JP2544173B2 - Sound receiving device - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a sound receiving device that intercepts an interfering sound incident from the outside and detects only a desired sound.

INDUSTRIAL APPLICABILITY The present invention is desired as an input microphone of a general voice recognition device in which it is theoretically desired to operate by suppressing acoustic signals other than the target sound pressure as much as possible, and it is also desired that the sound can be expanded without causing howling. As a microphone for conference phones or general public address systems, and as a microphone for various music or dramas where it is desired to separate and pick up the sounds of each part without interfering with each other. It is suitable for use as a super close-talking microphone with extremely high performance, which is desirable in high-temperature environments.

〔Overview〕

The present invention provides a sound receiving device having means for removing an external noise component contained in an acoustic signal detected by a central microphone, wherein a plurality of sub microphones including two sound receiving elements are arranged on a boundary surface surrounding the central microphone. By performing signal processing on the outputs of the sub microphones to obtain a noise component, noise coming from outside the boundary surface is removed and a super close-talking microphone is realized.

[Conventional technology]

The technology to block the external noise other than the desired sound and improve the signal-to-noise ratio is a fundamental one in acoustic engineering.
Conventionally, various methods have been proposed. If they are classified, (1) those that utilize direction selectivity (directional microphone), (2) those that utilize the difference in statistical properties between signal and noise (various filters), (3) adaptive signal It is considered that there are four types: processing technology (multi-channel denoising method) and (4) spherical wave effect (close-talking microphone).

In the case of the method (1), in the case of an acoustic signal that has to spread over a wide ratio band, the directivity actually obtained cannot be sharpened, and at most, it is limited to bidirectionality and its deformation degree. In general, the ability to improve the signal-to-noise ratio is generally extremely insufficient. Further sharpening of directivity is possible in principle by making the size of the entire microphone considerably larger than the wavelength, but it becomes impossible to realize constant directivity over the entire frequency band. .
Normally, noise has the main energy in the low frequency range. In order to deal with this, the size of the microphone is about several meters, and it cannot be put to practical use at all. Even if this is forced, the directional characteristic becomes abnormally sharp at a high frequency, which is unsatisfactory. In any case, (1)
It is difficult to achieve the purpose by the method of, and in reality, this method is regarded as inappropriate.

The method (2) focuses on the difference in the statistical properties of so-called signals and noise, particularly the difference in the power spectrum. The theoretical limit of this method is given by the well-known Winer filter, and it is proved from the information theory that it is impossible to improve the signal-to-noise ratio beyond that. For many of the acoustic signals we face, where the signal and noise spectra overlap, the Wiener filter has a fairly small limit, with very little improvement of at most a few dB. Therefore,
If possible, it is completely unsuitable for practical purposes where an improvement of several tens of dB is required.

In the method (3), the output of a microphone arranged near the noise source so that only the noise is picked up is passed through an adaptive filter having a variable weighting coefficient, and the filter output is mixed with the signal and the noise. Reduced from the main microphone output. The objective is achieved by controlling the variable weighting factors of the adaptive filter according to the algorithm so that the difference output is minimized, in other words, the noise component is minimized. As a device developed from this method, there is a device that sharply adaptively controls directivity by using a large number of microphones. In both cases, the signal-to-noise ratio is usually improved to 20 dB, and at the best, close to 30 dB, which is far superior to other methods.

However, it is a problem that the microphone must be placed near the noise source, and that when the directivity is adapted using a large number of microphones, it is necessary to know the distance and direction of the noise source in advance. These conditions are not always met, but rather unforgivable. The disadvantages are that there is this unrealistic limit and that the improvement of the signal-to-noise ratio is still insufficient for some purposes.

Method (4) is actually most often used,
It has the advantage of being light. However, since the change in the curvature of the spherical wave, which is not so large, is used, the signal-to-noise ratio actually obtained is larger than the method (1) but smaller than the method (3). Further improvement is required, but it is difficult in principle.

[Problems to be solved by the invention]

In the above-described conventional signal-to-noise ratio improving technique, the microphone used weights the signal and the external noise by utilizing the directivity and other characteristics, but essentially adds them linearly and converts them into an electric signal. . Therefore, the above methods (1) and (4) cannot obtain sufficient properties (if the "power" operation, which is a non-linear operation, can be applied to the microphone output, the directivity is sharpened at any time. However, such an operation cannot be used for the following reasons), and the methods (2) and (3) face a very difficult problem in information theory that the noise once mixed is separated and removed. . This linearity condition is important and must be satisfied by any sound receiving system. This is not only because it is a universal requirement in engineering, but because the sound receiving system must absolutely avoid nonlinear distortion.

An object of the present invention is to solve the above problems and to provide a sound receiving device that can pick up a desired acoustic signal with a large signal-to-noise ratio even in a noisy environment while maintaining linearity.

[Means for solving problems]

The sound receiving device of the present invention includes, as a sub-microphone group, sound receiving elements arranged at a plurality of points on a virtual boundary surface surrounding the central microphone, and as a signal processing means, external noise is generated for each sound receiving element. Sound pressure detecting means for detecting the sound pressure of, a first coefficient means for multiplying the value of this sound pressure by a coefficient, an averaging means for averaging the sound pressure values multiplied by this coefficient over the entire boundary surface, Delay means for delaying the signal output by the averaging means by a time corresponding to the distance between the central microphone and the sound receiving element, and noise subtracting means for subtracting the delayed output of the averaging means from the output of the central microphone. It is characterized by having.

The signal processing means further includes a time differentiating means for obtaining a time derivative of the sound pressure from the output of the sound pressure detecting means, and a second coefficient means for multiplying a value of the time derivative by a coefficient, and the averaging means has a coefficient of It is desirable to include means for adding the multiplied time derivative to the sound pressure value multiplied by the coefficient.

Further, the signal processing means further includes sound pressure gradient detection means for detecting the sound pressure gradient in the normal direction of the boundary surface, and third coefficient means for multiplying the value of this sound pressure gradient by a coefficient, and the averaging means is It is desirable to include means for further adding the sound pressure gradient value multiplied by the coefficient.

The first to third coefficient means may be arranged in the input signal path output signal of the respective associated means. The delay means may be arranged anywhere in the signal path from the sound receiving element to the noise subtraction means.

The boundary surface is preferably a spherical surface. To divide it, it is desirable to use regular polyhedrons. In this case, since the same amount of delay is given to each sound receiving element, one delay device can be arranged as a delay unit after the averaging unit.

If it is not a regular polyhedron, the coefficient values of the first to third coefficient means and the delay amount of the delay means are appropriately selected.

[Action]

The sound receiving device of the present invention focuses on spatial selectivity as a new fifth idea in order to avoid external noise while maintaining linearity. Here, the spatial selectivity is defined as a property that is sensitive only to a sound emitted from a sound source within a certain spatial area, but is insensitive to a sound incident from a sound source outside the area.

That is, considering a virtual boundary surface in the space, appropriately processing the outputs of the plurality of sub microphones arranged on it and the main microphone placed at one point on the boundary surface, and arriving from the outside of the boundary surface. It has a performance that is completely insensitive to the noise generated. That is, spatial selectivity is obtained. At the same time, the sound receiving device of the present invention has a very high sensitivity to a sound source extremely close to the sub microphone on the boundary surface,
Super close talk characteristics are obtained. Also, when the sub microphone on the boundary surface is omitted, the super close talk characteristic is similarly obtained in the direction of the lack.

〔Example〕

FIG. 1 is a block diagram of a sound receiving device according to the first embodiment of the present invention.

This sound receiving device includes a central microphone 100 and a group of sub-microphones 2 arranged around the central microphone 100.
00 and a real-time signal processing circuit 300 that removes an external noise component included in the output of the central microphone 100 by the output of the sub microphone group.

The sub microphone group 200 includes a plurality of points on a virtual boundary surface surrounding the central microphone 100, which is a positive M in this embodiment.
A set of two sound receiving elements 1- arranged at a central point of each surface of the face piece (M = 4, 6, 8, 12 or 20)
1, 2-1 to 1-M and 2-M are included.

The signal processing circuit 300 includes a pair of sound receiving elements 1-1 and 2
For each of -1 to 1-M and 2-M, adders 3-1 to 3-3 are provided as sound pressure detection means for detecting the sound pressure of external noise and first coefficient means for multiplying the value of the sound pressure by a coefficient. M and coefficient multipliers 5-1 to 5-M are provided, and an adder 10 and a coefficient multiplier 11 are provided as averaging means for averaging the sound pressure values multiplied by the coefficients over the entire boundary surface. The delay unit 12 is provided as a delay unit that delays the signal output by the unit by a time corresponding to the distance between the central microphone 100 and the sound receiving elements 1-1, 2-1 to 1-M, and 2-M. A subtractor 13 is provided as noise subtracting means for subtracting the output of the converting means from the output of the central microphone 100.

The signal processing circuit 300 is further provided with a differentiator 7-
1 to 7-M, and coefficient multipliers 8-1 to 8-M are provided as second coefficient means for multiplying the value of this time differential by a coefficient. Further, subtracters 4-1 to 4-M are provided as sound pressure gradient detecting means for detecting the sound pressure gradient in the normal direction of the boundary surface, and a coefficient is provided as a third coefficient means for multiplying the value of the sound pressure gradient by a coefficient. The multipliers 6-1 to 6-M are provided. Corresponding to this configuration, the averaging means includes adders 9-1 to 9-M as means for adding the time differential value multiplied by the coefficient and the sound pressure gradient value to the sound pressure value multiplied by the coefficient. .

FIG. 2 shows the arrangement of the central microphone 100 and the sub microphone group 200 when M = 12.

The central microphone 100 is arranged in the center of the regular dodecahedron, the sound receiving elements 1-1 to 1-12 are arranged outside the central portion of each surface of the regular dodecahedron, and the sound receiving elements 2-1 to 2- 12 are arranged inside the central part of the corresponding surface. In FIG. 2, twelve pairs of sound receiving elements 1-1, 2-1 to 1-12, 2-12.
Are shown as C _{0 to} C ₅ and D _{0 to} D ₅ .

Here, the spatial selectivity will be described. Since this is an unprecedented concept of sound reception, I will briefly explain that this is possible by going back to the wave equation.

FIG. 3 is a diagram for explaining the principle of space selectivity and shows various vectors at the continuous boundary surface Γ and the observation point p.

If there is a sound source intensity distribution ρ (, t) in space, the sound pressure ψ (, t) at the point p in Fig. 3 is the wave equation, Meet. However, is a coordinate vector of the point p, and t is time. In the region Ω closed by the virtual boundary surface Γ, starting from the equation (1), the Green function G (, t |
_O , t _O ), Is obtained. Here, ▲ ▼ _O is a surface element vector.

As a specific Green's function, G on the free-field of three-dimensional _{(, t | o, t o} ) = (1 / R) δ (R / c- (t-
t _o )), and using this to transform Eq. (2), Is obtained. Here, δ (...) Is the Dirac delta function, and _o and t _o are the position vector and time with respect to the sound source used in the definition of the Green's function (which does not necessarily match the actual sound source shown by + in FIG. 3). Also =-
_o , R = ||

(3) perform the integration of the expression of the second term on the right side, further outward normal vectors of the boundary surface and _o, ▲ according to the usual definition of surface elements vector ▼ = because it is _o dS,
Equation (3) is Can be rewritten.

From the definition of the δ function, the left side of the equation (4) means a sound coming from only the sound source in the closed space. It does not include sounds coming from external sound sources, that is, external noise.

FIG. 4 two-dimensionally shows the arrangement of the central microphone 100 and the sub microphone group 200. In this arrangement, the central microphone 100 causes the sound pressure ψ from the sound source 40 in the area to
(, T) is picked up, while ∂ψ / ∂n _o , ψ and ∂ψ / ∂t _o of the external noise 41 are measured by the sub-microphone group 200 appropriately distributed on the boundary surface Γ ( _o ). , (4), the system can be made spatially selective.

However, in practice continuously _{∂ψ / ∂n o, ψ, ∂ψ} / ∂
It is not possible to measure t _o . Therefore, the boundary surface is equally divided into M pieces, a representative point is selected at the center, and a sound receiving element is arranged at that point, and ∂ψ / ∂n _o , ψ∂ψ at the representative point.
We obtain / ∂t _o and approximate equation (4). In this case, the equation in the discrete distribution sound receiving element corresponding to the equation (4) is expressed by abbreviating the left side of the equation (4) by the symbol α (r _q , t), It is expressed as Here, the subscript i means a quantity related to the i-th sub microphone. Also, Is. Here, is a vector from the i-th sub microphone to the central microphone. Also, let the spherical surface be M
The surface segment typified by the i-th sub-microphone when equally divided into pieces is defined as Γ _i .

The subscript _toi = t−R _i / c at the right end of the equation (5) is a symbol [].
It means that the amount inside is earlier than t by R _i / c in time. That is, the quantities in the symbols [] must be time-adjusted after being observed and measured with a delay of R _i / c and then subtracted from the output ψ (o, t) of the central microphone. Since it is a simple configuration to make the amount of delay constant, it is convenient to make the boundary surface a spherical surface and place the central microphone at the center of the sphere. Select the boundary surface as a sphere,
Further, according to the division method by the regular polyhedron, a _i , b _i , and c _i are constant independent of i, which is more convenient.

FIG. 5 is a diagram showing x, y, z coordinate axes and related vectors when a spherical surface having a radius a is adopted as the boundary surface. O is the center of the sphere, and also serves as the origin here.
Q is a point sound source located at an arbitrary position on the x axis (either inside or outside the sphere). _q is the position vector of the point Q, is the point Q
To a point on the sphere, _o is a position vector of a point on the sphere, and | is a vector from that point to the center of the sphere, which has a relation of = _-o , || = | _o | = a. _o is a unit vector in the normal direction from the point on the sphere to the outside.

If the radius of the sphere is a, then R = || = a and _o
Since the directions of and are opposite, ( _.o ) =-a. Therefore, let the speed of sound be c, Becomes Therefore, equation (5) is Simplified as

A method of arranging the sub microphones that satisfies the above-described principle and a configuration for performing signal processing on the output of the sub microphones will be described. Further, the present invention is not limited to the spherical surface, because it is based on the idea of arranging the sub-microphones on the boundary surface of a closed arbitrary shape to achieve the object. _i , b _i , c
_i is, in the case of a spherical arrangement of sub microphones independently is constant in its number i, are exceedingly advantageous for constituting the device, since the spherical configuration method is determined that it would be actually used widely, the first The case where the spherical surface is used as the boundary surface will be specifically described.

The spherical surface is divided into M pieces with the same curved surface. For this reason, it is easy to think of starting from a regular polyhedron (however, it is known that there is a method of equal division other than regular polyhedron in geometry). In the case of regular polyhedron, only M = 4, 6, 8, 12, 20 exists. In FIG. 2 described above, the case of a regular dodecahedron is shown. The target spherical surface is inscribed in the dodecahedron and is equally divided into 12 pieces on a plane passing through the center O and each edge.

According to the equation (7), it is necessary to know the spatial differential ∂ψ _i / ∂n _oi of the sound pressure in the normal direction of the spherical surface. For this purpose, two minute sound receiving elements may be arranged at two points at a small distance Δr _o in the radial direction at each sound receiving position, and the difference between the outputs of the two may be divided by Δr _o . This is a well-known method that is naturally performed with, for example, a velocity microphone. The magnitude of Δr _o is
It may be selected to be a fraction of the highest frequency wavelength used.

The above-described first embodiment device performs signal processing according to equation (7). Explaining this content with reference to the principle described above, the spherical surface is first equally divided into M pieces. Sound receiving element 1 at the center of each equally divided surface and slightly outside
-1 to 1-M are arranged, and the sound receiving elements 2-1 to 2-1 are located slightly inside.
2-M is arranged. Sound receiving elements 1-1, 2-1 and 1-2
And 2-2, and the likewise similarly, M pairs of sound receiving elements up to 1-M and 2-M are placed at a distance of Δr _o in the normal direction at each position. The signal processing circuit 300 includes M signal processing units for each pair of sound receiving elements. Since the signal processing units have exactly the same structure, the signal processing unit related to the first sound receiving element 1-1, 2-1 pair will be described.

If the adders 3-1 add the outputs of the sound receiving elements 1-1 and 2-1 and the coefficient multiplier 5-1 multiplies by 1/2, the sound receiving element 1
The sound pressure component ψ _i (i = 1) of Expression (7) is obtained as the average of the outputs of −1 and 2-1. This is further differentiated by differentiator 7-1,
If multiplied by a factor multiplier 8-1 a / c, (7) the component (a / c) regarding the time derivative of the formula _{(∂ψ i / ψt o) |} i = 1 to the output of the coefficient multiplier 8-1 can get. On the other hand, the difference between the outputs of the sound receiving elements 1-1 and 2-1 is obtained by the subtractor 4-1 and multiplied by a / Δr _o by the coefficient multiplier 6-1. The term a (∂ψ _i / ∂n _oi ) | _{i = 1} related to is obtained. If these three types of components are algebraically added by the adder / subtractor 9-1, [a (∂ψ _i / ∂n _oi ) + ψ _i + (a / c) (∂ψ _i / ψt
_o )] is obtained.

The adder 10 adds the same number of M signals obtained by each signal processing unit.
, And multiplying by 1 / M in coefficient multiplier 11, the output is Becomes If this is further delayed by a / c by the delay unit 12 and subtracted from the sound pressure output ψ (o, t) of the central microphone 100 by the subtractor 13, Is obtained and the purpose is achieved.

FIG. 6 shows the space selection characteristics of the device of this embodiment. The vertical axis is the central microphone 10 with a point source separated by a distance r _q.
The output of this device, which is normalized by setting the output of 0 as "1", that is, shows the relative gain. The horizontal axis KQ is equal to (2π / λ) r _q (where λ is the wavelength), and is the sound source distance normalized with respect to the wavelength of the incident sound. Parameter K is (2π /
λ) a, the sphere radius normalized with respect to wavelength.

Clearly, the relative gain decreases as the sound source moves away beyond the radius of the sphere. The cutoff sharpness is sufficiently steep at about 35 dB per octave in this example for M = 12. This tendency increases with the value of M (11.5 dB per octave distance at M = 4, per octave distance at M = 20)
37.5dB). The lower limit of the cutoff characteristic is obtained by KQ → ∞, and its value increases as the normalized radius K increases, that is, as the frequency or radius increases. For example, a = 2.8c
If m and frequency are 4kHz, then M = 12 and K = 2,
It can be expected that the relative gain will be reduced by 40 dB at a distance of 40 cm from the sound receiving device. As such, there are significant spatial selection characteristics. Further, as is clear from the above description, this space selection characteristic is remarkable and easy to realize as the frequency is lower. This is a property opposite to the sound devices known so far, and is a convenient feature for blocking general noise having a property of high energy in a low frequency region.

FIG. 7 shows a change in relative gain when the sound source passes near the sub microphone arranged on the boundary surface.

Since the sub microphones are discretely present on the boundary surface, advantageous effects other than the spatial selectivity or the cutoff characteristic, whose principle is explained by assuming the continuous boundary surface, are generated.
That is the close-talk property described here. As shown in FIG. 7, it is apparent that the sensitivity is increased near the boundary. Therefore, this sound receiving device can operate and be used as a so-called super close-talking microphone that significantly attenuates external noise.

FIG. 8 shows the spatial characteristics when one of the sound receiving element pairs arranged on the boundary surface is removed (M = 11).

When a large number of sound receiving element pairs are arranged on the boundary surface, the sensitivity of some sound receiving element pairs can be increased or decreased to give the spatial selectivity directivity. In FIG. 8, the upper curve shows the spatial characteristics in the direction of the missing sound receiving element pair, and the lower curve shows the spatial selection characteristics in the other directions. The relative gain is increased in the direction where the sound receiving element pair is lacking, and the attenuation is large in the other directions. In this case, even in the direction of large attenuation, the spatial selectivity becomes sharper than in the case of M = 12 where the sound receiving element pairs are aligned. However, in practice, it can still have a sufficient blocking characteristic, and at the same time, can be used as a close-talking microphone for the direction in which the sub microphone is absent.

FIG. 9 is a block diagram of the sound receiving device according to the second embodiment of the present invention.

This embodiment apparatus is used when the shape of the boundary surface is not spherical. The difference from the first embodiment is that the coefficient multiplier 14-i is inserted in the signal path from the coefficient multiplier 5-i to the adder / subtractor 9-i, and the coefficient values of the coefficient multipliers 6-i and 8-i are Different,
That is, instead of the delay device 12, a delay device 12-i is inserted between the adder / subtractor 9-i and the adder 10.

When the shape of the boundary surface is not spherical, depending on the value of i,
The coefficient terms a _i , b _i , c _i and the boundary to center microphone 10
The distance R _{i to} 0 changes. Therefore, as shown in the second embodiment, the coefficient value of the coefficient multiplier of each channel and the delay value of the delay unit are adjusted to it.

Specifically, the coefficient multiplier 5-i of the i-th channel
Now all multiply by a factor of 1/2 and this output is the coefficient multiplier 1
Multiply-(1 / 4π) b _i by 4-i. Coefficient multiplier 6-i,
For 8-i, (1 / 4π) a _i / (Δr),-(1/4)
π) multiply by c _i . The delay device 12-i introduces the delay amount R _i / c.

Although only specific circuit configurations have been described in the above embodiments, the present invention can be similarly implemented with any circuit configurations as long as equivalent signal processing is performed.

〔The invention's effect〕

As described above, the sound receiving device of the present invention can easily realize spatial selectivity and close contact characteristics, and can pick up a desired signal with a high signal-to-noise ratio even in a noisy environment, It is extremely effective when applied to an input microphone for voice recognition and a microphone capable of preventing howling in an acoustic system accompanied by loud sound.

[Brief description of drawings]

FIG. 1 is a block diagram of a sound receiving device according to a first embodiment of the present invention. FIG. 2 is a diagram showing an arrangement example of a central microphone and a sub microphone group. FIG. 3 is a diagram explaining the principle of space selectivity. FIG. 4 is a diagram showing a two-dimensional arrangement of a central microphone and a sub microphone group. FIG. 5 is a diagram showing x, y, z coordinate axes and related vectors when a spherical surface having a radius a is adopted as a boundary surface. FIG. 6 is a diagram showing a space selection characteristic of the embodiment apparatus. FIG. 7 is a diagram showing a change in relative gain when a sound source passes near a sound receiving element arranged on a boundary surface. FIG. 8 is a diagram showing the spatial characteristics when one set of the sound receiving element pairs arranged on the boundary surface is removed. FIG. 9 is a block diagram of a sound receiving device according to the second embodiment of the present invention. 1-1 to 1-M, 2-1 to 2-M ... Sound receiving element, 3-1
~ 3-M ... Adder, 4-1-4-M ... Subtractor, 5-
1-5-M, 6-1-6-M, 8-1-8-M ... Coefficient multiplier, 7-1-7-M ... Differentiator, 9-1-9-M ...
… Adder / subtractor, 10 …… Adder, 11 …… Coefficient multiplier, 12, 12
-1 to 12-M ... Delay device, 13 ... Subtractor, 14-1 to 14-
M …… Coefficient multiplier, 100 …… Center microphone, 200 ……
Sub-microphone group, 300 ... Signal processing circuit.

Claims (3)

(57) [Claims] 1. A central microphone, a group of sub-microphones arranged around the central microphone, and real-time signal processing means for removing an external noise component contained in the output of the central microphone by an output of the group of sub-microphones. In the sound receiving device provided with, the sub microphone group includes sound receiving elements arranged at a plurality of points on a virtual boundary surface surrounding the central microphone, and the signal processing means includes respective sound receiving elements. Sound pressure detection means for detecting the sound pressure of external noise, first coefficient means for multiplying the sound pressure value by a coefficient, and averaging the sound pressure values multiplied by this coefficient over the entire boundary surface. Means and delay means for delaying the signal output from the averaging means by a time corresponding to the distance between the central microphone and the sound receiving element; Noise subtraction means for subtracting the output of the averaging means from the output of the central microphone. 2. The signal processing means includes a time differentiating means for obtaining the time derivative of the sound pressure from the output of the sound pressure detecting means, and a second coefficient means for multiplying the value of the time derivative by a coefficient, and the averaging means. The sound receiving device according to claim 1, wherein the sound receiving device includes means for adding a time differential value multiplied by a coefficient to a sound pressure value multiplied by the coefficient. 3. The signal processing means includes sound pressure gradient detecting means for detecting a sound pressure gradient in the normal direction of the boundary surface, and third coefficient means for multiplying the value of the sound pressure gradient by a coefficient, and an average value. The sound receiving device according to claim 2, wherein the conversion unit includes a unit that further adds the sound pressure gradient value multiplied by the coefficient.