RU2284585C1 - Method for measuring speech intelligibility - Google Patents

Abstract

FIELD: computation engineering.

SUBSTANCE: method involves carrying out object-oriented measurement tract correction. To carry out it, some test signal frequencies are transmitted of given level in turn. Real level of the acoustic signals is measured at each frequency at their transmission place. The acoustic signal levels are compared to given levels at the same frequencies, measured acoustic signals levels deviations from the given ones are measured and frequency characteristic of the measurement tract is adjusted to compensate the difference between the measured and given acoustic signal levels at their transmission place.

EFFECT: high accuracy and reliability of measurements.

2 dwg

Description

The field of technology.

The invention relates to speech recognition, specifically to measuring speech intelligibility in extremely adverse conditions, in noise, at low signal-to-noise ratios, and is intended mainly to determine the degree of protection of premises against unauthorized leakage of acoustic speech information in real conditions.

The prior art.

Various methods are known for measuring speech intelligibility. The simplest, but also the least accurate, is to evaluate intelligibility by the ratio of a given integral level of an acoustic signal and a measured integral noise level at a listening point. More accurate methods are expert listening at the place of reception of acoustic signals composed of specially selected phrases, words or syllables (programs) that are emitted at the place of speech generation, followed by processing the results to determine intelligibility (“Acoustics”, reference edited by M.A. .Sapozhkova, - M .: Radio and communications, 1989; GOST B 20775-75. "Voice transmission over communication paths equipped with security equipment. Speech intelligibility requirements. Measurement method"; ON Novoselov, A.F. Fomin "Fundamentals of the theory and calculation of information KSR Control Systems ", M .: Engineering, 1991). In particular, in the indicated works of P. M. Polkovsky, A. D. Tkachenko “Electroacoustic paths with feedback” and GOST 20775-75 describe a similar measurement method, the apparatus used and the mathematical processing methods. The method is technically simple, but not reproducible and accurate in the conditions of weak signals, i.e. with legibility of speech syllables less than 25%, because it depends, in particular, on the intelligence of experts, and most importantly, it is time-consuming and requires the participation of a large number of experts.

A more objective method is based on the sequential emission of individual tonal components of the audible spectrum at the place of transmission and the determination by any experts of the threshold of their audibility at the place of reception, and in fact at the place of reception there can only be microphones connected through amplifiers to the head phones of experts (NB Pokrovsky "Calculation and measurement of speech intelligibility", - M .: Communication, 1962). The level of acoustic radiation of each component is increased to the threshold of hearing by its expert. The fixed radiation levels are mathematically processed taking into account the human hearing curve and thus determine the minimum level of the speech signal to obtain a given speech intelligibility at the reception site. Naturally, the obtained information can also be used to determine the maximum level of the acoustic signal at which speech will be guaranteed to be illegible at a given place of reception, i.e. a predetermined level of protection against leakage of acoustic information is provided. However, this method is laborious, since it also requires the participation of many experts and time. In addition, intelligibility is assessed not at the real level of acoustic signals, but at those that correspond to the threshold of audibility in a given place. This is often unacceptable, since, as you know, spurious channels of transmission of acoustic information (walls, floors, windows, furniture, doors, etc.) usually unevenly weaken the frequency components of speech and the degree of protection of information from leakage, calculated from the results of the described measurements, can not match the real ones.

Currently, the most common hardware methods based on the tonal method of speech intelligibility (for example, P.M. Polkovsky, A.D. Tkachenko "Electroacoustic paths with feedback", GOST B 20775-75, Utility model No. 27259). They provide for the measurement of intelligibility at any signal to noise ratio exclusively by technical means, without the use of expert listening. The most modern device according to the patent for utility model No. 27259, publ. 01/10/2003. It is closest to the claimed method according to most of the essential features and is therefore selected as a prototype.

The following is a detailed description of the prototype method.

The method involves placing the source of acoustic test signals at the location of the source of the speech signal, and the receiver at the location of the receiver of an unauthorized leak of acoustic information. A sequence of tonal sound frequencies of a given level is emitted with pauses between them, and the radiation frequencies are distributed over the middle frequencies of the N bands into which the audible speech spectrum is divided. During the emission of each frequency by the receiver of the test signal through a band-pass filter tuned to this frequency, the total acoustic signal formed at the place of reception of this frequency and acoustic noise in this band is received and measured. During a pause, the receiver receives and measures only the noise signal in this band. According to the measurement results, the power values of the received signal of this tonal frequency are determined based on the independent linear addition of the signal and noise powers, and the signal-to-noise ratios at this tonal frequency are determined. The procedure is then repeated at the next tonal frequency, then at the next until the signal-to-noise ratios at all tonal frequencies of the test signal are determined. For certain signal-to-noise ratios at all tonal frequencies of the test signal, speech intelligibility of a signal against real noise is calculated based on well-known techniques (see, for example, an article by A.A. Kolesnikov et al. "Correlation theory of speech intelligibility" in the journal " Questions of radio electronics ", series" General issues of radio electronics ", 1995, issue 1, pp. 3-9; NB Pokrovsky" Calculation and measurement of speech intelligibility ", - M .: Communication, 1962).

In other words, the essence of the method, like most other hardware methods for measuring intelligibility, consists in sequentially emitting test N tonal acoustic vibrations in the speech frequency range, receiving the received signals at a given receiving point, N-band measurement of signal-to-noise ratios at all N frequencies of test signals and the subsequent calculation of them the coefficient of speech intelligibility - according to well-known methods. The essence of the method is illustrated in figure 1 - an example of a simplified structural diagram of the implementation of the method, where 1 is an N-frequency test signal generator, 2 is an acoustic emitter (loudspeaker), 3 is an acoustic signal receiver (microphone), 4 is an N-band signal / signal ratio meter noise, 5 - calculator of the level of speech intelligibility.

The device consists of a series-connected generator of test signals 1, a loudspeaker 2 located at the actual location of the sound source, a microphone 3 located at the listening position, an N-band meter 4 signal-to-noise ratios at each of the N frequencies of the test signals and level 5 calculator speech intelligibility.

The device operates as follows. After placing the speaker 2 and the microphone 3 in the indicated places, the device is turned on. Generator 1 generates a test acoustic signal of a given level, consisting of a sequence of N frequencies spaced apart from the middle frequencies of N bands into which the audible speech spectrum is divided. Loudspeaker 2 emits the indicated signals. Microphone 3 converts the received signals into electrical signals; an N-band signal-to-noise ratio meter 4 determines the signal-to-noise ratio at each frequency of the test signal. Thus, at the input of the calculator 5 of the intelligibility level, the measured signal-to-noise ratios for all N radiated frequencies are received in turn. The intelligibility level calculator 5 for the measured signal-to-noise ratios calculates speech intelligibility. The calculation is based on known relationships and theories, the essence of which is that first, the intelligibility factors at each frequency of the test signal are determined from the above measured signal-to-noise ratios. They determine the average (arithmetic mean) value of the intelligibility factor. Then, according to a certain arithmetic mean value of the intelligibility coefficient and the known curve of the dependence of verbal intelligibility on the value of the intelligibility coefficient, verbal intelligibility of speech is determined (for more details, see, for example, an article by V.K.Zheleznyak et al. Some methodological approaches to assessing the effectiveness of protecting speech information in g. " Special equipment "No. 4, 2000, pp. 39-45).

The described method is widely used for measuring speech intelligibility, i.e. to determine the quality of listening. However, at present, the fight against eavesdropping, i.e. reduction of intelligibility in places of possible information leakage, prevention of unauthorized listening. In this case (in contrast to the task of determining the quality of listening), the signal-to-noise ratio is small. The application of the described method to control these parameters is not always acceptable due to its lack of accuracy. The total acoustic leakage signal can also be generated in non-linear ways, therefore, to ensure the reliability of the measurement, it is necessary to produce it at the real level of the signal at the transmission site, which is very small at the receiving point. It is also known that the level of the real total acoustic signal at the transmission point (the location of the source of the protected information) depends not only on the radiation power, but also on the secondary acoustic signals reflected from the structural elements of the premises (walls, ceiling, window panes, furniture), from reverberation noise, from vibration, from magnetic, from other ways, for example, artificial acoustic noise created by technology, engineering structures, human activities (L.A. Lependin "Akusti ka ", Moscow: Higher School, 1978, chap. VII, pp. 346-370).

This dependence is non-linear and frequency-dependent, which significantly reduces the reliability of measurements of the intelligibility factor. Moreover, it is fundamentally impossible to exclude the indicated deviations of the actual level of acoustic radiation from the given in a previously unexplored room. In addition, the acoustic characteristics of even a pre-examined room at the radiation site can noticeably change with changes in the atmosphere, temperature, day and night, the presence and movement of equipment, furniture, people, and much more. The listed secondary acoustic signals and interference significantly affect the results of measuring speech intelligibility, reducing their accuracy and reliability.

This is a big disadvantage of the described method.

The objective of the invention.

The present invention is to improve the accuracy and reliability of measuring speech intelligibility.

The solution to the problem is the essence of the invention.

The problem is solved in that in the known method for measuring speech intelligibility, which involves placing the source of acoustic test signals at the location of the source of the speech signal, and the receiver at the location of the receiver of unauthorized leakage of acoustic information, the serial radiation of the N-frequency test acoustic signal of a given level, and radiation frequencies are distributed at medium frequencies of N bands into which the audible spectrum of speech is divided, receiving and measuring acoustically signals produced at the site of reception of all N at the test signal frequency, determining from measurements of the signal / noise ratio at each frequency of the test signal, the subsequent calculation of the level of speech intelligibility in certain signal / noise ratio,

Significant changes and additions have been made, namely:

- before the emission of the test acoustic signal, an object correction of the frequency response of the measurement path is performed, for which

- alternately emit all N tonal frequencies of the test acoustic signal of a given level and at each frequency the actual level of acoustic signals is measured at the place of their emission,

- compare the measured levels of acoustic signals with predetermined levels at the same frequencies and determine the deviation of the measured levels of acoustic signals from the set,

- adjust the frequency response of the measurement path to compensate for the difference between the given and measured levels of acoustic test signals in the place of their real radiation.

Disclosure of the invention.

The essence of the invention is to provide measurements at each frequency precisely at given levels of acoustic signals, since possible deviations from specified levels due to reflections, resonances, nonlinearity of the emitter, etc. in advance, before the start of measurements are taken into account, measured and compensated at the place of reception (listening). We called this operation a general term - object correction, since it eliminates, mainly, specific measurement distortions associated with the features of objects (rooms).

An example of a simplified structural diagram of a device that implements the described method is presented in figure 2, where the following notation is adopted: 1 - generator of an N-frequency test signal, 2 - acoustic emitter (loudspeaker), 3 - receiver of acoustic signals (microphone), 4 - N -band signal-to-noise ratio meter, 5 - speech intelligibility level calculator, 6 - acoustic signal level meter at the radiation site, 7 - comparison device, 8 - adjustable frequency response corrector.

The device consists of a series-connected generator 1 of an N-frequency test signal, a loudspeaker 2 located at the actual location of the sound source, a microphone 3 located at the listening position, an adjustable equalizer 8 frequency response, an N-band meter 4 signal-to-noise ratios and a calculator 5 levels of speech intelligibility. Directly in front of the acoustic emitter 2 is an acoustic signal level meter 6, the output of which is connected to a comparison device 7. The operation of the device is synchronized by a clock synchronizer.

The method is implemented as follows.

Place the source of the acoustic test signal (loudspeaker 2) at the point of placement of the real source of the acoustic signal, and the receiver of the acoustic signal (microphone 3) at the point of possible placement of the receiver of acoustic information. Before the beginning of speech intelligibility measurements, an object correction of the frequency characteristic of the measurement path is performed, for which, alternately, the tonal frequencies of the test signal of a given level are generated by the generator 1 and the loudspeaker 2, and the actual levels of the acoustic signal in the place of their emission are measured by meter 6, and the measured levels of acoustic signals with preset levels at the same frequencies, determine the deviation of the measured levels from preset and adjust The corrector 8 corrects the frequency response of the intelligibility measurement path to compensate for the difference between the given and measured levels of the test signals in the place of their real radiation.

Then, the measurement of intelligibility begins, for which the test acoustic signals of a given level are generated by the generator 1 and emit by the emitter 2, which are, most often, a sequence of tonal frequencies distributed over the middle frequencies of the N bands into which the audible spectrum is divided, with pauses between the frequencies. The receiver 3 receives the received acoustic signals at the measurement point during emissions and pauses, makes an object correction of the frequency response of the measurement path by corrector 8. Then, divide them with narrow-band filters 4 into the frequencies corresponding to the frequencies of the test signal, measure the received signal levels, and then determine signal-to-noise ratios at each frequency of the test signal and compute 5 intelligibility of speech against real noise based on the correlation theory of analysis speech agility (L.A. Lependin "Acoustics", M .: Higher School, 1978, chap. VII, pp. 346-370), which is the development of formant speech intelligibility (see, for example, NB Pokrovsky "Calculation and measurement speech intelligibility ", - M.: Communication, 1962 and utility model patent No. 27259, publ. 10.01.2003).

The main difference between the claimed method and the prototype is the introduction of the object correction of the frequency response of the measurement path before measuring the intelligibility of the operation, for which the frequencies of the test signal of a given level are alternately emitted, the actual radiation levels in the place of their radiation are measured at each frequency, the measured levels of acoustic signals are compared with the given levels at the same frequencies, determine the deviation of the measured levels from the set and adjust the frequency response of the measurement path p zborchivosti to compensate for the difference between the set and measured levels of acoustic test signals in place of their actual emissions.

This method is unknown to us from available sources of information.

The combination of these differences is a creative achievement of the inventors, it is not obvious to a specialist.

Industrial applicability.

The claimed method is easily feasible both schematically and programmatically by circuit means well mastered by the industry, which follows from the above description of an example implementation.

In particular, the measurement of the actual level of the acoustic signal can be performed with a level meter or sound level meter, comparison with a given level can be performed visually (by comparison with a standard or table) or a subtractor. An adjustable frequency response corrector may be, in particular, a multi-band tone control. A device that implements the described method was built at the Informakustika Federal State Unitary Enterprise, it has successfully passed metrological and certification tests and is already used in measuring speech intelligibility. However, no practical difficulties associated with the use of the invention were encountered.

The real advantages over the prototype are to increase the reliability and accuracy of measuring the coefficient of speech intelligibility. In particular, the guaranteed lower level of determining the signal-to-noise ratio is 10 dB lower than the zero level of the ear auditory threshold (2 × 10 ^-5 Pa) and the intelligibility is reduced by 1.5 times, the measurement error is more than halved due to the elimination of the influence of secondary sources of acoustic signals. The reliability of determining speech intelligibility under conditions of a weak signal and high-level noise allows you to guarantee the absence of information leakage channels, since intelligibility is determined below the nominal value specified by regulatory and methodological documents.

Thus, in our opinion, the claimed technical solution meets all the criteria for inventions, it is new, non-obvious and industrially applicable.

Claims (1)

A method for measuring speech intelligibility, comprising placing a source of acoustic test signals of a given power at a point of placement of a source of a speech signal, and a receiver at a point of placement of a receiver of unauthorized leakage of acoustic information, sequential radiation of N-frequencies of test acoustic signals of a given level, the radiation frequencies being distributed over medium frequencies N bands into which the audible spectrum of speech is divided, the reception and measurement of acoustic signals formed at the volume at each frequency of the test signal, determination of signal-to-noise ratios at each frequency of the test signal from the results of measurements, and calculation of the speech intelligibility level from the signal-to-noise ratios defined above, characterized in that before the emission of the test acoustic signals an object correction of the measurement frequency response is performed, for which, in turn, the individual frequencies of the test signal of a given level radiate and at each frequency the actual level of acoustic signals is measured per month their emissions, compare the measured levels of acoustic signals with the given levels at the same frequencies, determine the deviation of the measured levels of acoustic signals from the set and adjust the frequency response of the measurement path to compensate for the difference between the set and measured levels of test signals in the place of their real radiation.

Images