WO2014024382A1 - Noise observation device and noise observation method - Google Patents

Abstract

A noise observation device comprises: a cross-correlation coefficient calculation unit (112) which, using four microphones (M0-M3) which are respectively positioned two-by-two upon the x-axis, the y-axis, and the z-axis, computes cross-correlation coefficients of sound which has reached each microphone (M0-M3) for each axis; a peak search processing unit (114) which extracts, in descending order of cross-correlation coefficient, a plurality of time lags which denote a peak trend for the cross-correlation coefficients; a segmentation processing unit (116) which collects fluctuations in a time region of the plurality of time lags and forms a set of time lags which is contiguous for each axis; a normalized cross-correlation coefficient calculation unit (122) which, for the set of time lags for each axis, computes normalized cross-correlations between different axes; and a segment integration processing unit (124) which combines sets of time lags with the same sound sources, from normalized cross-correlation coefficients.

Description

Noise observation apparatus and noise observation method

The present invention relates to a noise observation apparatus and a noise observation method suitable for use in an environment where a plurality of sound sources exist in an observation target area.

Conventionally, a prior art effective for automatic identification of aircraft flight noise observed under a flight route such as an aircraft has been known (see, for example, Japanese Patent Laid-Open No. 7-43203: Patent Document 1). In this prior art, the arrival direction vector (elevation angle, azimuth angle) of the moving sound source is calculated from the cross-correlation coefficient of the time delay of the sound reaching four microphones arranged at intervals on the X, Y, and Z axes. And the movement trajectory of the moving sound source is automatically identified from the obtained vector set.

According to the prior art identification method, even if the observation point is affected by the noise of the aircraft taking off and landing from the airport, it accurately grasps the effects of the noise of the aircraft taking off and landing from the airport, and distinguishes them from each other. A moving course can be identified with high accuracy.

Until now, it was sufficient to observe only the flight noise such as the noise from the sky, the take-off runway noise generated on the runway and the reverse noise at the time of landing. In the future, it will be necessary to observe the ground noise of the aircraft generated by aircraft operations and aircraft maintenance in the airfield. Ground noise includes, for example, noise generated by the operation of the auxiliary power supply unit (APU) in the parked aircraft and aircraft that are moving between the terminal and the runway (taxing) for propulsion. And noise generated when an aircraft performs an engine test operation in an engine test operation area in an airfield.

Furthermore, the noise observed around the airfield is complex, and various noises such as cars and sirens come from the surroundings at the observation point, so only the ground noise generated by the aircraft can be compared with other noise from the ground. It is difficult to detect pinpoints.

Aircraft noise includes transient single-shot noise that occurs on the airfield as the aircraft operates, engine test operations and APU operations that are observed around the airfield due to aircraft maintenance, etc. Noise continues for a long time, and there are quasi-stationary noises that are steady but have considerable level fluctuations, making noise identification more difficult.

In the sound source identification method using the three-axis cross-correlation method found in the above prior art (Patent Document 1), when noise is simultaneously generated from a plurality of sound sources, the cross-correlation coefficient is on each axis. The maximum time delay does not necessarily indicate the direction of arrival of sound from the same sound source. In addition, when a plurality of noises are overlapped with each other, only the sound having the maximum cross-correlation is used, and it is difficult to automatically identify other noises.

With such a background, there is a need for a technology that can identify noise using the cross-correlation method even when multiple noises occur simultaneously in a noise environment where there are multiple sound sources, such as in an airfield. Yes.
Japanese Unexamined Patent Publication No. 7-43203

The invention disclosed herein employs the following solutions.
In other words, this solution calculates the time delay cross-correlation coefficient for each axis at regular intervals, and for the time delay in which the cross-correlation coefficient shows a peak (maximum) tendency, the cross-correlation coefficient starts from the largest. In this method, a plurality of time delay variations are extracted in the time domain to form a continuous time delay set for each axis. At this time, the set of time delays formed for each axis is a set separated for each sound source when there are a plurality of sound sources. Then, regarding the time delay sets formed for each axis, if the cross-correlation between the different axes is now observed, the time delay sets having the same sound source can be combined therewith. As a result, it is possible to automatically separate and identify a plurality of simultaneously occurring noises.

For example, when multiple noises (landing and taxing, etc.) occur simultaneously in an airfield, the noise with the highest sound pressure level tends to dominate the maximum peak of the cross-correlation coefficient for each axis. As shown in the figure, when looking at a plurality of peaks in order from the top, attention is paid to the fact that one of the peaks shows the influence of noise (other sound sources) other than the maximum noise.

That is, when a sound source moves in the observation space, even if the cross-correlation coefficient is indicated by a time delay that is the maximum peak in a certain time zone, it is not the maximum peak in another time zone, It may be indicated by the time delay of other lower peaks. In this case, the time delay at which the cross-correlation coefficient reaches the maximum peak in the same time zone indicates other sound sources, so simply tracing the maximum peak always in the time domain does not identify the sound source. .

In view of this, this solving means extracts a plurality of time delay fluctuations in the time domain that are higher in the peak of the cross-correlation coefficient calculated at regular intervals (for example, from the first to the third), and continuously for each axis. We decided to form a set of time delays. Since a set of consecutive time delays represents the expected time delay variation (noise data indicating the direction of arrival) that is expected to be the same sound source in the time domain, if there are multiple sound sources in the observation space, each axis A set of time delays separated by sound source on the line is formed. Thereby, the noise data of a plurality of sound sources can be separated for each sound source on each axis.

Furthermore, a set separated on each axis is combined between different axes. That is, in the present invention, when a cross correlation coefficient is considered instead of individual time delay for a certain set, attention is paid to the fact that the sets considered to be the same sound source have very similar fluctuations in the cross correlation coefficient. is doing. This is presumably because the sound emitted from the actual sound source is affected by, for example, output fluctuations and weather changes, and these changes appear in common with the sound input to the microphone. Therefore, if a set of sets exists at the same time between different axes in a certain time domain, the sets are determined from the normalized cross-correlation coefficient when the time delay is 0 with respect to the time variation of the cross-correlation coefficient of each set. Can be combined. Since the sets combined between the axes represent the arrival directions of the same sound source, a plurality of sound sources existing in the observation space can be automatically identified therefrom.

The noise observation apparatus according to the present solution has a configuration including a calculation unit, an aggregation unit, and an integration unit. The noise observation method of the present solution can be executed using these configurations.
In other words, the calculation means uses two microphones arranged at intervals on a plurality of axes defined in an observation space where a plurality of sound sources exist, and uses a time delay that represents a difference in sound arrival time for each axis. The step of calculating the cross-correlation coefficient is performed at regular intervals. Further, the aggregation means extracts a plurality of time delay fluctuations in the time domain in descending order of the cross correlation coefficient for a plurality of time delays in which the cross correlation coefficient calculated every fixed time by the calculating means shows a peak tendency, A step of forming a set of continuous time delays for each axis is executed. Then, the integration unit executes a step of combining sets of time delays having the same sound source from the cross-correlation between different axes with respect to the sets of time delays for each axis formed by the aggregation unit.

This makes it possible to automatically identify sound sources using the cross-correlation method even when there are multiple sound sources in the observation space and multiple noises are generated at the same time.

In the noise observation device and the noise observation method of the present solution, the following various aspects can be provided in forming a continuous set of time delays.

The aggregation means can execute a step of confirming whether or not a plurality of time delays showing a peak tendency at each fixed time become initial values for each sound source prior to formation of a set of continuous time delays.

In the above aspect, after confirming the initial value at the time of noise generation from the upper peak of the cross-correlation coefficient, it is possible to appropriately form a set of continuous time delays over the subsequent time domain.

Moreover, the steps performed by the aggregation means can include the following steps.
(1) Among a plurality of time delays showing a peak tendency at every fixed time, whether or not there is a single value of time delay that makes a difference from a specific time delay showing a peak tendency before the fixed time less than a predetermined threshold Step to determine whether.
(2) A step of adding a unique value to the same set as a specific time delay if it is determined in (1) that the unique value exists.
(3) A step of calculating a virtual time delay by the least square method using at least a specific time delay (the latest several) when it is determined in (1) that there is no unique value.

If it is said aspect, about the time delay of the upper peak calculated every fixed time, after confirming whether it is a time delay by the same sound source as the last time delay, the time delay (it is expected that the sound source is the same) Only unique noise data) can be added to a continuous set. Thereby, the accuracy of the identification result can be improved and the reliability of the noise observation result can be increased. Also, if there is nothing that can be predicted as the same sound source in the time delay of the upper peak this time, obtain the virtual time delay by the least square method using the time delay for the last several pieces including the previous value, Can be used for the steps.

That is, the aggregation unit further executes the following steps.
(4) Since there are a plurality of time delays in which the difference from a specific time delay is less than a predetermined threshold, there is no single value, or there is a predetermined difference from a specific time delay among a plurality of time delays. Determining whether there is only one value because there is no time delay that is less than the threshold value.
(5) If it is determined in (4) that there is no single value because there are multiple time delays where the difference from the specific time delay is less than the predetermined threshold, the difference from the virtual time delay is Determining whether there is a specific value that is less than a specific threshold;
(6) A step of adding the specific value to the same set as the specific time delay when it is determined in (5) that there is the specific value.
(7) In the above (5), it is determined that there is no specific time, or in the above (4), there is one time delay in which the difference from the specific time delay is less than a predetermined threshold among the plurality of time delays. Adding a virtual time delay to the same set as a particular time delay if it is determined that there is no unique value because it does not exist.

In this case, the virtual time delay calculated in (3) above represents the latest time delay variation in the time domain. Therefore, when there is a plurality of time delays and the only value cannot be narrowed down, it is possible to add the ones close to the virtual time delay to the set after predicting the same sound source (6). On the other hand, if there is nothing close to the virtual time delay and there is no single value, the virtual time delay can be added to the set to continue the assembly (7).

Further, the aggregation means can further execute the following steps.
(8) An end determination step of ending the formation of the set by deleting the virtual time delay for the predetermined number of times when the virtual time delay is continuously added to the set for a predetermined number of times at regular times.

In this case, if the last time delay in the set is a virtual time delay and all of the immediately preceding predetermined numbers are virtual time delays, the virtual time delay for a predetermined number of consecutive is deleted, and for the set Further aggregation can be terminated. As a result, it is also possible to end (end processing) the collection of continuous time delays in accordance with the end time of a noise event or the like.

Further, the aggregation means deletes a predetermined number of virtual time delays in the end determination step of (8) above, and as a result of completing the formation of the set, the number of time delays included in the set is equal to or less than a specified number. If so, the set can be invalidated.

In the case of the above-described aspect, it is possible to avoid mixing noise in subsequent identification by invalidating those having insufficient absolute numbers as a set. Thereby, identification accuracy can be improved and the reliability of noise observation results can be further increased.

According to the noise observation apparatus and the noise observation method disclosed above, a plurality of sound sources that are simultaneously generated in the observation space can be automatically separated and identified.

Also, since the calculation load can be reduced by processing periodically rather than processing after accumulating a large amount of calculation results, real-time processing using a computer can be easily realized.

FIG. 1 is a schematic diagram showing one embodiment when a noise observation device is installed in an airfield, FIG. 2 is a diagram schematically showing the configuration of the noise observation apparatus and the noise identification method based on the cross-correlation method. Fig. 3 is a diagram explaining the noise event detection method for single noise, along with the temporal change of the noise level under the channel, FIG. 4 is a diagram explaining the noise event detection method for quasi-stationary noise, along with the temporal change of the noise level in the airfield (or nearby), FIG. 5 is a simplified model diagram for explaining the principle of the identification method. FIG. 6 is a flowchart illustrating an example of a procedure of sound source separation processing executed by the sound source separation processing unit. FIG. 7 is a schematic diagram showing ranking of time delays due to the upper peak of the cross-correlation coefficient, FIG. 8 is a flowchart showing a procedure example of the same sound source segmentation processing for each axis, FIG. 9 is a diagram showing the variation of the time delay on the X axis in the simplified model, FIG. 10 is a diagram showing an example in which the variation in time delay on the X-axis shown in FIG. 9 is separated into sound source segments, FIG. 11 is a diagram illustrating an example when the simplified model is applied to a moving sound source, FIG. 12 is a diagram showing fluctuations in time delay in the X axis and the Y axis. FIG. 13 is a diagram showing an example of segment separation in the X-axis. FIG. 14 is a diagram showing the fluctuation results of the time delay in the X axis and the Y axis for the measured data. FIG. 15 is a diagram showing a result of separating the time delay variation in the X axis and the Y axis into segments for the measured data. FIG. 16 is a diagram showing a variation pattern of the cross-correlation coefficient of each segment. FIG. 17 is a diagram illustrating a calculation example of a normalized cross-correlation coefficient between overlapping segments in the time domain.

Hereinafter, embodiments will be described with reference to the drawings.

FIG. 1 is a schematic diagram showing one embodiment when a noise observation apparatus is installed in an airfield. In a target area such as an airfield (or its surroundings), noise coming from the sky as the aircraft travels, take-off / landing noise generated on the runway, and reverse noise during landing (hereinafter referred to as “flight noise”). ) In addition to the noise associated with aircraft operations and aircraft maintenance in the airfield, there is a noise environment that is mixed with noise (hereinafter referred to as “ground noise”) associated with taxing, engine trial operation, and APU operation. Is formed.

As shown in FIG. 1, the noise observation apparatus can be used with a microphone unit 10 installed at an observation point in an airfield. An observation unit (not shown) is connected to the microphone unit 10.

Within the target airfield, there are various areas that are noise sources, such as the landing area 20, the taxing road 30, the lander 40, the take-off machine 50, and the engine trial operation area 60 that run or fly on the runway 25. Exists. In the airfield, various noises are generated from these places and arrive at the observation point from each direction. The noise observation apparatus of the present embodiment is suitable for an application for automatically identifying a plurality of noises arriving at an observation point by using the microphone unit 10. In the following, explanation will be made for each area that is a source of noise.

[APU]
From the parking area 20, noise is generated due to the operation of an auxiliary power unit (APU: Auxiliary Power Unit). The auxiliary power unit is a small engine that is used as a power source for supplying compressed air, hydraulic pressure, electric power, and the like into the parked aircraft AP.

[Taxing]
The taxing road 30 is a runway on which the aircraft AP moves between the parking area and the runway 25. From the aircraft AP during taxing, the engine is operated to obtain a propulsive force necessary for ground running, and noise is thereby generated.

[Landing sound]
The landing aircraft 40 enters and descends toward the runway 25 when the aircraft AP arrives, and, in many cases, performs reverse engine injection (reverse) on the runway 25 for deceleration, and finally the runway. Generates noise associated with the flight until 25.

[Takeoff sound]
The take-off aircraft 50 starts to run at the start position of the runway 25 when the aircraft AP departs, and generates noise associated with the operation until it flies, rises and flies off in the middle of the runway 25.

[Engine test run]
Further, in the engine trial operation area 60, noise is generated along with the trial operation performed for confirming the operation of the aircraft AP engine (main engine).

Although not shown in FIG. 1, the following noises are also generated in the airfield.

[Touch and Go]
For example, when the aircraft AP performs a flight form (touch and go) in which the aircraft AP enters the runway 25, takes a landing, decelerates, decelerates, then increases the engine output and takes off again for takeoff and landing training, etc. appear.

[Hovering]
When the helicopter flies and takes a flight form in which it is almost stationary, noise accompanying this occurs.

[City area]
In addition, when there is an urban area 70 in the vicinity of the airfield, for example, other ground noise is generated due to various social activities (transportation operation, road traffic, civic life, etc.) in the urban area 70.

FIG. 2 is a diagram schematically showing the configuration of the noise observation apparatus and the noise identification method based on the cross-correlation method. The noise observation apparatus has a function of performing calculation processing using the microphone unit 10 and identifying noise by a cross-correlation method.

[Microphone unit]
The microphone unit 10 includes, for example, four microphones M0, M1, M2, and M3. The individual microphones M0 to M3 are arranged on the X axis, the Y axis, and the Z axis that are virtually determined in the observation space. It is arranged on the axis and at the origin of the three-axis coordinate system. Specifically, the microphone M0 is disposed at the origin, and another microphone M1 is disposed on the Z axis extending in the vertical direction from the origin. In addition, another microphone M2 is installed on the Y axis extending in the horizontal direction from the origin and opening 90 ° with the X axis, and another microphone M3 is installed on the X axis extending in the horizontal direction from the origin. Has been. The microphone unit 10 holds the relative positions of the microphones M0 to M3 in an installed state while mechanically fixing the individual microphones M0 to M3. As described above, two microphones are disposed on each axis on the X axis, the Y axis, and the Z axis in the observation space.

In addition, the microphone unit 10 includes a microphone MB different from the four microphones M0 to M3 described above. The four microphones M0 to M3 are for noise identification by the cross correlation method, while the microphone MB is for measuring ambient noise. That is, the microphone MB is used to measure the noise level at the observation point alone, for example.

[Observation unit]
The noise observation apparatus includes an observation unit 100, and a microphone unit 10 is connected to the observation unit 100. The observation unit 100 includes, for example, computer equipment including a central processing unit (CPU), a semiconductor memory (ROM, RAM), a hard disk drive (HDD), an input / output interface, a liquid crystal display, and the like (not shown).

[Noise discrimination method by cross-correlation method]
Next, a noise identification method using the cross-correlation method using the four microphones M0 to M3 will be described. In addition, since the noise identification method by a cross correlation method is already well-known, the outline is demonstrated here.

For example, consider the identification of a sky sound. When two microphones M1 and M0 are vertically installed on the vertical line (Z-axis line) in the observation space, the interval between them is defined as d (m). When the sound of the flying aircraft AP enters at an elevation angle θ, the time difference τ [s] when the sound reaches the two microphones M1 and M0 is expressed by the following equation (1), where the sound speed is c [m / s]. ). The time difference τ is a time delay when the cross-correlation of the sounds that reach the microphones M1 and M0 has a maximum value.
τ = d / c · sin (θ) (1)
From the above equation (1), the elevation angle θ of the sound source viewed from the observation point can be obtained.

When it is considered that the direction of sound arrival is sufficiently high (θ> 0), this elevation angle θ information can be used for identifying flight noise (see Patent Document 1 cited in the prior art). That is, for example, when the noise level detected by the microphone MB exceeds a certain threshold value (when a noise event occurs), if the elevation angle change θ (t) is recorded at the same time as the sound arrival direction data, it is designated in advance. It is possible to determine that the noise of the sound arrival direction data larger than the elevation angle is the flight noise caused by the aircraft AP.

[Calculation of arrival direction vector]
In addition to the vertical direction, the azimuth angle δ can be obtained by calculation in addition to the elevation angle θ if the sound arrival direction is expanded not only in the vertical direction but also in the three axes of the XY, YZ, and ZX axes. It is. Then, by obtaining the elevation angle θ and the azimuth angle δ, it is possible to calculate a noise arrival direction vector (unit vector) in a three-axis observation space (vector space) with the observation point as a reference. Further, the moving product of the sound source (aircraft AP) (from which direction to which direction) can be more reliably known with the cross product of the calculated arrival direction vectors as a reference.

[Application of flight sound discrimination method to ground noise discrimination]
As described above, if the three-axis arrival direction vector of sound can be calculated, when the elevation angle θ indicates the ground, the arrival direction of noise from the ground can be determined from the azimuth angle δ. However, in this discrimination method, when there are a plurality of sound sources in the observation space and a plurality of noises are generated at the same time, the time delay τ when the cross-correlation coefficient for each axis is maximum is not necessarily the same. It does not necessarily indicate the same sound source. Therefore, when a plurality of noises are generated simultaneously, the sound arrival direction may not be accurately calculated.

Various studies on the identification of multiple sound sources have been reported so far, but none of them can be separated when the movement of the sound source is complicated, or processing performed after accumulating huge amounts of data. Therefore, there is a problem that real-time processing cannot be performed.

For this reason, in the present embodiment, an algorithm simplified as much as possible is adopted in consideration of processing in real time inside the noise observation apparatus. Hereinafter, the noise observation method used in this embodiment will be described in more detail.

[Configuration as a noise monitoring device]
The observation unit 100 includes a noise event detection unit 102 and a direction-of-arrival vector calculation unit 106 as functional elements thereof, and further includes a sound source separation processing unit 110 and a separated sound source integration unit 120 including a plurality of functional elements.

Among these, the noise event detection unit 102 detects the ground noise level generated in the target area based on the noise detection signal from the microphones MB, M0 to M3, for example. Specifically, the result of digital conversion of the noise detection signal is sampled, and the noise level value (dB) at the observation point is calculated.

[Single noise / Quasi-steady noise]
Aircraft noise can be broadly divided into single noise and quasi-stationary noise. Of these, single noise is transient noise that occurs once, such as noise observed in the vicinity of an airfield as the aircraft AP operates. In the case of ground noise, taxing noise is often observed as a single noise.

The quasi-stationary noise is noise that continues for a long time and is steady but accompanied by a considerable level fluctuation. When it is generated from the aircraft AP, it is regarded as ground noise of the aircraft. Specifically, this includes engine test operation observed in the vicinity of an airfield accompanying the maintenance of the aircraft AP, operation noise of the APU, noise during standby before takeoff at the end of the runway, and the like. In addition, helicopter idling and hovering noises often continue constantly and may be observed as quasi-stationary noises.

Although not particularly illustrated, the noise event detection unit 102 registers a condition (threshold level) for detecting a noise event of single noise or quasi-stationary noise from the noise level value in the observation unit 100. The noise event detection unit 102 applies the calculated noise level value (dB) to the registered conditions, detects a single-shot flight noise or ground noise event, or detects a quasi-steady noise ground noise event. be able to. An example of noise event detection will be described later.

The arrival direction vector calculation unit 106 calculates the sound arrival direction vector (elevation angle θ, azimuth angle δ) by the above-described three-axis cross-correlation method based on the detection signals from the four microphones M0 to M3. Moreover, the arrival direction vector calculation unit 106 records the elevation angle θ (t) and the azimuth angle δ (t) represented by a time function as sound arrival direction data.

[Sound source separation processing section]
The sound source separation processing unit 110 includes a cross correlation coefficient calculation unit 112, a peak search processing unit 114, and a segmentation processing unit 116. The sound source separation processing unit 110 has a function of separating the time delay obtained for each axis into sound sources on each axis.

Among these, the cross-correlation coefficient calculation unit 112 calculates a time delay representing arrival time differences of a plurality of upper sounds of the cross-correlation peak for each axis based on detection signals from the four microphones M0 to M3 at regular intervals. Further, by calculating for each axis, a cross-correlation coefficient R (axis, i, τ) is obtained here (i is a time index for each fixed time: i = i + 1, updated regularly, axis = X, Y, Z) ). The calculation result here is also provided to the arrival direction vector calculation unit 106 described above.

The peak search processing unit 114 searches for a plurality of time delays indicating a peak tendency for the cross-correlation coefficients for each axis calculated by the cross-correlation coefficient calculation unit 112, and the highest number (for example, the first number) from the maximum peak is searched. Rank the time delay from place to place 3). In order to rank the cross-correlation coefficient peaks for each axis, a plurality of ranked time delays τ _{axis, i, j} are obtained here (j = 1,..., M: for example M = 3). .

The segmentation processing unit 116 collects a plurality of time delay fluctuations higher than the peak value extracted by the peak search processing unit 114 in the time domain, and collects the same sound source and expected ones in the same set. A process for forming a time delay set is performed. Hereinafter, such processing is referred to as “segmentation”, and the formed set is referred to as “segment”. Details of processing by the sound source separation processing unit 110 will be described later with reference to another flowchart.

[Separated sound source integration section]
Next, the separated sound source integration unit 120 includes a normalized cross correlation coefficient calculation unit 122 and a segment integration processing unit 124. The separated sound source integration unit 120 has a function of combining segments for each axis formed by the segmentation processing unit 116 with the same sound source. Since the segments formed by the segmentation processing unit 116 are separated for each sound source on each axis as described above, in order to identify sound sources in the observation space, it is necessary to combine the sets between each axis. is there. Therefore, here, the segments between the axes are combined using the fluctuation of the correlation coefficient R (τ).

For this reason, the normalized cross-correlation coefficient calculation unit 122 generates a time delay for the time variation of the cross-correlation coefficient R (τ), not the individual time delay τ, for the segment formed on each axis. The normalized cross-correlation coefficient R (0) when there is no (τ = 0) is calculated here.

Then, the segment integration processing unit 124 integrates segments having a sufficiently large calculated cross-correlation coefficient R (0) as segments of the same sound source. Details of processing performed by the separated sound source integration unit 120 will also be described later.

In addition, the observation unit 100 includes an identification result output unit 130. The segments integrated by the segmentation processing unit 116 are provided to the identification result output unit 130. The identification result output unit 130 identifies a plurality of types of sound sources from the information of the segments integrated into the same sound source and the arrival direction vector calculated by the arrival direction vector calculation unit 106, and outputs the result. The output result can be displayed on a display device (not shown) or transmitted as data to an external computer of the observation unit 100, for example.

[Noise event detection method]
FIG. 3 is a diagram illustrating a noise event detection method for single noise, along with a temporal change in the noise level under the channel. The observation unit 100 calculates the background noise level (BGN) at the observation point by, for example, continuously detecting the noise level in the noise event detection unit 102.

単 Single noise is generated as transient noise when the aircraft AP passes over the sky as described above. Accordingly, the temporal change in the single noise level increases with time, and increases to a level 10 dB higher than the background noise level at time t1. Thereafter, the noise level reaches the maximum value (Nmax) and becomes the background noise level (BGN) again.

In this case, the observation unit 100 starts noise event detection at the noise event detection unit 102 from time t1. That is, when the noise level of the microphone MB rises to a level 10 dB higher than the background noise level (BGN), noise event detection processing is started.

In the noise event detection unit 102, a threshold level (Na) for determining that single noise has occurred is set in advance. Therefore, the noise event detection unit 102 identifies single noise only when the observed value exceeds the threshold level (Na). In this example, since the observed value actually exceeds the threshold level (Na), the noise event detection unit 102 determines that the single noise generation time is at time t3 when the noise level reaches the maximum value (Nmax). it can.

Further, at this time, the noise event detection unit 102 determines the time t4 when the noise level is lowered by 10 dB from the maximum value (Nmax) as the end time of the single noise. As a result, the period from the time t1 (start time) to the time t4 (end time) is the period during which the noise event is being detected (detection process).

Then, the noise event detection unit 102 cuts out a period when the noise level is higher than the value lower than the maximum value (Nmax) by 10 dB, and determines this as a noise event section. The noise event section is regarded as the time when single noise continues at the observation point.

Next, FIG. 4 is a diagram explaining the noise event detection method for the quasi-stationary noise along with the temporal change of the noise level in the airfield (or in the vicinity). Here again, the observation unit 100 calculates the background noise level (BGN) at the observation point by continuously detecting the noise level in the noise event detection unit 102.

[Detection of quasi-stationary noise]
Assume that a quasi-stationary noise is generated by the aircraft AP in the airfield. Before a certain time t12, the observation value at the observation point rises to a level (NP1) 10 dB higher than the background noise level (BGN) due to, for example, movement of the taxing path 30 or the like. After that, the noise level further increases, the quasi-steady high level is maintained while being maintained for a long time, decreases to a level (NP2) 10 dB higher than the background noise level (BGN), and again the background noise level (BGN). become.

In this case, the observation unit 100 starts detection of the noise event at time t12 in the noise event detection unit 102. In other words, the noise event detection process is started when the level rises to 10 dB higher than the background noise level (BGN) (NP1). However, in the case of quasi-stationary noise, no threshold level is set.

And the noise event detection part 102 cuts out the period when the observed value was 10 dB higher than the background noise level (BGN), and determines this as the noise event section. The noise event section in this case is regarded as the time when the quasi-stationary noise continues when the observation point continues for a certain long time.

Here, detection of noise events that occur only once, such as takeoff, landing, and overpass, is a useful technique for identifying whether or not it is aircraft noise in continuous monitoring of actual aircraft noise, etc. It has become. On the other hand, as described above, the quasi-stationary noise caused by the engine test performed by the aircraft AP in the airport or the APU may have a relatively long duration (quasi-stationary noise section). In addition, since the quasi-stationary noise may include a plurality of sound sources in the noise section, it is difficult to automatically identify what kind of noise event it is after the quasi-stationary noise section is detected.

In the present embodiment, the sound source separation processing unit 110 and the separated sound source integration unit 120 described above use the maximum value of the cross-correlation coefficient (the number of upper peaks) for a plurality of sound sources generated simultaneously, and the direction of arrival of each sound. Implement a method to identify Also, with this method, identification processing is possible simultaneously with the occurrence of a noise event, and so-called “real time processing” is possible. Hereinafter, details of the identification method in the present embodiment will be described.

[Principle explanation]
First, the principle of the identification method in this embodiment will be described.
FIG. 5 is a simplified model diagram for explaining the principle of the identification method in the present embodiment. In the simplified model, three microphones M0, M2, and M3 are arranged on two axes (XY axes) in an anechoic chamber AR. That is, here, in order to focus on ground noise in the airfield, the microphone M1 on the Z axis is omitted, and the observation space is simplified in two dimensions. Then, an observation horizontal plane PL is virtually defined on two XY axes, and two fixed sound sources SS1 and SS2 are arranged there.

Under the above conditions, sound is output from the two sound sources SS1 and SS2 before and after time, and the sound source separation processing unit 110 performs segmentation by sound source.

[Sound source separation processing]
FIG. 6 is a flowchart illustrating a procedure example of the sound source separation processing executed by the sound source separation processing unit 110. The sound source separation processing unit 110 can execute the sound source separation processing unit at regular intervals (for example, every 200 ms) by, for example, timer interruption. Hereinafter, it demonstrates along the example of a procedure.

Step S10: First, the sound source separation processing unit 110 defines the current time index i. As described above, the time index i is updated by 1 increment every fixed time (i = i + 1).

Step S12: Next, the sound source separation processing unit 110 performs cross-correlation coefficient calculation processing for each axis by the cross-correlation coefficient calculation unit 112 described above. In this process, the cross-correlation coefficient calculation unit 112 calculates a cross-correlation coefficient R (axis, i, τ) for each axis (here, X-axis and Y-axis, and so on). In the actual configuration, axis = X, Y, Z, but the simplified model omits the Z axis. The subsequent processing is executed for each axis.

Step S14: That is, the sound source separation processing unit 110 performs the peak search processing for each axis by the peak search processing unit 114 described above. In this processing, the peak search processing unit 114 searches for the upper peak of the cross-correlation coefficient R (axis, i, τ) for each axis, and ranks the values in descending order. If the rank is j, a plurality of time delays τ _{axis, i, j} ranked from the first peak to the third peak as described above are obtained (j = 1, 2, 3). The subsequent processing is executed for each peak Peak (τ _{axis, i, j} ).

[Ranking by top peak]
FIG. 7 is a schematic diagram showing ranking of time delays due to the upper peak of the cross-correlation coefficient. For example, when there are a plurality of sound sources in the observation space (a single sound source may be used), if the cross-correlation coefficient R (τ) is calculated on each axis, a mutual phase relationship is obtained with a plurality of time delays τ1, τ2, and τ3. The number shows a peak (maximum) tendency, and the first peak R (τ1), the second peak R (τ2), and the third peak R (τ3) are observed in the upper rank. Thereby, a plurality of time delays τ1, τ2, and τ3 can be ranked in descending order of the peak of the cross correlation coefficient. Note that the result of ranking performed for each axis by the peak search processing unit 114 is τ _{axis, i, j} described above.

Step S16: The sound source separation processing unit 110 uses the segmentation processing unit 116 to execute a sound source initial value determination process for each axis and each peak. In this process, the segmentation processing unit 116 confirms whether or not each Peak (τ _{axis, i, j} ) is an initial value of the sound source. For example, some tau _{axis, i,} the particular tau _axis which is expected to the same sound source immediately before _{j, if i-1, j} is absent, segmentation processing unit 116 the _{Peak (τ axis, i, j} ) the It is regarded as the start point τ _{axis, s, k of the} sound source (s is one of i and k is one of rank j). Thereafter, segmentation is determined for this initial value.

Step S18: Next, the sound source separation processing unit 110 causes the segmentation processing unit 116 to perform the same sound source segmentation processing for each axis. In this process, the segmentation processing unit 116 determines whether or not the current τ _{axis, i, j} is the same as the specific τ _{axis, i−1, j} before the fixed time _, and the sound source, When the sound sources are the same, they are set as one segment. If τ _{axis, i−1, j} one time before has already been part of the segment, this time τ _{axis, i, j} is added to the same segment, and the segment is grown in the time domain. The specific contents of the process will be described later using still another flowchart.

Step S20: Then, the sound source separation processing unit 110 executes the segment-by-segment end determination process by the segmentation processing unit 116. This process is performed for each of all the segments being formed. For example, in the segmentation processing unit 116, the current τ _{axis, i, j} is τ _virtual , and γ of the immediately preceding γ (including γ = 10) including τ _{axis, i, j} is τ _virtual . If this is confirmed, all the 10 consecutive τ _virtuals are deleted, and the initial value τ _{axis, s, k} to τ _{axis, i−γ, j} are set as one segment (termination process). Note that “τ _virtual ” will be described later. At this time, if the number of τ constituting the segment is small to some extent (for example, less than 50), the segment is invalidated.

[Same sound source segmentation for each axis]
FIG. 8 is a flowchart showing a procedure example of the same sound source segmentation processing for each axis. Hereinafter, the contents of the process will be described according to a procedure example.

Step S100: The segmentation processing unit 116 has one time delay of this time that is regarded as the same sound source as any one of the initial values τ _{axis, s, k} or specific segmented τ _{axis, i−1, j.} (Unique value) Determine whether it exists. Specifically, the following equation (2) is calculated.

here,
α: Constant depending on the moving speed of the sound source (predetermined threshold)

Step S102: As a result, when there is only one (only) j ′ that satisfies the formula (2) (step S100: Yes), the segmentation processing unit 116 uses the time delay τ _{axis, i, j ′} . Segment. Specifically, the time delay τ _{axis, i, j ′} is added to the members of the same segment as τ _{axis, i−1, j} .

Step S104: On the other hand, if there are two or more j ′ satisfying the expression (2) or none exists (step S100: No), the segmentation processing unit 116 uses the previous τ to minimize Τ _virtual is calculated from the square method. At this time, it is not necessary to use all τ immediately before, and several pieces of data are sufficient. That is, it is sufficient here to obtain the latest variation of τ as τ _virtual . The order of the least squares method is empirically effective as the second order.

Step S106: The segmentation processing unit 116 checks whether there is a current time delay that is regarded as two or more identical sound sources.
Step S108: As a result, when there are two or more j ′ satisfying the expression (2) (Step S106: Yes), the segmentation processing unit 116 determines whether or not there is a time delay that can be segmented. to decide. Specifically, the following equation (3) is calculated.

here,
β: preset constant (specific threshold), value can be set experimentally or empirically.

Step S110: When there is an optimum τ _{axis, i, j ′} that satisfies the above expression (3) at the minimum (step S108: Yes), the segmentation processing unit 116 uses the optimum τ _{axis, i, j ′} . Segment. Specifically, the optimum τ _{axis, i, j ′} is added to the members of the same segment as τ _{axis, i−1, j} .

Step S112: On the other hand, if there is no j ′ satisfying Expression (3) (Step S108: No), or if there is no j ′ satisfying Expression (2) (Step S106: No), the segment The segmentation processing unit 116 performs segmentation using τ _virtual calculated in the previous step S104. Specifically, τ _virtual is added to the members of the same segment as τ _{axis, i−1, j} .

[Changes in time delay]
FIG. 9 is a diagram illustrating a variation in time delay on the X axis in the simplified model. The horizontal axis in FIG. 9 indicates the number of time indexes, and the vertical axis indicates the time delay on the X axis. Also, the white circles shown in FIG. 9 indicate the time delay ranked in the first peak of the cross-correlation coefficient, and the hatched circles indicate the time ranked in the second peak of the cross-correlation coefficient. Indicates a delay.

[Execution condition 1]
In this example, in the above simplified model (FIG. 5), the sound is first output from the first sound source SS1, and the sound is output from the second sound source SS2 later to generate a plurality of sound sources SS1, SS2. After outputting the sound simultaneously for a while, the output of the first sound source SS1 was stopped first. At this time, since the two sound sources SS1 and SS2 are fixed in the observation horizontal plane PL, the time delays of the detection signals obtained from the two microphones M0 and M3 on the X axis are the microphones M0 and M3 and the sound sources SS1, SS2. It is considered that a substantially stable value is shown based on the relative positional relationship with SS2.

[Time index number 0 to Ti1]
Therefore, when the output of the first sound source SS1 is started from the beginning of the time domain (time index number = 0), the time delay on the X axis is only the one ranked in the first peak of the cross-correlation coefficient. It is represented by In the time domain until the sound is output from the second sound source SS2 (time index number 0 to Ti1), a segment is formed only by the time delay ranked in the first peak of the cross-correlation coefficient. Can do.

[Time index number Ti1 to Ti2]
When the output is started from the second sound source SS1, a plurality of sound sources exist in the observation horizontal plane PL. In this case, the time delay on the X-axis is represented as a plurality of ranks ranked in the first peak and the second peak of the cross correlation coefficient. In this case, the time delay (> 0) ranked in the first peak corresponds to the first sound source SS1, and the time delay (<0) ranked in the second peak is in the second sound source SS2. It corresponds. Therefore, even in this time domain, the time delay ranked to the first peak of the cross correlation coefficient is continuously added to the segment of the same sound source SS1.

On the other hand, the time delay ranked in the second peak of the cross-correlation coefficient when the time index number is Ti1 is regarded as an initial value for the second sound source SS2. In the subsequent time domain, the time delay ranked at the second peak of the cross-correlation coefficient is added to the segment of the sound source SS2.

[Time index number Ti2 or later]
When the output of the first sound source SS2 stops, the number of sound sources SS2 becomes one after that. In this case, the time delay on the X-axis is represented only by the one ranked in the first peak of the cross correlation coefficient. However, in the subsequent time domain, the time delay ranked to the first peak corresponds to the second sound source SS2, so from here the time delay ranked to the first peak of the cross-correlation coefficient Is added to the segment of the sound source SS2.

[Separation by sound source]
FIG. 10 is a diagram showing an example in which the variation in time delay on the X axis shown in FIG. 9 is separated into sound source segments. Among these, (A) in FIG. 10 shows the segment of the first sound source SS1, and (B) in FIG. 10 shows the segment of the second sound source SS2.

In this way, in this embodiment, the time delay variation on each axis can be separated into segments for each sound source. Although the X axis is shown here, the Y axis can be similarly divided into segments for each sound source.

[Example of application to moving sound source]
FIG. 11 is a diagram illustrating an example when the simplified model is applied to a moving sound source. In this application example, two moving sound sources SS1, SS2 are arranged in the observation horizontal plane PL. Further, the angles (X axis and Y axis) of the microphone unit 10 are different from the simplified model.

[Execution condition 2]
For example, of the two sound sources SS1 and SS2 arranged in the anechoic room AR, the first sound source SS1 is moved from the vicinity of one wall toward the other wall and then moved again to the vicinity of one wall. Let Conversely, the first sound source SS2 was moved from the vicinity of the other wall toward one wall and then moved to the vicinity of the other wall. Further, these two sound sources SS1, SS2 were moved in parallel at the same time.

[Results of time delay fluctuation]
FIG. 12 is a diagram illustrating a variation in time delay in the X-axis and the Y-axis. Among these, (A) in FIG. 12 shows the fluctuation of the time delay on the X axis, and (B) in FIG. 12 shows the fluctuation of the time delay on the Y axis. In the figure, the horizontal axis represents the number of time indexes, and the vertical axis represents the ranked time delay τ. In addition, here, the time delay ranked from the top first peak to the third peak is displayed. Similarly, in FIGS. 12A and 12B, white circles indicate the time delays ranked in the first peak of the cross-correlation coefficient, and hatched circles indicate the cross-correlation coefficient. The time delay ranked in the second peak is shown. Black circles indicate time delays ranked in the third peak.

When multiple sound sources SS1 and SS2 move, even in an ideal environment such as an anechoic room AR, if the time delay variation is extracted in the time domain, the cross-correlation coefficients at the first to third peaks Multiple time delays are observed, ranked up to. Even in this case, by applying the sound source separation processing according to the present embodiment, it is possible to separate the time delay variation on each axis into segments for each sound source.

[Example of segment separation]
FIG. 13 is a diagram illustrating an example of segment separation along the X axis. Among these, (A) in FIG. 13 represents a segment corresponding to the sound source SS1, and (B) in FIG. 13 represents a segment corresponding to the sound source SS2. Similarly, the horizontal axis in the figure indicates the number of time indexes, and the vertical axis indicates the ranked time delay. In FIG. 13, black rhombus marks shown in (A) and (B) indicate τ _virtual (virtual time delay) used in the segmentation process (the same applies hereinafter).

In this way, by applying the sound source separation processing of the present embodiment, even when the time delay variation on each axis is complicated (FIG. 12), these can be separated into segments for each sound source. I understand that there is.

[Example of application to an airfield]
Up to this point, an example of application to a simplified model using an anechoic chamber AR has been described. In the following, an example of application when an actual airfield is used as an observation space will be described. Here, the method of this embodiment is applied to noise data measured by the microphone unit 10 with the observation point being the side of the runway 25 (FIG. 1) in an actual airfield. Note that the positional relationship between the runway 25 and the taxing road 30 may be different from that shown in FIG.

[Results of time delay fluctuation]
FIG. 14 is a diagram showing a result of fluctuation in time delay in the X axis and the Y axis for the measured data. 14A shows the time delay variation on the X axis, and FIG. 14B shows the time delay variation on the Y axis. In addition, here, the fluctuation results in the situation where the landing sound is observed in the middle of the taxing sound are shown. The white circles in FIG. 14 indicate time delays ranked in the first peak of the cross-correlation coefficient, and the hatched circles indicate time delays ranked in the second peak of the cross-correlation coefficient. The black circles indicate time delays ranked in the third peak of the cross correlation coefficient.

As shown in FIG. 14, when the fluctuations in the time delay on each axis are extracted in the time domain for the actually measured data in the airfield, the cross-correlation coefficients are first to third at many points. Multiple time lags ranked to peak have been observed. Further, even if the entire fluctuation result of the time delay is looked at in a post-processing manner, it can be understood that it is extremely difficult to identify from where to where the fluctuation is due to the same sound source.

In addition, in this embodiment, by performing segmentation by real-time processing as described above, it is possible to separate the time delay variation on each axis into segments for each sound source.

[Example of segment separation]
FIG. 15 is a diagram illustrating a result of separating the time delay variation on the X-axis and the Y-axis into segments for the measured data. Among these, (A) and (B) in FIG. 15 show examples of time-delayed segment separation on the X axis, and (C) and (D) in FIG. 15 show examples of time-delayed segment separation on the Y-axis. ing. Similarly, the white circles in FIG. 15 indicate the time delays ranked in the first peak of the cross correlation coefficient, and the hatched circles are ranked in the second peak of the cross correlation coefficient. Shows a time delay. The black rhombus marks indicate τ _virtual (virtual time delay) used in the segmentation process.

Also for the actual measurement data, by executing the segmentation process described above, the time delay variation is separated into, for example, four segments X1, X2, X3, and X4 on the X axis, and three segments Y1 and Y2 on the Y axis. , Y3.

Here, since the segments are basically separated by sound source, the number of segments after separation basically represents the number of sound sources. However, in view of the situation where landing sound is generated during the generation of taxing sound, noise with a large sound pressure level in the middle (landing sound) becomes dominant, and the continuity of the time delay from before is broken, As in the separation example of FIG. 15, it is fully possible that the number of segments does not match the number of sound sources.

In the present embodiment, segments separated on each axis are further integrated with the same sound source between different axes.

[Segment integration processing]
That is, the segment integration processing is executed by the above-described separated sound source integration unit 120. In this process, the segments separated on each axis as described above are combined between different axes. In the present embodiment, segments between different axes are integrated using fluctuations in the cross-correlation coefficient R (τ).

FIG. 16 is a diagram showing a variation pattern of the cross-correlation coefficient R (τ) of each segment. Among these, (A), (B), (C) in FIG. 16 show the variation pattern of the cross-correlation coefficient R (τ) of the X-axis segment, and (D), (E), (F) in FIG. Indicates a variation pattern of the cross-correlation coefficient R (τ) of the Y-axis segment. Also, the white circle in the figure indicates the first peak value of the cross-correlation coefficient R (τ), the hatched circle indicates the second peak value, and the black circle indicates the third peak value. Yes.

In the present invention, when a cross-correlation coefficient R (τ) is considered with respect to a certain segment rather than a time delay τ, a segment considered to be the same sound source has a very close variation pattern of R (τ). ing. This is because the sound generated from an actual sound source has fluctuations such as changes in the engine output, or is affected by the weather before reaching the observation point. It is thought to be caused by appearing in

Therefore, when a set of segments exists between different axes in a certain time domain, the normalized cross-correlation coefficient R (0) when τ = 0 is calculated, and segments having the same sound source are normalized to each other. Correlation coefficient R (0) increases to some extent. For this reason, in this embodiment, combinations of segments satisfying R (0)> 0.9 are integrated as those of the same sound source. In addition, when there are a plurality of combinations in which R (0)> 0.9, the combination in which R (0) has the maximum value is integrated.

[Normalized cross-correlation coefficient]
FIG. 17 is a diagram illustrating a calculation example of a normalized cross-correlation coefficient between overlapping segments in the time domain. In FIG. 16, segments on the X axis are arranged in the vertical direction and segments on the Y axis are arranged in the horizontal direction, and the normalized cross-correlation coefficients between the segments are shown in a 4 × 3 matrix.

[Example of segment integration]
In FIG. 17, it can be seen that if a combination having a normalized cross-correlation coefficient greater than 0.9 is selected, four combinations of X1-Y1, X2-Y2, X3-Y3, and X4-Y3 can be obtained. Reflecting this result, an example of segment integration between axes is shown in FIG. 16 by dashed arrows.

〔inspection result〕
As a result of comparison with the noise event at the time of actual measurement, it was confirmed that the combination of X1-Y1, X3-Y3, and X4-Y3 was actually a taxing sound and X2-Y2 was a landing sound.

On the other hand, in FIGS. 16A and 16B, the peak value of the taxing sound is not detected between the time indexes Tia and Tib because the actual noise level of the landing sound is higher than the noise level of the taxing sound. This is because it was large enough.

As described above, in the identification method of the present embodiment, the time delay is divided into sound source-specific segments by extracting the time delay variation in the time domain due to the peak of the cross-correlation coefficient on each axis, and The segments between the axes are integrated using the variation of the cross-correlation coefficient for each segment. This makes it possible to separate the sound arrival directions of the sound sources from a plurality of noise sources that are generated simultaneously in the observation space.

In addition, the number of aircraft that emit significant noise levels that could not be identified only by the method using only the first peak of the cross-correlation coefficient and the fluctuation of the sound pressure level by the identification method of this embodiment. Can be recognized automatically. Thereby, for example, in noise level evaluation of single noise, information for inferring the influence of background noise can be obtained. At the same time, since the direction of arrival of each sound can be obtained, it is possible to reliably identify each sound source by using the structure information of the airport such as a runway, a taxiway, and a surrounding road.

In the above-described one embodiment, the observation on the Z-axis is omitted for simplification, but it is naturally possible to use the three-axis correlation of the X-axis, the Y-axis, and the Z-axis when implementing the present invention. . In particular, with regard to time delay on the Z-axis, the peak tendency of the cross-correlation coefficient mainly due to ground reflection is prominent, and if such a tendency is utilized, the ground noise identification of the aircraft can be made more accurate. Very useful in terms.

The disclosed invention can be implemented with various modifications without being limited to the above-described embodiments. In one embodiment, the airfield is the target area. However, the noise observation apparatus and the noise observation method of the present invention can use an observation space (target area) other than the airfield.

In addition, the conditions (α, β, 0.9) related to the formation of segments and the integration of segments mentioned in one embodiment are examples, and the setting of the conditions is matched to the characteristics of the observation target area and the noise source. Can be changed as appropriate.

In one embodiment, the cross-correlation coefficient is calculated for each scheduled interruption, but the interval of “every scheduled” may not be constant. For example, the calculation is performed with an interval of 200 ms at a certain fixed time, but the next fixed time may be after 100 ms shorter than 200 ms, or conversely after 300 ms longer than 200 ms.

In one embodiment, a taxing sound and a landing sound are exemplified as the plurality of noises, but the plurality of noises may be other combinations. Further, the disclosed invention can be applied even when three or more noises are generated simultaneously.

[Explanation of symbols]
DESCRIPTION OF SYMBOLS 10 Microphone unit 100 Observation unit 102 Noise event detection part 106 Arrival direction vector calculation part 110 Sound source separation processing part 112 Cross correlation coefficient calculation part 114 Peak search processing part 116 Segmentation processing part 120 Separation sound source integration part 122 Normalization mutual phase relation Number calculation unit 124 Segment integration processing unit 130 Identification result output unit

Claims (12)

1. Using two microphones arranged at intervals on multiple axes defined in the observation space where multiple sound sources exist, the cross-correlation coefficient of the sound that reaches each microphone for each axis is fixed at regular intervals. A calculating means for calculating
   For a plurality of time delays in which the cross-correlation coefficient calculated every fixed time by the calculating means shows a peak tendency, a plurality of time delay fluctuations extracted in descending order of the cross-correlation coefficient are collected in the time domain, and each axis is An aggregation means for forming a set of continuous time delays;
   A noise observation apparatus comprising: an integrating unit that combines time delay sets having the same sound source from a cross-correlation between different axes with respect to a set of time delays for each axis formed by the collecting unit.
2. The noise observation apparatus according to claim 1,
   The aggregation means includes
   Prior to the formation of a set of continuous time delays, it is confirmed whether or not a plurality of time delays showing peak trends at regular intervals are initial values for each sound source.
3. In the noise observation apparatus according to claim 1 or 2,
   The aggregation means includes
   Among a plurality of time delays that show a peak tendency at each fixed time, if there is a unique value of a time delay in which the difference from a specific time delay that showed a peak tendency before the fixed time is less than a predetermined threshold, the unique value Is added to the same set as the specific time delay, and when the unique value does not exist, the virtual time delay is calculated by the least square method using at least the specific time delay and the previous time delay.
4. In the noise observation apparatus according to claim 3,
   The aggregation means includes
   In the case where the only value does not exist because there are a plurality of time delays in which the difference from the specific time delay is less than a predetermined threshold, a specific value in which the difference from the virtual time delay is less than the specific threshold If there is, the specific value is added to the same set as the specific time delay, while the specific time does not exist, or the difference from the specific time delay among a plurality of time delays is less than a predetermined threshold If the unique value does not exist because there is no time delay, the virtual time delay is added to the same set as the specific time delay.
5. The noise observation apparatus according to claim 4,
   The aggregation means includes
   When the virtual time delay is continuously added to the set for a predetermined number of times at regular intervals, the virtual time delay for the predetermined number of times is deleted and the formation of the set is terminated.
6. The noise observation apparatus according to claim 5,
   The aggregation means includes
   If the number of time delays included in the set is equal to or less than a specified number as a result of deleting the predetermined number of virtual time delays and completing the formation of the set, the set is invalidated.
7. Using two microphones arranged at intervals on multiple axes defined in the observation space where multiple sound sources exist, the cross-correlation coefficient of the sound that reaches each microphone for each axis is fixed at regular intervals. A calculation step for calculating
   For a plurality of time delays in which the cross-correlation coefficient calculated every fixed time in the calculation step shows a peak tendency, a plurality of time delay fluctuations extracted in descending order of the cross-correlation coefficient are collected in the time domain, and each axis is An assembly step that forms a continuous time-delayed set;
   A noise observation method comprising: an integration step of combining time delay sets having the same sound source from cross-correlation between different axes with respect to a set of time delays for each axis formed through the aggregation step.
8. The noise observation method according to claim 7,
   Prior to the formation of a set of continuous time delays in the assembly step, the method further includes a confirmation step of confirming whether or not a plurality of time delays showing a peak tendency at regular time intervals become initial values for each sound source.
9. The noise observation method according to claim 7 or 8,
   The aggregation step includes
   Judgment is made whether there is a single value of the time delay that is less than the predetermined threshold, and the difference from the specific time delay that showed the peak tendency before the fixed time is less than a predetermined threshold among the multiple time delays that show the peak tendency at each fixed time And steps to
   Adding the unique value to the same set as the specific time delay if it is determined that the unique value exists;
   If it is determined that the unique value does not exist, at least a step of calculating a virtual time delay by a least square method using the specific time delay and a previous time delay is included.
10. The noise observation method according to claim 9,
    The aggregation step includes
    The unique value does not exist because there are a plurality of time delays in which the difference from the specific time delay is less than a predetermined threshold, or the difference from the specific time delay is a predetermined value among the plurality of time delays. Determining whether there is no unique value because there is no time delay that falls below the threshold of
    When it is determined that the only value does not exist because there are a plurality of time delays in which the difference from the specific time delay is less than a predetermined threshold, the difference from the virtual time delay is less than the specific threshold. Determining whether there is a specific value,
    If it is determined that there is the specific value, adding the specific value to the same set as the specific time delay;
    When it is determined that there is no specific time, or there is no time delay in which a difference from the specific time delay is less than a predetermined threshold among a plurality of time delays, and the unique value does not exist If so, the method further includes adding the virtual time delay to the same set as the specific time delay.
11. The noise observation method according to claim 10,
    In the aggregation step, when the virtual time delay is continuously added to the set at a fixed number of times at regular intervals, an end determination step for deleting the virtual time delay for the predetermined number of times and terminating the formation of the set is further included Have.
12. The noise observation method according to claim 11,
    In the end determination step,
    If the number of time delays included in the set is equal to or less than a specified number as a result of deleting the predetermined number of virtual time delays and completing the formation of the set, the set is invalidated.

Images