KR20100001726A - Method for determining the representative point of cluster and system for sound source localization - Google Patents

Abstract

PURPOSE: A method for determining a representative point of a cluster and a system for sound source localization are provided to improve sound source localization estimation by solving problem on a position discordance. CONSTITUTION: A plurality of coordinates is selected within a cluster area(S131). The sum total of distance with the coordinate within the same cluster is computed based on the specific coordinate and the specific coordinate in which value is minimized pitches as the representative point of cluster. The specific coordinate is selected between the coordinate of the selected multiple(S132). The sum total of distance within the same cluster is calculated based on the selected specific coordinate(S133). The sum total of distance with coordinates is calculated by varying the specific coordinate. The specific coordinate in which the sum total of distance was minimized pitches as the representative point of cluster(S134).

Description

Representative point selection method for sound source position estimation and sound source position estimation system using the method {Method for Determining the Representative Point of Cluster and System for Sound Source Localization}

The present invention relates to a method for estimating the position of a sound source using a plurality of microphones, and a sound source position estimation system using the method.

The position of the sound source can be estimated by using the power, time difference, phase, etc. of the sound input to the microphone. Since a single microphone cannot accurately estimate the position or direction of the sound, a plurality of microphones are used.

1 is a view showing the arrangement of a plurality of microphones according to the prior art. Figure 1 (a) is a view showing the arrangement of the microphone linearly, Figure 1 (b) is a view showing the arrangement of the microphone in a circle, Figure 1 (c) is a view showing the arrangement of more than 100 microphones to be.

Microphone arrays can be applied to many applications, including sound enhancement techniques that amplify only the sound that occurs at a specific speaker or location, and sound source localization that tracks the location as the speaker speaks. Technology, and a source separation technology that separates a mixed voice by each speaker's voice by speaking several speakers at the same time.

The most important aspect of the sound source position estimation technique above is estimation performance, which is determined by factors such as microphone performance and number, microphone placement, noise and echo, and number of speakers. Degradation occurs.

That is, when the microphone performance is excellent or the number of microphones is large, the sound source position estimation performance is improved, but the larger the noise and echo, the lower the estimation performance. In addition, performance can be improved through microphone placement suitable for a specific application, but performance decreases as ambiguity increases as the number of talkers increases.

You can increase the number of microphones to improve the position estimation performance of the sound source, but you cannot place a large number of microphones in every application. Therefore, in order to ensure sound source estimation performance with a relatively small number of microphones (4 to 8), it is necessary to select an appropriate placement method of the microphone and a sound source position estimation algorithm suitable for each application.

The sound source position estimation technique using the microphone array in the three-dimensional space is summarized into three types as follows.

1. Method using time difference of arrival (TDOA)

2. Method using steered beam former

3. Method using high resolution spectral estimation

Documents that may refer to the description related to the sound source position estimation technique as described above are as follows. In addition, it contains a description of the Steeped Response Power (SRP), Steeped Response Power-phase transform (SRP-PHAT) algorithm to be used in the process of the present invention.

[1] M. Omologo and P. Svaizer, “Use of the crosspower-spectrum phase in acoustic event location”, IEEE Transactions on Speech and Audio Processing, vol. 5, no. 3, pp. 288-292, 1997.

[2] R. Ben, K. Waleed, and S. Claude, “Real time robot audition system incorporating both 3D sound source localization and voice characterization”, IEEE International Conference on Robotics and Automation, pp. 4733-4738, 2007.

[3] M. Togami, T. Sumiyoshi, and A. Amano, “Stepwise phase difference restoration method for sound source localization using multiple microphone pairs”, International Conference on Acoustics, Speech, and Signal Processing, vol. 1, pp. 117-120, 2007.

[4] J. H. DiBiase, “A High-Accuracy, Low-Latency Technique for Talker Localization in Reverberant Environments Using Microphone Arrays”, Ph.D. thesis, Brown University, 2000.

[5] R. Schmidt, "Multiple emitter location and signal parameter estimation", IEEE transactions on antennas and propagation, vol. 34, pp. 276-280, 1986.

[6] A. Johansson, G. Cook, S. Nordholm, “Acoustic direction of arrival estimation a comparison between ROOT-MUSIC and SRP-PHAT”, IEEE Region 10 Conference on Convergent Technologies for the Asia-Pacific, vol. B, pp. 629-632, 2004.

Of the above three sound source position estimation techniques, it will be described mainly in the above method 1 and 2 related to the present invention. First, the method using the arrival delay time will be described with reference to the drawings.

2 is a diagram illustrating a sound source generation and arrival delay time in a two-dimensional space.

As shown in FIG. 2, the sound generated from the sound source 20 located in a specific part of the space is input to the two microphones 22 and 24 planarly. Since it is assumed to be a two-dimensional space, the sound is input in a plane. In FIG. 2, the sound reaches the microphone 2 24, which is closer to the sound source 20, and arrives at the microphone 1 22 as late as the arrival delay time τ.

In Figure 2, what we want to know is to know the direction of the sound source, that is, the angle θ value between the two microphones 22 and 24 and the sound source 20. In FIG. 2, the expression (1) can be obtained by using the distance between the two microphones (d _mic ) and the distance traveled by the sound source by the arrival delay time τ.

c is the speed of sound. By modifying Equation 1 above as Equation 2 below, an angle θ between two microphones and a sound source can be obtained.

In Equation 2, the speed c of sound and the distance d _mic between two microphones are known, but the arrival delay time τ is unknown. One method of calculating τ is the Generalized Cross Correlation (GCC) method.

3 is a diagram illustrating a method of estimating arrival delay time using a Generalized Cross Correlation (GCC) method. FIG. 3A is a diagram illustrating signals input to two microphones 22 and 24, and FIG. 3B is a diagram showing cross correlation calculation results of signals input to two microphones.

The GCC method uses cross-correlation between signals input to two microphones, and performs a cross-correlation operation while moving two signals on a time axis, and estimates a shift value when the value reaches the maximum as an arrival delay time.

When a signal delayed by +2 seconds between two microphones 22 and 24 is inputted as shown in FIG. 3 (a), a cross-correlation operation is performed while moving the two signals on the time axis, thereby obtaining a graph as shown in FIG. have. As shown in (b) of FIG. 3, since the highest value is shown in the value of -2, the arrival delay time is -2.

By using the method illustrated in FIGS. 3 (a) and 3 (b), the arrival delay time τ is obtained and substituted into Equation 2 to obtain the angle θ value between the microphones 22 and 24 and the sound source 20. It can be estimated.

4 is a diagram schematically illustrating a sound source direction estimation method using four microphones.

When the arrival delay time between two microphones measured when a sound source is generated is τ, a hyperbolic is made by taking all the points where the arrival delay time can be τ in three-dimensional space. Using two more microphones, using a total of four microphones, two hyperbolic is formed as shown in FIG. As shown in FIG. 4, a straight line is formed at a portion where the two hyperbolic surfaces meet, and this direction is assumed to be a sound source direction in three-dimensional space.

However, using four microphones as shown in Figure 4 can only know the direction of the sound source and does not know how far apart. In addition, since there are two straight lines generated by two hyperbolic, it is not possible to know whether the direction in which the sound is generated is forward or backward. In order to solve this problem, the number of microphones should be increased, which will be described in FIG. 5.

5 is a diagram schematically illustrating a sound source position estimation method using eight microphones.

As shown in FIG. 5, when eight microphones are used, two straight lines are created for every four pairs of microphones, and one of the two straight lines meets or one of the two closest lines exists. We estimate it as the location where sound source occurred in space.

However, the method shown in FIG. 5 has a disadvantage in that it is difficult to implement the method shown in FIG. 5 with a single device because the microphone must be located in two physically separated areas. That is, when four microphones are one set, two sound source position estimation apparatuses equipped with such microphone sets should be installed and separated into spaces. In addition, there is a disadvantage that the sound source position estimation performance is sharply degraded when several speakers are speaking at the same time.

A method using a steered beam former among the sound source position estimation techniques using the microphone array described above is as follows.

6 is a diagram illustrating the steered beamformer.

As shown in FIG. 6, the sound source position estimation method using a steered beamformer adds signals from multiple microphones when sound is generated at an arbitrary point to find out where the sound is generated. It is a way.

When adding each signal, various manipulations can be applied, which can be classified into Delay-and-sum beam forming waterproof formula and Weight-and-sum beam forming waterproof formula.

First, the Delay-and-sum beam forming waterproof equation will be described. Assuming that a sound source occurs at any point q _s in three-dimensional space, the distance between q _s and each microphone is different. That is, assuming that the arrival delay time of the signal arriving at the n-th microphone at any one point q _s is τ _n , it is expressed by the equation considering the arrival delay time.

Each microphone is time-synchronized to account for the arrival delay of the sound source, which reduces ambient noise and amplifies the original signal.

The weight-and-sum beam forming method adds one operation to the delay-and-sum beam forming method. Due to the nature of the input device or the location of the microphone and the sound source, certain microphones may have more reliable information, and another microphone may contain less reliable information than certain microphones.

For example, if you have eight microphones, microphone 1 has the best performance, or if microphone 1 is directly exposed to the sound source, microphone 1 provides more important information than the other seven microphones. It is likely to contain. In such a case, when the signal input to each microphone is calculated, performance can be expected to be improved by giving a weight to the signal input to the first microphone, which is expressed as follows.

Compared with the delay-and-sum beam forming method described above, it can be seen that a weight filter called G _n (ω) is added. In addition, unlike Equation 3, Equation 4 is composed of capital letters, indicating that it is expressed in the frequency domain. This is because the Fourier transform transforms the signal axis from the time axis to the frequency axis, and the frequency axis is easier to manipulate the signal than the time axis.

FIG. 7 is a diagram illustrating a spatial power graph measured using a steered response power (SRP) method. FIG.

One of the methods used to track the position of a sound source using a signal received by a beam former is a steep response power (SRP). SRP is a method of obtaining the output power of the beam former for all positions in space, and estimating the position where the output power is maximum as the generation position of the sound source. Equation for calculating the output power of the beam former is as follows.

와 같다.In the above equation, Ψ _lk (ω) is a weight function,

Same as

FIG. 7 illustrates an example of obtaining power for all points in space where a sound source is estimated using the above equation. In FIG. 7, the most raised portion corresponds to the coordinate of the point where the output power is calculated to be high.

If the position where the output power is maximum is q _s ', q _s ' can be obtained by the following Equation 6.

8 is a graph illustrating performance comparison of sound source position estimation algorithms.

The algorithms compared in FIG. 8 are Steered Response Power (SRP), Generalized Cross Correlation-PHAse Transform (GCC-PHAT), and Steeped Response Power-PHAse Transform (SRP-PHAT).

As described above, the SRP method is a method of obtaining the output power of the beam former for all positions in the space, and estimating the position where the output power is maximum as the generation position of the sound source. The SRP-PHAT method is a method of applying a phase transform (PHAT) filter to the SRP algorithm. The SRP-PHAT method divides the space to be searched into small blocks having a predetermined size and calculates the output power at the position of each block to represent the largest output power. This is a method for estimating that a sound source has occurred at a location.

Since the SRP method or the SRP-PHAT method needs to calculate the output power of every block in space, it requires a large amount of computation that is difficult to use in a real-time system. Therefore, a method to reduce the computational amount for use in a real-time system has been sought, and one of them is to cluster the space into multiple blocks.

The GCC-PHAT method is a method in which an arrival delay time is obtained by using cross correlation between signals input to a microphone, and then an additional operation such as a weight filter is performed.

The horizontal axis (x axis) of FIG. 8 represents an error threshold value, and the vertical axis (y axis) represents a sound source position estimation error rate according to each error threshold value. In FIG. 8, when the error threshold value is 0.5 meters (that is, when the estimated sound source position and the actual sound source position are within 0.5 meters, the position of the sound source is accurately estimated), the error rate of the GCC-PHAT method is 70%. The SRP method has an error rate of about 24% and the SRP-PHAT method has a 0% error rate. As can be seen in Figure 8, in general, the SRP is better than the GCC-PHAT performance, it can be seen that the SRP-PHAT is better than the SRP.

9 is a diagram illustrating a representative point selection method of clustering according to the related art.

When using an SRP method or an SRP-PHAT method having a smaller error rate than the GCC-PHAT method, a clustering operation for splitting a location of a sound source into a plurality of block spaces is performed. As a criterion for dividing into a plurality of block spaces, various methods can be used, but a clustering method of dividing points having the same arrival delay time into one block space has been widely used.

In order to apply the SRP and SRP-PHAT methods to a specific location in space, the coordinates of the clustered blocks must be recorded and the recorded coordinates are the representative points of the block. The most common method of selecting representative points is to use the center of gravity.

9 illustrates a triangle as an example of a clustered block. Since the location of the sound source is estimated in space, the clustered blocks may be three-dimensional (3D), but are illustrated as two-dimensional triangles in FIG. 9 for convenience of description.

As shown in FIG. 9, the point where the line segments connecting the midpoint of each side of the triangle and the opposite vertex intersect is the center of gravity of the triangle. The coordinates of the center of gravity are then taken as representative points of the clustered blocks.

10 is a view showing a problem of the representative point selection method according to the prior art.

Shapes of spatial blocks having the same arrival delay time between specific microphone pairs may be generated in various ways. Thus, if a clustered space block is formed as shown in FIG. 10, the center of gravity of the block is formed outside the block. In FIG. 10, the wedge-shaped spatial block has been described as an example, but the spatial block shape in which the center of gravity is formed outside the block may exist in various ways.

If a representative point is formed on the outside of the block, the representative point is insufficient to represent the position of the block. When the output power as shown in FIG. 7 is obtained for each block in the clustered space using the SRP or SRP-PHAT method, the position and difference of the specific block represented by the point where the output power is maximized and the center of gravity occur. Because it is easy.

11 is another diagram illustrating a problem of the representative point selection method according to the prior art.

As shown in FIG. 11, four microphones and clusters a, b, c are shown. 11 and 2 are black in the microphone illustrated in FIG. 11 because the spaces are clustered based on arrival delays based on 1 and 2 among the 1 to 4 microphones.

When a cluster having a shape like that shown in FIG. 11 is formed in a space, the centers of clusters a, b, and c are obtained, and the representative points a, b, and c are respectively located outside the region of the cluster. do. Representative point a of cluster a exists in the region of cluster b, and representative point b of cluster b exists in the region of cluster c.

According to the cluster and representative point selection method of the shape shown in Figure 11, the sound source position estimation process will be described as follows.

First, the center of gravity of each cluster is stored in the position reference table as the representative point of each cluster. Then, when a sound occurs at a specific point in space, the arrival delay time (TDOA) between the sound waves reaching the plurality of microphones is measured for each microphone pair. Based on the measured arrival delay time, find out the representative point of each cluster by referring to the location reference table, and select the location of the representative point with the highest output power by performing SRP or SRP-PHAT method on each found point. do. Then, the position of the selected representative point is estimated as the position where the sound is generated. In Fig. 11, it is assumed that the representative point b is selected as the generation position of the sound source because the output power is the highest. Actually, the position where the sound is generated will exist inside the cluster b, but since the representative point b exists inside the cluster c, the problem is that the position of the sound source finally estimated by the sound source position estimation algorithm will exist in the region of the cluster c. do. That is, the method of obtaining the representative point of the cluster with the center of gravity causes a difference from the actual sound source generation position, thereby degrading the accuracy of the sound source position estimation.

Accordingly, an object of the present invention is to perform a more accurate output power calculation by solving the position mismatch between the representative point performing the operation and the figure forming the cluster in performing the SRP algorithm or the SRP-PHAT algorithm when the position of the sound source is estimated. The present invention provides a method for improving the accuracy of sound source position estimation.

Another object of the present invention, in estimating the position of a sound source using the SRP or SRP-PHAT method, the sound source position estimation using the above method of selecting a representative point that accurately represents the position of the cluster regardless of the shape of the cluster In providing a system.

Representative point selection method according to the present invention for achieving the above object, the step of selecting a plurality of coordinates in the cluster region and the sum of the distance to the other coordinates in the same cluster based on a specific coordinate of the selected plurality of coordinates And calculating specific coordinates of which the value is the minimum as a representative point of the cluster.

Preferably, the process of calculating the sum of distances with other coordinates includes selecting a specific coordinate from among a plurality of selected coordinates, and calculating a sum of distances with other coordinates in the same cluster based on the selected specific coordinates. And a step of calculating a sum of distances from other coordinates by changing a specific coordinate as a reference, and repeating by changing a specific coordinate as a reference to a process of calculating the sum of the distances. It is done.

In addition, in the sound source position estimation system using the representative point selection method according to the present invention, a plurality of microphones and spaces for receiving sound waves of sounds generated in the space are divided into a plurality of clusters, and each Selecting a plurality of coordinates in the region of the cluster, and determine the specific coordinates that the minimum value calculated by the sum of the distances with other coordinates in the same cluster based on a specific coordinate of the selected plurality of coordinates as the representative point of the cluster In addition, a sound source position estimating unit is used as coordinates for performing an algorithm for estimating a position of the sound source.

Preferably, the sound source position estimating unit divides a target space for performing sound source position estimation into a plurality of clusters, selects a plurality of coordinates within a region of each cluster, and selects a plurality of coordinates based on a specific coordinate among the selected plurality of coordinates. Position reference table creation unit for determining a specific coordinate whose minimum value is calculated as the sum of distances from other coordinates in the same cluster as the representative point of the cluster and recording them in a table; arrival delay time of sound waves reaching a plurality of microphones arrival delay determination unit for determining a time difference of arrival and a sound, a steep response power (SRP) algorithm or SRP-PHAT (Steered) for a cluster having the arrival delay time measured by the arrival delay time determination unit A response power-phase transform algorithm is performed and the coordinates of the cluster are determined by referring to the location reference table. It characterized in that it comprises a RP algorithm performing unit.

In performing the sound source position estimation, when dividing a space into a plurality of clusters using arrival delay times of sound waves reaching a plurality of microphones, a large number of clusters may be generated according to the number and number of microphones. In selecting a representative point of the cluster, when using the conventional method using the center of gravity, it is difficult to accurately estimate the position of the sound source when the center of gravity is located outside the cluster, depending on the shape of the cluster, By selecting a plurality of coordinates in a cluster according to the method and selecting a point at which the sum of the distances between the selected points is the minimum as the representative point of the cluster, accurate position estimation is possible as compared with the conventional method.

In the sound source position estimation according to the present invention, a representative point selection method and a sound source position estimation system using the method will be described with reference to the drawings.

12 is a diagram illustrating a sound source position estimation system according to an embodiment of the present invention.

As shown in FIG. 12, the sound source position estimation system according to the present invention includes a plurality of microphones 120 and a sound source position estimating unit 121, and the sound source position estimating unit 121 includes a position reference table preparing unit ( 122), the arrival delay time determination unit 125, and the SRP algorithm execution unit 126. In addition, the position reference table preparing unit 122 includes a clustering unit 123 and a representative point calculating unit 124.

The plurality of microphones 120 function to receive sounds generated in the space. In the present invention, at least two or more microphones are required because the position of the sound source is estimated using the time difference of the sound waves reaching the microphone.

The position reference table creating unit 121 calculates a representative point for each spatial cluster having the same arrival delay time of the sound waves.

The clustering unit 123 divides the space into regions having the same arrival delay time of sound waves at a specific frequency (clustering, clustering). The arrival delay time of the sound wave may vary depending on the frequency of the sound wave or the arrangement of the microphones. Therefore, after arrangement of the microphones, it is preferable to consider the possible frequencies of sound waves in clustering for spaces having the same arrival delay time. That is, the spatial cluster having the same arrival delay time when the expected frequency of the sound wave is a and the spatial cluster when the expected frequency is b will not be the same.

The representative point calculator 124 calculates a representative point of the clustered area. The method of calculating the representative point is based on a method of selecting a specific coordinate at which the sum of distances with all other coordinates in the same cluster is minimum based on the specific coordinate in the cluster. A more detailed description of the operation will be described later with reference to the drawings.

The arrival delay time determining unit 125 calculates the arrival delay time τ of sound waves arriving at a pair of microphones among the plurality of microphones 120. Since two or more microphones are used, a pair of two microphones is connected to one pair. Calculate the arrival delay of sound waves for all combinations of microphone pairs. At this time, the arrival delay time is calculated in consideration of the frequency of the sound wave.

The SRP algorithm execution unit 126 performs a function of performing the SRP algorithm for each cluster based on the location reference table. The execution algorithm is not limited to the SRP algorithm, but other algorithms such as SRP-PHAT are possible.

13 is a diagram illustrating a representative point selection method according to an embodiment of the present invention.

As shown in FIG. 13, a cluster having the same TDOA is specified (S130). In the present invention, since the space is divided using the arrival delay time, the SRP algorithm is selected for the divided cluster, and the position of the sound source is estimated. Therefore, the cluster needs to be specified first.

Thereafter, a plurality of coordinates are selected in the specified cluster region (S131), and one specific coordinate is selected from the selected coordinates (S132). For example, after a region having the same TDOA is bound to each other and designated as clusters A, B, C, D ..., a plurality of coordinates B-1, B-2, B-3, .. Select, and select B-1 among the selected coordinates first. In selecting a plurality of coordinates in a cluster area, the number of coordinates to be selected may be arbitrarily determined by a user performing the method according to the present invention. If the number of coordinates to be selected is large, the operation for selecting representative points of each cluster will be increased, but it may have a more accurate character of representative points. If the number of coordinates to be selected is small, the reverse effect will be obtained.

The sum of the distances between the selected specific coordinates and the other coordinates is calculated (S133). In the above example, if the specific coordinate selected is B-1, the distance between B-1 and B-2, the distance between B-1 and B-3, the distance between B-1 and B-4, ... . The distance sum operation is performed for each coordinate in the same cluster. That is, the sum operation of the distance between B-2 and other points B-1, B-3, B-4, ..., and B-3 and other points B-1, B-2, B-4 Calculate the sum of the distances between, ...) based on all points B-1, B-2, B-3, ... selected inside the cluster B.

After the sum operation of the distances based on all coordinates in the same cluster (S133) is performed, the reference point having the minimum sum of distances is determined as the representative point of the cluster (S134). When the result of calculating the sum of distances from other points based on B-2 has a minimum value, B-2 is regarded as the representative point of cluster B.

When the representative point is selected, the representative point of the cluster is recorded in the position reference table (S135). It is determined whether the location reference table for all the clusters in the space has been created (S136), and the representative point selection process for all the clusters in the space is repeated until the creation is made.

14 is a view showing a representative point selected according to an embodiment of the present invention.

As shown in FIG. 14, there is one blackened point in the center of the wedge-shaped figure, and there are a plurality of blurred points.

In Fig. 14, the blacked out points represent representative points selected according to the method of the present invention. In FIG. 10, the representative point is located outside the cluster, but in FIG. 14, the representative point of the cluster exists inside the cluster.

In FIG. 14, the plurality of blurred points represent a plurality of points selected in the cluster, and the coordinates selected according to the process of S131 of FIG. 13 are exemplarily illustrated.

15 is a diagram illustrating an embodiment of a method for performing sound source position estimation using the representative point selection method of the present invention.

As shown in FIG. 15, a cluster having the same arrival delay time TDOA is specified in a space in which the position of the sound source is to be estimated (S150). A plurality of coordinates are selected in the specified cluster, and the sum of distances with other coordinates is calculated based on the specific coordinates among the selected coordinates, and the coordinates having the minimum sum of the distances are determined as representative points (S151). The process of determining the representative point is as described above with reference to S131 to S134 of FIG. 13.

The process of recording the representative points determined for each cluster in the position reference table is repeated (S152), and the process of recording the representative points for all clusters in the space is completed (S153).

Subsequently, when a sound is generated in a specific part of the space (S154), the TDOA is measured for each combination of the microphone pairs (S155). To apply a location estimation algorithm to clusters with measured TDOA, we use a location reference table that contains a record of the coordinates of each cluster. When the SRP algorithm or the SRP-PHAT algorithm is performed on each of the plurality of clusters in the space (S156), and the cluster having the highest output power is selected (S156), the selected cluster is estimated as the position where the sound source is generated (S158). ).

Although the present invention has been described in detail through the representative embodiments, those skilled in the art to which the present invention pertains can make various modifications without departing from the scope of the present invention. I will understand. Therefore, the scope of the present invention should not be limited to the described embodiments, but should be defined by the claims below and equivalents thereof.

1 is a view showing an arrangement of a plurality of microphones according to the prior art.

2 is a diagram illustrating a sound source generation and arrival delay time in a two-dimensional space.

3 is a diagram illustrating a method of estimating arrival delay time using a Generalized Cross Correlation (GCC) method.

4 is a diagram schematically illustrating a sound source direction estimation method using four microphones.

5 is a view schematically showing a sound source position estimation method using eight microphones.

6 is a schematic representation of a steered beam former;

FIG. 7 is a diagram illustrating a spatial power graph measured using a steeped response power (SRP) method. FIG.

8 is a graph illustrating performance comparison of sound source position estimation algorithms.

9 is a diagram illustrating a representative point selection method of clustering according to the related art.

10 is a view showing a problem of the representative point selection method according to the prior art.

11 is yet another view showing a problem of the representative point selection method according to the prior art.

12 is a view showing a sound source position estimation system according to an embodiment of the present invention.

13 is a view showing a representative point selection method according to an embodiment of the present invention.

14 is a view showing a representative point selected in accordance with an embodiment of the present invention.

15 is a view showing an embodiment of a method for performing sound source position estimation using the representative point selection method of the present invention.

Claims (9)

In the method of selecting a representative point of a cluster related to the position estimation technique of a sound source, Selecting a plurality of coordinates within the cluster region; And Representing a sound source position estimation comprising calculating a sum of distances with other coordinates in the same cluster based on a specific coordinate among the selected plurality of coordinates and determining a specific coordinate whose value is minimum as a representative point of the cluster. Point selection method. The method of claim 1, wherein the determining of the representative point comprises: Selecting a particular coordinate from the selected plurality of coordinates; Calculating a sum of distances from other coordinates in the same cluster based on the selected specific coordinates; Calculating a sum of distances from other coordinates by changing specific coordinates as a reference; Repeating by changing specific coordinates based on a process of calculating a sum of distances; And And determining, as a result of the iterative operation, a specific coordinate having the minimum sum of distances as a representative point of the cluster. The method of claim 1, wherein the cluster, A representative point selection method of sound source position estimation, characterized in that the space is divided based on the arrival delay time that occurs when the sound wave reaches the microphone pair. The method of claim 1, wherein the selected plurality of coordinates, A representative point selection method of sound source position estimation, characterized in that the arrival delay time generated when a sound wave reaches a microphone pair is selected among the same points. The method of claim 1, And recording the representative points determined for each cluster in the form of a table. A plurality of microphones for receiving sound waves of sounds generated in space; And The target space to estimate the sound source position is divided into a plurality of clusters, and a plurality of coordinates are selected within the region of each cluster, and the distance from other coordinates in the same cluster is determined based on a specific coordinate among the selected plurality of coordinates. And a sound source position estimating unit configured to determine a specific coordinate at which the sum is minimized as a representative point of the cluster, and perform a position estimation algorithm of the sound source using the determined representative point. The method of claim 6, wherein the sound source position estimation unit, The target space to estimate the sound source position is divided into a plurality of clusters, and a plurality of coordinates are selected within the region of each cluster, and the distance from other coordinates in the same cluster is determined based on a specific coordinate among the selected plurality of coordinates. A position reference table creating unit which determines a specific coordinate at which the sum value is the minimum as a representative point of the cluster and writes it to a table; An arrival delay time determining unit determining a time difference of arrival of sound waves reaching the plurality of microphones when a sound is generated; And A steered response power (SRP) algorithm or a steered response power-phase transform (SRP-PHAT) algorithm is performed on the cluster having the arrival delay time measured by the arrival delay time determining unit, and the coordinates of the cluster are referred to the position. Sound source position estimation system comprising an SRP algorithm performing section that refers to the table. The method of claim 7, wherein the position reference table creation unit, A clustering unit for dividing spaces into spatial regions having the same arrival delay time of sound waves; And In calculating a representative point of each cluster divided by the clustering unit, a plurality of coordinates are selected within a specific cluster, and a sum of distances from other coordinates in the same cluster is calculated based on a specific coordinate among the plurality of coordinates. And a representative point calculator configured to determine the reference coordinate at which the sum is the minimum as a representative point of the cluster. The method of claim 6, wherein the plurality of coordinates to be selected in each of the cluster region, A sound source position estimation system, characterized in that the arrival delay time generated when a sound wave reaches a microphone pair is selected from the same point.

Images