Improving the methodological principles for assessing the security of premises from speech information leakage.

Improving the methodological principles for assessing the security of premises from speech information leakage..

Improving the methodological principles for assessing the security of premises from speech information leakage.

Candidate of Technical Sciences
V. L. Kargashin
Special Equipment, No. 6, 2001

When carrying out practical work on protecting vibroacoustic channels of speech information leakage from premises, much attention should be paid to the validity of the requirements for the integral level of specially formed masking interference. As a rule, guaranteed protection cannot be ensured by implementing only passive protection measures, i.e. by strengthening sound insulation and vibration insulation of structures, introducing sound absorption and vibration absorption into speech signal leakage paths. In such situations, it is necessary to use active protection measures based on the creation of additional vibration and acoustic interference in information leakage channels. In this case, secondary acoustic noise is created in the protected and adjacent premises, which can be classified as a factor interfering with the normal work of personnel in the premises [1]. Despite the fact that the appearance of such noise is a justified payment for protection, abnormal conditions for the work of employees are objectively created from the point of view of industrial sanitation. Consequently, reducing the integral level of side acoustic noises during active protection measures in vibroacoustic channels is an important task of protective measures, since with a high level of side acoustic noises, situations are not excluded when active protection will be used only in exceptional cases from the standpoint of the interests of the protected person, reducing the effectiveness of the measures taken.

The indicator of vibroacoustic leakage channel protection based on estimated speech intelligibility proposed in [2] provides significantly greater opportunities for developing requirements for passive and active protection measures than the traditional indicator based on the signal-to-noise ratio. This is due to the fact that estimated speech intelligibility is a functional whose value can be optimized based on various criteria acceptable for practical reasons.

The functional is a certain weighted sum of an argument, which for formant intelligibility has the following form [3]:

,

where  — are the lower and upper frequency limits of the speech range,  — distribution of formants over the frequency range, therefore,  as the distribution density,  — the level of formant sensation, which is determined by the signal-to-noise ratio,  — the formant perception coefficient, which does not depend on frequency, but only on the signal-to-noise ratio.

The level of formant perception is determined by the expression:

,

where  — speech signal spectrum,  — speech and formant spectrum ratio,  — interference spectrum.

In practical tasks, the level of formant perception is calculated depending on the values ​​of signals and interference in dB, which does not change the ability to optimize the functionality for any of the parameters.

Thus, formant intelligibility is an integral of the frequency function within limited limits, which allows us to solve problems of intelligibility optimization according to various criteria. The most understandable practical problem that can be set when carrying out protective measures is formulated as a search for the interference spectrum , ensuring an intelligibility value no greater than the specified standard  with a minimum integral interference power in the entire frequency range . In other words, a given level of speech information protection in terms of formant speech intelligibility can be implemented by interference with different spectrums and power levels, including those with a minimum integral level. The ability to implement protection using interference with a minimum level allows for minimal side effects of special masking interference on a person, i.e. to maximize the comfort of work combined with safety. Moreover, an optimal solution means that for a certain specific case this solution is the only one. The task itself can be expanded, provided that the frequency characteristics of speech signal attenuation and interference to the point of installation of the interference emitter and back to the protected or adjacent premises are taken into account. For passive protection purposes, optimization is possible, for example, by the indicator of ensuring protection with minimal integral attenuation in the speech signal frequency band.

It is obvious that the complete solution of the problem by the methods of variational calculus is quite complicated if we operate with the spectra of signals and interference, as well as the calculation of intelligibility by the integral formula. In this regard, in practice, the integral is replaced by a weighted sum, and all calculations are carried out in standard frequency bands. Note that in [3] it was proposed to use 20 frequency bands, in each of which the condition is satisfied, where  — upper and lower bounds i-th frequency range. Since the distribution of formants over the frequency range is uneven, such equal-articulation distribution of intelligibility leads to frequency band boundaries that differ significantly from the standard ones. As a rule, all measurements in acoustics are carried out in octave or third-octave frequency bands, but in [2] it is proposed, on the contrary, to select standard frequency boundaries to obtain different formant weights over the speech spectrum. Thus, the calculation of formant speech intelligibility is carried out according to the following formula:

,

where  — weighted contribution of the corresponding frequency band to intelligibility, and ,  — the formant perception coefficient, which can be expressed as a function of the signal and noise levels measured in frequency bands, and these levels can be measured in dB, as is customary in acoustics,  — standard frequency band number,  — number of standard frequency bands taken into account in calculations (5 in the octave analysis recommended in [2]).

Since the functional or its approximation are continuous functions of the signal and interference levels, the standard value of the protection index, strictly speaking, can be quite arbitrary within the approximations of the functions used. The method proposed in [2] is analytically complex enough to be used even for a limited number of frequency bands. Since the development of standards is a formal procedure and the theoretical and experimental foundations of the emergence of certain relationships between the measured values ​​are not essential to the consumer, simpler methods for calculating formant intelligibility are possible, which formally lead to the achievement of protection requirements, but due to the simplicity of the calculation, they allow obtaining visual methods for solving the protection problem, including its optimization.

Let us consider from this standpoint the calculation method of a similar parameter adopted in the USA [4]. In this case, formant intelligibility is called the articulation index and the formalization of the calculation is so great that it is not possible to determine the visual physical foundations of the accepted calculation methods. The standard indicator is assessed based on the results of signal and interference measurements in one-third octave frequency bands, but octave analysis can also be used.

The essence of the method is as follows. The speech signal level is taken as a standard, 12 dB higher than the average speech level. The signal/interference ratio values ​​are measured in octave or one-third octave frequency bands DIFFi in dB. If DIFFi ? 30, then DIFFi = 30 dB, if DIFFi ? 0, then DIFFi = 0. The resulting differences are multiplied by the frequency-dependent weighting coefficients WFi, which yields the value of the articulation index by frequency bands:

.

The total articulation index is calculated using the formula:

.

The obtained value is an indicator that is functionally related to speech intelligibility and can be used to standardize various tasks, in particular to assess the quality of sound paths.

The values ​​of speech signal levels at a distance of 1 meter from the source and weighting factors are given in Table 1. For comparison, similar indicators for the method [2] are also given there, and the speech levels according to the US standard are reduced by 12 dB for comparison. Table 1 also shows the relative weighting contributions of octave frequency bands calculated by the formula , which are essentially equivalent to the weighting contributions of formants.

Table 1

A significant difference in the spectra of average speech is striking, which is most likely caused not by real differences in the spectra, but by the additions embedded in them, corresponding to the domestic concept of the formant parameter of speech. Most interestingly, the sum of the relative coefficients for the American standard is 0.999, that is, practically 1, and for the domestic one it is 0.91. The imprecise equality of this sum 1 for both standards is puzzling, since it fundamentally does not allow obtaining full speech intelligibility for a significant excess of signal over noise. The loss of 9% intelligibility in the domestic method is caused by the fact that the octave band 8000 is not included in the calculationHz. Apparently, this frequency band in the American standard is taken into account by formal additional weighting of all coefficients.