JP2005077341A - Sound listening discrimination device and sound listening discrimination method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To collect a generated sound from an observation object or a sound propagating in a solid or liquid, and to grasp a continuous change relative to the phenomenon. <P>SOLUTION: A sound propagating in a medium which is the observation object is collected by a sound pressure value or a frequency width at prescribed time intervals as long as a prescribed time. Then, a sound data block in each prescribed band is produced, and it is determined whether or not the sound data block is converted into observation object various quantities through a quantization software via calibration capable of evaluating the phenomenon or the block belongs to the pertinent object phenomenon through an encoding software, to thereby grasp the state of the observation object phenomenon. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a sound identification device and a sound identification method, and in particular, collects sound propagating through a solid or liquid as sound data of sound pressure or frequency level, and quantifies and encodes sound data chunks. The present invention relates to a sound identification device and a sound identification method that can discriminate an event.

  Conventionally, in steel structures and concrete structures such as aircraft, nuclear power plants, large civil engineering facilities, etc., to detect cracks, crack detection, fatigue check, etc. of components, internally generated sound is measured to identify the occurrence phenomenon A measurement technique called AE (Acoustic Emission) is used. In this AE, a detection element that detects a frequency in an ultrasonic range higher than an audible band is generally used to collect generated sound.

  On the other hand, a lot of measurement and analysis has been carried out in the field of earthquake-resistant engineering and mechanical vibration engineering, and in the field of practical use for vibration and vibration area sound for audible band sound in relation to human hearing. For example, there is known a mechanical operating sound acquisition device or the like that determines the quality of operating characteristics of a rotating shaft based on the operating sound of a shaft during rotation of an automobile engine, a motor, or the like (see Patent Document 1).

Japanese Patent Application Laid-Open No. 11-248527.

  By the way, the sound generated by the operation of the machine such as the squeak noise of the above-mentioned mechanical equipment and the rotation sound of the motor can be understood from the situation of the sound propagating through the solid / liquid. In addition, sounds generated by static structures (moving load sound acting on structures), sounds generated by natural events (wind, wave, liquid sand, landslides, etc.), animal activity There are sounds (for example, fish swimming sounds, human pulse sounds, microbial fermentation sounds, etc.), and various sounds generated in daily life.

  Many of these sounds that propagate in solids and liquids propagate faster and farther than in air, but are greatly attenuated when propagating from solids and liquids into air. Therefore, it is necessary to collect data in the medium. In addition, the frequency of sound generated in many of these events is excellent from the audible band to 100 Hz or less. At present, no measurement / analysis technique has been established for easily realizing continuous analysis of 5 to 100 Hz and low-frequency sounds, which are intermediate between these audible band sounds and vibration domain sounds. Therefore, the object of the present invention is to solve the above-mentioned problems of the prior art, and various generated sounds using solid and liquid media can be collected directly and continuously and become the source of the generated sounds. An object of the present invention is to provide a sound identification device and a sound identification method capable of identifying an event.

  In order to achieve the above-mentioned object, the present invention is realized by a sound identification technique. This audio identification technology is based on the assumption that the generated sound is composed of the sound intensity (sound pressure) and sound pitch (frequency) distribution and its time-varying characteristics. This is a technique for identifying, specifying, etc., an observation target event of a generated sound by measuring, quantifying and encoding the data. Specifically, it is possible to determine changes in the state of target natural objects and artificial structures, detection of abnormalities, detection of damage, monitoring of progress of deterioration due to fatigue, etc., prediction of precursors such as collapse / collapse. Natural objects and artificial structures to be observed are classified according to medium, solid (concrete structure, steel structure facility, ground, etc.), liquid (water, river water, human waste, oil, etc.), and moisture in the middle A high content of woody, herbaceous, living body (human body, animal) can be assumed.

  The discrimination by the quantification of the data is performed by quantifying the sound in a form proportional to the observation target event and comparing it with the calibration for converting into various quantities of the target. On the other hand, in the discrimination by encoding, the sound data is arranged and encoded so that the difference between various events is easily determined, and it is determined which event belongs to the event to be distinguished.

  That is, the present invention collects sound propagated through a medium to be observed and collects the sound propagated through the medium microphone at a predetermined time interval over a predetermined time with a sound pressure value or a frequency width. Measurement control calculation means for creating sound data chunks for each, quantification calculation means for quantifying the sound data chunks and converting them into various quantities to be observed through calibration capable of evaluating events, and encoding the sound data chunks And coding operation means for determining whether or not the target event belongs, and the state of the observation target event is grasped.

  As a method invention using this audio identification device, the sound propagated in the observation target medium is collected at a predetermined time interval over a predetermined time with a sound pressure value or a frequency width, and a sound data block for each predetermined band is created. Quantify or encode the sound data chunk and convert the event into various quantities to be observed through calibration that can evaluate the event, or determine whether it belongs to the target event, and grasp the status of the event to be observed It is characterized by performing.

  The measurement time for measuring the sound pressure / frequency of the sound propagated through the medium is preferably set to about 1 to 100 seconds.

  According to the present invention, the sound generated from the observation object, the solid, the sound propagating in the liquid is obtained as a data chunk, and the data obtained by processing and quantizing and coding the data chunk, There is an effect that it is possible to identify a continuous change in phenomenon.

Hereinafter, embodiments of a sound identification apparatus and a sound identification method of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing a schematic configuration of a sound identification apparatus according to the present invention. As shown in the figure, the sound identification device 10 is installed in a place where there is no influence of precipitation, such as indoors, so as to receive a signal from a waterproof medium microphone 1 installed at a measurement position such as outdoors. It is comprised from the form of the measured box type measuring device. As shown in the block diagram of FIG. 1, the internal configuration includes an amplifier circuit 11 that amplifies a weak signal detected by a medium microphone, a band-pass filter 12 that measures sound by frequency band, and an A / D converter 13. By connecting the operation analysis unit 14 that analyzes signal data based on the installed software (application) and peripheral devices (not shown), data can be output to the outside, and data can be recorded on various media. Output / recording unit 15 and a display unit 16 that conveys operation procedures, analysis results, system information, and the like to the user. Specifically, the arithmetic analysis unit 14 includes a control board on which a control CPU and an arithmetic chip are mounted. In addition to the measurement control software, the arithmetic analysis unit 14 includes quantification software created according to the target event. Wear and encoding software are stored in ROM.

  In the sound identification apparatus of the present invention, data collection and data processing are performed by a plurality of application software. FIG. 2 is a flowchart showing the relationship between each software and each processing flow of the sound identification technology. First, the measurement control software functions to collect the propagation sound data of the target event. This measurement control software is set at regular intervals (1 to 100 seconds) with respect to the sound pressure and frequency for each band of the band (vibration region sound, low frequency sound, audible band sound and full band sound) to which the propagation sound of the target event belongs. The collection of sound pressure and frequency data groups in the degree). In addition, the algorithm for statistical processing and analysis processing for creating data blocks for quantification and encoding is handled uniquely for each observation content and the content of the target event to be discriminated. After designing by measuring, the algorithm was constructed (designed), and quantification software and encoding software were created and handled. Specifically, the collected data group is handled as a data chunk (for example, about 10 to 3,000 data as the number of data), and these data chunks are quantified and encoded by performing statistical processing and analysis processing. It was decided. In addition, software for calibrating quantified data and target events and discriminating software with event discriminants of encoded data based on known reference data and data acquired and accumulated by learning ing. As a result, it is finally possible to convert the phenomenon of the target event into various quantities that can quantitatively grasp the phenomenon of the target event from the quantified data, and it is possible to determine the event to which the target event belongs by using the encoded data.

Here, a configuration concept of a data block to be quantified and encoded based on the collected data block of propagation sound will be described with reference to Table 1. In Table 1, data (n = _{1 to n} ) sampled at a predetermined measurement time (time TT _{1 to} TT n) are vibration sound (DA _{1 to} DA _n ), ultra-low frequency sound (DB) for each band. _{1 to} DB _n ), low frequency sound (DC _{1 to} DC _n ), audible band sound (DD _{1 to} DD _n ) and sound pressure value (P) and frequency defined as full band sound (DT _{1 to} DT _n ) A set of values (F) is a data chunk. The time measurement time (TT ₁ ~TT _n) can be arbitrarily set. Statistical processing and analysis processing are performed from these sound data chunks for digitization. In the present invention, as statistical processing and statistical analysis methods used for quantification, an average value, a variance value, etc., frequency distribution, cumulative frequency distribution, scatter diagram / average value, median value, maximum frequency / standard deviation, range (maximum deviation), There are quartile deviation, 10% value / distortion, kurtosis / correlation coefficient, correlation coefficient, spectrum, and various other methods.

  In encoding, the same data chunk processing is performed. However, since the purpose of encoding is to determine an event, a processing numerical value may not be essential. Therefore, it is also effective to set a threshold value for an event and perform encoding of the threshold value or encoding with a synthesis function, instead of using a statistical processing or statistical analysis method to obtain a numerical value.

  FIG. 3 is a flowchart showing a work flow from obtaining and quantifying and encoding in the sound identification technology. As shown in the figure, first, the change characteristics of the target event for which the auditory identification is performed are grasped. Specifically, the grasp of the change characteristic is to estimate the characteristic of the generated sound in the event (phenomenon) to be observed or discriminated and the change in sound accompanying the change in phenomenon. If it is difficult to simply estimate, it is preferable to perform preliminary measurement based on an experiment or the like simulating a phenomenon change to grasp the result. At this time, it is important to set a time width in which the same phenomenon continues or continues in the target event and the generation of sound due to the same reaction continues. Next, test measurement is performed at an actual site, and characteristics of sound (propagation sound) due to the target phenomenon are grasped. Construction and design conditions for algorithms, etc., to identify and organize the sound pressure level, sound pressure change level, dominant frequency range, etc. of the generated sound (propagation sound) at this stage, and identify the target object (observation, discrimination) Decide.

  Specifically, the set time for measuring the sound pressure / frequency of the generated sound (propagation sound) is preferably about 1 to 100 seconds. This set time is set according to the reaction of the target. For example, when the phenomenon is expected to change with a period of about 5 to 30 seconds as in the case of wind speed observation, the set time is set long. Thus, the change between about 10 to 100 seconds is small, and the measured value as the instantaneous value may be set short for effective data. The number of data measured within this set time and recorded in a recording unit such as a computer is about 20 to 200 data / second, and in the case of 100 seconds, it becomes 2,000 to 20,000 data in proportion. This data group is analyzed by statistical processing, statistical analysis, or a state equation or simulation model created for each phenomenon, and the quantity or coding of sound is performed. In this case, there is a feature in that not the sound pressure value but the frequency of the band necessary for grasping the reaction is continuously measured and the frequency value is used.

  Further, in order to ensure the performance of the apparatus that handles data chunks, a time constant, an amplification degree, etc. in the measurement frequency range are set. Here, in setting the measurement frequency range, a selection is made between a case where the low frequency sound including the vibration sound is centered and a case where the sound is centered on the audible range. In the case of focusing on low frequency sound, the reaction time constant needs to be slowed down to about 2 seconds, and in the case of audible range sound, it is necessary to speed up to about 0.2 seconds or less. In addition, the amplification degree determines the amplification degree of the input signal from the medium microphone installed in the target, and an appropriate amplification degree is obtained from six stages of 0, 20, 40, 60, 80, and 100 dB based on the change characteristic of the input signal. select.

  As the sound pressure sampling item, a necessary item is selected from the sound pressure items, and the sampling rate is also selected based on the reaction rate of the target event. Create a calculation formula or encoding algorithm for quantification in the form of statistical processing, analysis processing or numerical model creation of sound data obtained by deciding these various conditions and continuously measuring To do. Then, the data quantified through the quantification software is converted by calibration, or the data encoded through the encoding software is discriminated by the discriminant conditional expression. The result is output externally in a predetermined format.

  In quantification, as described above, the sound pressure and frequency change of the sound due to the reaction in which the same reaction continues for about 0.5 to 1 or more are measured for a certain period of time (about 1 to 100 seconds) and converted into data. From the data group (number of data: about 20 to 10,000), quantification is performed so that the observation target is linear, or encoding is performed so that the determination of the event becomes highly accurate. In this case, not only statistical processing but also statistical analysis is performed and, for example, sound is indexed to simplify the relational expression. Specifically, as shown by the following formula, a model experiment or the like is performed to derive a linear relational expression between the fluctuation amount of the phenomenon and the quantified value of the sound, and the sound is indexed.

  Hereinafter, examples of quantification formulas and examples will be described. For example, the observed quantities can be expressed by the following function formula.

Y = K (X−X ₀ )
Where, Y: various observed quantities (for example, precipitation during rain, river sediment)
K: Coefficient as calibration
X: Value obtained by quantifying the sound pressure and frequency measurement values
X ₀ : X value when the observed quantities are close to 0 (background value)

  Here, in order to quantify the sound data block, it is preferable that the measured value is not used as it is, but a form close to a linear expression is used, and a mathematical expression that can be calibrated by multiplying by a conversion coefficient is preferable. Thereby, the precision of the apparatus for continuous data acquisition can be maintained. Even when the observed quantities are close to 0, a background sound corresponding to background noise is measured, so that cancellation is made in the equation to improve data accuracy. In some cases, as shown in the following formula, the dimensions of the formula format are raised and the various quantities are converted based on a plurality of quantified values.

Y = f (X ₁ × X ₂ ... × X _n )
Where Y: various observed quantities (eg soil moisture)
X ₁ : Quantification value of sound (1)
X _2: 〃 (2)
_Xn : 〃 (n)

Hereinafter, as an example of quantification, an example of creating a calibration by a conversion formula based on an experiment for precipitation and raindrop particle size distribution by measurement of precipitation sound during rainfall will be shown.
R = a {(P−P ₀ ) / F _K }
Rd ₅₀ = b ₁ (F _K ) ^b2
Where R: Precipitation (mm / min)
Rd ₅₀ : 50% particle size of raindrop (mm)
P: Measurement sound pressure (μPa)
P ₀ : Sound pressure when there is no rain (μPa)
F _K : Raindrop sound frequency (Hz)
a, b ₁ , b ₂ : conversion factors (experimental values)

  Examples of other events include observation of soil water content (moisture content ratio) by measurement of ground penetration sound, observation of snow density by measurement of penetration sound in snow, and observation of amount of sand flow by measurement of sand flow sound in rivers. , Observation of debris flow by measuring vibration noise in Sabo dam concrete enclosure, Observation of foundation penetration by measurement of vibration sound in enclosures such as piers and river facilities, Observation of stable state of structure by measurement of vibration sound of building blocks , Observation of ground semi-liquefaction by measuring ground around concrete structure, Measurement of microbial metabolic rate by measuring sound generated during fermentation by microorganisms, Observation of progress of deterioration of concrete structure, Observation of pre-collapse sign, Tunnel covering There are observations of the progress of cavitation on the back of the work, observation of the landslide phenomenon, etc.

  At the time of encoding, for example, it is only necessary to be able to distinguish the state of event A and event B, and encoding is performed using numerical values or arbitrarily defined classification symbols (for example, A to Z) from the sound pressure / frequency data block. Specifically, in the process of performing an experiment or test measurement on the target event, a conversion algorithm from a data chunk to encoding is created. In this case, the determination of the event is incomplete in simple data comparison, but if the complete determination is pursued too much, the accuracy may be lowered. Therefore, it is preferable to analyze the sound data chunk and process it so that there is a space between the event A and the event B.

  Hereinafter, an embodiment will be described in which it is determined whether the TV program viewed by the samples A and B is the same program as the program-specific sound data (channel X). For example, a TV program has a unique sound (program-specific sound data) that changes for each program and each broadcast time. This program specific sound data is composed of continuous changes in frequency and sound pressure. Therefore, the measurement data of the viewing program sound viewed by the viewer (samples A and B) collected by the sound identification apparatus of the present invention and the program specific sound (program broadcast on channel X) set in advance. ) Is compared with the data stored in the learning device (learning device data), statistically processed and encoded, so that the program viewed by the samples A and B is the program specific sound data (channel X). An embodiment for determining whether or not the same program will be described below.

Specifically, the data block obtained by sampling for a certain period of time is compared with the data of the learning device in the following procedure to make a determination.
(1) The following four items (FD1 to FD4) are obtained from the data chunk of the viewing program sound of the program viewed by the sample A. The results are shown in Table 2.

(2) Substituting FD1 to FD4 obtained from learning device data at the same measurement time and FD1 to FD4 obtained in (1) into the following equations, and calculating each of FD1 to FD4. The results are shown in Table 3.
FD1 to FD4 = −100 × (X−X ₀ ) ² +1
Where X: FD1 to FD4 obtained in (1)
X ₀ : FD1 to FD4 obtained from learning device data

(3) The values obtained in (2) are summed (up to 4 points), and ○ (same program) and x (different programs) are determined by the following discriminant (reference). The results are shown in Table 4.
If total score> 3.9 ○
Other than that ×

  In addition to the above-described embodiments, application examples in each field can be considered for the audio identification technology that assumes encoding. The following devices are provided depending on the installation location of the audio identification device and the necessity of the power supply. Is possible. For example, as a device that does not use a commercial power supply in the outdoors (including underwater), fluid pipe leakage identification device, steel tower, tower abnormal sound identification device, rooftop facility, stability identification device for signboard attachment state, temporary facility stability As a device that uses a commercial power source as an identification device, a traffic presence / absence identification device (example: pass check for cars, trains, etc.), approach, intruder identification device (purpose example: building premises security), abnormal underwater sound identification device ( (Example: Management of aquariums, pools, etc.) Further, as a device using a commercial power source indoors, the above-described TV program sound identification device (purpose example: TV program audience rating survey), consumer identification device (purpose example: life management of elderly living alone), intruder identification device (Purpose example: Building security), Fermentation sound identification device (Normal reaction management), Customer entry / exit identification device, Abnormal behavior identification device (Purpose example: Security assistance), etc. Is possible.

The block block diagram which showed schematic structure of the sound identification apparatus of this invention. The flowchart which showed the function of the software in the sound identification apparatus of this invention. The flowchart which showed the flow of the process for quantification in the audio | voice listening identification method of this invention, and an encoding.

Explanation of symbols

1 Medium microphone 10 Audio identification device

Claims (3)

  A medium microphone that collects sound propagated through the medium to be observed, and the sound propagated through the medium microphone is collected at a predetermined time interval over a predetermined time with a sound pressure value or a frequency width, and a sound data block for each predetermined band Measurement control calculation means for creating the sound data block, quantification calculation means for quantifying the sound data chunk and converting it into various quantities to be observed through calibration capable of evaluating the event, encoding the sound data chunk, and corresponding event And a coding operation means for determining whether or not the object belongs to the sound identification device.   The sound propagated through the observation target medium is collected at a predetermined time interval over a predetermined time with a sound pressure value or frequency width, and a sound data block for each predetermined band is created, and the sound data block is quantified or encoded, A sound identification method characterized in that an event is converted into various quantities to be observed through calibration that can be evaluated, or whether or not the event belongs to a corresponding target event, and the state of the observation target event is grasped.   The method according to claim 2, wherein the measurement time for measuring the sound pressure and frequency of the sound propagated through the medium is set to about 1 to 100 seconds.

Images