligible audio signal into the communication channel, usually lying in the same frequency band as the original. Knowing the nature of the change and the type of interference, at the receiving end of the voice communication channel protected in this way, its influence is neutralized with additional purification and amplification of the restored speech signal. Thus, on the lower panel of Fig. 3 shows the result of removing quasi-harmonic interference from the useful mixture, significantly exceeding the energy level of the RS of interest, produced by means of the tools of the Lazur software.

Above – setting up a powerful quasi-harmonic interference in the speech signal;
Below – removing interference from the useful mixture at the receiving end of the communication channel.

Fig. 3. Masking of speech by quasi-harmonic interference

 There are various types of implementation of this type of masking: when the interference is comparable in power to the original RS or significantly exceeds it, when the interference is noise, quasi-harmonic or speech-like, etc. The issues of choosing the type of interference when constructing active acoustic protection devices and assessing the effectiveness of acoustic (speech) information protection are considered in detail in [2, 10, 15].

Under modificationspeech we will understand such a transformation of the original speech signal, first of all its phonetic function, with the purpose of achieving its unintelligibility and/or unrecognizability according to a known given law, when the parameters of this transformation at the transmitting end of the communication channel are either known in advance or are extracted from the original signal itself and do not change during the entire communication session. At the receiving end, these transformation parameters are either also known in advance or are extracted from the received modified signal with the purpose of restoring the unintelligible RS according to the same previously known law.

Note that it is not always necessary to restore the original signal in the form it was originally in at the receiving end. For example, this applies to the RS synthesized from a graphic image restored from a closed sonogram image without taking into account the original values ​​of the phase spectral components. Then the waveforms (oscillograms) of the original and restored RS will be different, but their intelligibility and sound will be absolutely identical. Here, the properties of human auditory perception are fully manifested, which weakly depends on the phase relationships of the simplest narrow-band components of a complex sound signal. Hence the conclusion: if the images of correctly calculated dynamic spectrograms of different acoustic signals are similar, then they will sound (be perceived by ear) in the same way.

The main task that is solved in technical speech closure using the described approach is such a change in the Pirogov phonetic function of the original RS, in which the modified speech will be absolutely unintelligible. This task is solved by dynamically changing the envelope of the amplitude spectrum of the RS, that is, ultimately, by modifying its formant structure. The methods described in [11] are quite applicable for assessing the final intelligibility of the closed and restored RS.

Examples of practical implementation of methods of such modification of RS include the simplest inversion of speech in the channel band of the tonal frequency. A more labor-intensive and previously unheard of procedure in practical applications of inversion of the cyclic shift of the spectrum envelope while maintaining the unchanged harmonic structure of the original RS is shown in Fig. 4. More complex laws of mutually inverse transformations of RS with the aim of achieving its intelligibility are also known (see Fig. 5).

It is also possible to implement combined methods of technical closure of the RS: modification of the RS with simultaneous imposition of interference. An example is spectrum inversion plus quasi-harmonic interference.

Above — sonogram of the original fragment of speech;
In the center — sonogram of the speech signal with inversion by spectrum;
Below — sonogram of the speech signal with inversion and cyclic shift of the envelope of the speech spectrum while maintaining its harmonic structure.

Fig. 4. Technical closure options

Above – sonogram of the original speech fragment;
In the center – sonogram of the speech signal with modification of the spectrum envelope by means of its dynamic twist;
Below is a sonogram of the central fragment of the speech signal, modified by the “chess” frequency-time function.

Fig. 5. Technical closure options (continued)

It should be noted that in some publications speech maskers, such as spectrum inverters and the like, are classified as the simplest type of analog static scramblers, in which the speech conversion «key» does not change during the entire session or during a group of communication sessions. Despite the fact that the described technologies for modifying the Pirogov phonetic function and the spectrum envelope associated with it can be used to create dynamic scramblers, when the conversion key changes during a session from one frequency-time section of the signal to another, nevertheless, in our opinion, static scramblers should still be classified as technical closure tools and considered as speech maskers. That is, when we talk about closing speech messages, we mean the use of constantly acting specific speech conversion laws that introduce incomprehensibility into the RS and are implemented in speech maskers, and when we talk about encrypting the RS, we mean the use of cryptographic algorithms.

Indeed, there are many disputes about the division of methods and means of cryptographic encryption and technical speech suppression. We believe that the technologies of technical suppression of the RS, based on the exploitation of the properties of human auditory perception, have the right to independent consideration within the framework of the general problem of information protection, outside the technologies of cryptographic protection, and, moreover, not as a certain subtype of it. Here it is appropriate to draw analogies with the methods and means of information compression, which, as is known, although in most cases they are a stage of preliminary processing of cryptographic encryption of data, nevertheless, in many applications they imply their separate use.

Secret transmission of speech messages

At present, OBRS measures can be aimed not only at preventing unauthorized interception of protected speech information, but also at concealing the very fact of its transmission, by using for these purposes standard technical means (STM), conventional, traditional information exchange protocols and publicly available communication channels (POC).

In recent years, this area of ​​information security in computer telecommunication systems, called “stegology” (sometimes “stealthology”), has been actively developing throughout the world.

Recently, such part of stegology as “steganography” has become especially popular. It is used in the field of concealing confidential information in graphic images transmitted over computer networks. At the same time, the progress achieved in the field of developing devices for transmitting voice messages, as well as in computer technology, opens up new possibilities for both the covert transmission of confidential information in analog and digital audio signals and speech, and for covert transmission in information containers of various kinds, based on the use of dynamically developing technologies of multimedia, computer and cellular telephony [6]. This direction of digital technologies in the field of protecting confidential information, covertly present inside or on top of an openly transmitted audio signal, is now commonly called “steganophony”.

Currently, computer steganophony methods based on the use of natural noises, which contain digital arrays obtained by standard methods of conversion from analog acoustic and video signals, are widely used. These noises are quantization errors and cannot be completely eliminated. The use of noise bits to transmit additional confidential information allows creating a hidden data transmission channel. The least significant digits of the sample values, which are noise from the point of view of measurement accuracy and carry the least amount of information contained in the sample, are usually considered as noise bits. Such bits are usually called least significant bits (LSB) [4,9].

One of the most common methods of steganophonic concealment of confidential information is the method based on the use of the NZB of audio (and/or any other multimedia) data. In [9] it was shown that such natural flows of NZB audio files are not random and have a certain grouping of consecutive zeros and ones, which is violated when additional information is introduced. Certain statistical criteria were developed to detect the fact of concealment of a confidential information message in the NZB of audio signals.

The statistical analysis of audio data conducted in [4, 9] allowed us to identify a number of significant properties that affect the security of confidential data and, accordingly, the security of such data using noise bits. Among such properties, the following should be highlighted:

• non-uniformity of sample sequences;
• the presence of certain dependencies between bits in samples;
• the presence of certain dependencies between the samples themselves;
• inequiprobability of conditional distributions in a sample sequence;
• the presence of long series of identical bits;
• the presence of a correlation between the LSB and the most significant bits.

These properties are observed to varying degrees in most audio files and can be used to construct various statistical criteria that determine the fact of concealment of information in the NZB. That is why such methods of computer steganophony are used in practice less and less often.

Today, the following requirements can be proposed for concealment of confidential speech information (CSI) and the placement of steganophonic markers in signals, arrays and data formats of various natures:

• the perception of signals and data with embedded CRI should be practically indistinguishable from the perception of the original, “open” message contained in the given signal or array;
• confidential speech data transmitted via the OCS, camouflaged by various signals or implicitly contained in their parameters, should not be easily detected in these carrier signals by the widely used methods and technical means of analysis currently available;
• in a number of applications, the setting and detection of steganophonic markers should not depend on the synchronization of these processes and on the presence of any standards;
• special methods for setting and detecting steganophonic markers should be implemented on the basis of standard computing technology or special software and hardware based on it;
• it must be possible to embed and detect signs of authenticity in an acoustic (speech) signal that appear when it is illegally copied or modified, regardless of the type of presentation and transmission of this signal (analog or digital);
• it must be possible to conceal the CRI in data arrays, regardless of the type of information presented in them.

We will give some examples of the use of the proposed approach to sound processing through the processing of its graphic image in computer steganophony tasks, described in detail in [6].

Thus, it is possible to transmit and store a speech signal in another audio or video signal in a way that is imperceptible to the ear, and also to combine steganophony technologies with steganography technologies, “dissolving” images of dynamic acoustic spectrograms in specified images – “containers”, with their subsequent development, synthesis and pronunciation at the receiving end of the OCS.

Sonogram images can be used to transmit and store speech on paper media as stegomarkers. When implementing «speech signature» technologies, related to the protected document in meaning and content approximately the same as an electronic digital signature, 2 to 4 minutes of speech of telephone quality sound in the form of various patterned drawings can be applied to a standard sheet of paper. In this case, the authenticity of the document can be established not only in the presence of appropriate signatures and seals, but also by the information contained in the «speech signature», by scanning, synthesizing and voicing which you can hear the key points of the document content, voiced by the responsible person. Mismatch between the voiced information and the information contained in the document indicates its falsification. It is practically impossible to forge a «speech seal» or «speech signature». Note that such a cheap “speech signature” technology can be implemented on standard office equipment: a computer with a sound card plus a printer and scanner.

With the help of the proposed approach to processing audio signals, it is possible to implement a large number of the most diverse methods of computer steganophony, exclusive for each specific task of covert transmission of CRI or marking with sounds or speech.

It should be noted that the considered methods of setting steganophonic markers and covert transmission of the CRI in most cases will not require synchronization of the processes of their introduction — detection or availability of comparison standards, as a result of which they can be used in communication channels not only when receiving and transmitting signals and data, but also in storage modes. Therefore, they can find application in analog and digital answering machines, standard voice mail systems, computer telephony, etc., as well as when transferring «steganophonically» processed records on audio, video cassettes and diskettes.

If we talk about the transmission of confidential information in sounds and speech, then the conducted assessments of the permissible values ​​of the speed of covert transmission of confidential information in audio signals have shown that today these values ​​do not exceed 100 bits/s. For now, these are the maximum values ​​that can be achieved with various methods of concealing confidential information in speech or acoustic signals by means of appropriate processing of graphic images of their dynamic spectrograms. Nevertheless, it can be assumed that such speeds will most likely be quite sufficient for the prompt transmission of important, confidential messages during speech communication between two subscribers over a telephone line or by means of receiving and transmitting audio cassettes containing audio signals — “containers” with an information bookmark, as well as other applications. Indeed, at such speeds, approximately three pages of text and about ten black-and-white photographs can be covertly transmitted in one minute of speech signal during telephone conversations.

It is possible that new methods of masked transmission of confidential information in acoustic signals – “containers” based on the proposed methods of processing graphic images of sound will appear, as a result of which the information efficiency of computer systems for covert data transmission can increase significantly.

Thus, based on the above, it can be assumed that in the future, one of the promising areas of protecting speech messages in communication channels and dedicated areas can be considered the creation and development of computerized speech masking systems along with or in combination with traditional technologies for the semantic protection of speech messages, namely, the encryption of speech signals based on cryptographic algorithms.

The choice of specific methods and means of speech masking as one of the types of semantic protection of speech messages will depend on the practical requirements for the speech protection system and the technical characteristics of the speech information transmission channel.

Literature

1. Abalmazov E.I. Methods and engineering and technical means of counteracting information threats//Grotek, 1997, 248 p.
2. Abalmazov E.I. New technology for protecting telephone conversations//Special technology, No. 1, 1998, pp. 4–8.
3. Agapiyev A.N., Milashenko V.I. Identification of users of computing systems based on speech technologies//Confident, No. 6, 1999, pp. 37–45.
4. Ukov V.S., Romantsov A.P. Evaluation of the level of secrecy of multimedia steganographic channels for storing and transmitting information//Special technology, No. 6, 1999, pp. 52–59.
5. Dvoryankin S.V. Computer technologies for protecting speech messages in telecommunication channels. — M.: MTUCI, 1999, 52 p.
6. Dvoryankin S.V. Covert transmission of confidential information in audio signals and speech//“BDI”, No. 2 (30), 2000, pp. 12 – 16.
7. Dvoryankin S.V. Digital processing of images of dynamic spectrograms of audio signals in problems of ensuring the security of speech communication//Special equipment, No. 3, 2000, pp. 37 – 45.
8. Dvoryankin S.V., Devochkin D.V. Methods of closing speech signals in telephone channels//Information protection. Confidential, No. 5, 1995, pp. 45 – 54.
9. Dvoryankin S.V., Romantsov A.P. Statistical method of steganophonic analysis of audio files.//Abstracts of reports of the IV interregional scientific and technical seminar «Application of plastic cards and information protection». — M.: MNTORES im. A.S. Popov, 1999, pp. 51 — 54.
10. Zolotarev V.I. New solution for protecting confidential negotiations//Special equipment, No. 5, 1999, pp. 26 — 30.
11. Kalintsev Yu.K. Speech intelligibility in digital vocoders. — M.: Radio and communication, 1991, 140 p.
12. Kravchenko V.B. Protection of speech information in communication channels//Special equipment, No. 4, 1999, pp. 2–9, No. 5, 1999, pp. 2–11.
13. Petrakov A.V., Lagutin V.S. Protection of subscriber traffic – Moscow: Radio and communication, 2001, 504 p.
14. Sudarev I.V. Cryptographic protection of telephone messages//Special equipment, No. 2, 1998, pp. 47–55.
15. Khorev A.A., Makarov Yu.K. On the assessment of the effectiveness of acoustic (speech) information protection//Special equipment, No. 5, 2000, p. 46 – 56.