JPH0126598B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to a multipoint sound receiving device that selectively receives acoustic signals using a plurality of microphone elements. <Prior art> When a target acoustic signal (hereinafter referred to as the target signal) is received, unnecessary acoustic signals such as noise and voice (hereinafter collectively referred to as noise) are received at the same time, resulting in an increase in the SN ratio. This may cause a decrease in performance, howling, etc. Solving this phenomenon has been an important issue in public address systems, PA (public address) systems, etc. To solve this problem, directional microphones have often been used. However, in this case, since the directivity pattern is fixed, there are many problems in practical use, such as the position of the caller being limited by the direction of the microphone. In addition, in recent years, direct array microphones have been used to achieve sharp directivity (RL Wallance et al.: U.S. Patent No. 4311874 (1982), Kido et al.: Proceedings of the Acoustical Society of Japan, Showa
October 1953, p. 181). However, since the setting theory of this method is limited to plane waves, the operation does not follow the logic when the sound waves are spherical waves, as is the case in many real cases. The drawback was that it required a large microphone array. <Summary of the invention> The object of the invention is to be able to be configured on a small scale, and to be able to prevent movement of target signal sources and noise sources.
An object of the present invention is to provide a multipoint sound receiving device that can adaptively select a target signal. According to this invention, each output of a plurality of microphone elements is delayed by the delay means, each having a different delay time, and these delayed signals are weighted and outputted by the weight addition means. A virtual target signal is electrically generated, and the virtual target signal is correlated with each of the plurality of signals obtained by delaying the outputs of the plurality of microphone elements in the same manner as described above, and a predetermined evaluation amount is minimized. Based on the correlation, the coefficients in the load addition are set so that As a result, noise is suppressed as the weighted addition output, and the target signal is extracted. <Embodiment> FIG. 1 shows an embodiment of the present invention, in which a multi-point sound receiving section 1 consisting of a plurality of microphone elements 1 ₁ to 1 _N includes a delay circuit 2 and an adding section consisting of adding circuits 3 ₁ to 3 _N. Connected to 3. The output of the delay circuit 2 is connected to a load adder 4. A virtual target signal generator 5 is provided, and its output side is connected to virtual target signal delay circuits 6 ₁ to 6 _N
The output side of the delay circuit 6 is connected to the adder 3. The output side of the adder 3 is connected to a delay circuit 7, and the output side of the delay circuit 7 is connected to a weighting factor determining section 8. The output side of the virtual target signal generator 5 is connected to the weighting factor determination section 8 through a delay element 9, and the weighting factor determination section 8 is connected to the setting input side of the weighting addition section 4. First, the basic operation of this embodiment will be explained.
In the multi-point sound receiving section 1, signals _{u 1} (t), . . . _{u N (} t) in which a target signal and noise are superimposed are received by N microphone elements 1 1 . Next, these signals are input to the delay circuit 2. Here, delay circuit 2
As shown in FIG.
1 ₁ ... 11 _N , and each delay unit has M of the input signal and the output signal of each delay element 11.
+1 signal is output. Therefore, for N input signals u ₁ (t), ..., u _N (t), L (L=
Obtain x〓 ₁ (t), ..., x〓 _L (t) of N×(M+1)). In FIG. 1, the load adding section 4 adds the output signals x〓 ₁ (t), . . . x〓 _L (t) of the delay circuit 2 with a load. This load addition _is performed using the following formula y(t)=〓 ^T・XI～(t ₎ =L 〓 i= ¹ h i・x〓 _i (t) (1) using the load coefficient h ₁ ,..., h _L however,

[Equation] T: Expressed as a transposed matrix, which obtains the output y(t) of this device. At this time, the input signal ui(t) (i=1, ...,
An output y(t) is obtained in which the noise components contained in N) are suppressed and the target signal is extracted. The following two assumptions are made when determining the load coefficient. (Assumption-) The arrival time difference of the target signal between each microphone element is known. (Assumption-) The target signal has a silent time period, and only the noise to be suppressed is received at that time. Here, the arrival time difference under the assumption is the difference in time at which the target signal (sound wave), which is generated due to the spatial arrangement of the microphone elements, arrives at each microphone element. For example, in FIG. 3, when microphone elements 1 ₁ and 1 ₂ are provided at distances d ₁ and d ₂ from the target signal source 12, the arrival time difference τ is calculated by the following formula: τ = (d ₂ - d ₁ )/C C: A quantity expressed by the speed of sound (2). Therefore, if the sound wave of the target signal can be considered to be a plane wave, the arrival direction of the target signal is known, and if it is a spherical wave, the position of the target signal source is known, then the condition of assumption - is satisfied. It is. Furthermore, typical examples of audio signals having silent time intervals that satisfy the assumption - are voice signals and music signals. Now, the determination of the weighting coefficient is carried out using the following procedure under conditions that satisfy the assumption - that is, when there is no target signal and only the noise to be suppressed is received by the N microphone elements. Ru. First, the virtual target signal generator 5 generates the virtual target signal S'(t). Then this signal
S'(t) is added to the noise received by the N microphone elements 1 ₁ to 1 _N through the virtual signal delay circuit section 6. Here, the virtual target signal delay circuit section 6 is 1
N delay circuits 6 ₁ for one input S'(t)...
_6N , N types _{of delay} amounts τ _{1 ,} . The values of the delay amounts τ ₁ , ...τ _N are assumed to be known, and the actual target signals are transmitted to each microphone element 1 ₁ ,
...1N is assumed to be a value that satisfies the relationship of the time difference when reaching _N. Therefore, S'(t-τ ₁ ),...
...S'(t-τ _N ) - each microphone element output containing only noise u ₁ (t), ..., u _N
(t) in the adder circuit 3 means that a virtual target signal is emitted from the actual target signal source, and this is received along with noise by N microphone elements 1 ₁ ... 1 _N. This corresponds to simulating the situation. Next, the signal u〓i(t) (i=1, . . . N) obtained by adding the actual noise and the virtual target signal is passed through the delay circuit 7, which has the same structure as the delay circuit 2, to the L signals XI
(t) = (x ₁ (t), ... x _L (t)) Obtain ^T. At this time,
A virtual target signal S _′ (t
−τ ₀ ) with the root mean square error E as the evaluation quantity. However, -: Time average The weighting coefficient 〓 is determined based on the least squares principle so as to minimize this. The fact that equation (3) is zero indicates that the noise from the microphone elements 1 ₁ . . . 1 _N is completely suppressed. Since equation (3) is a quadratic function of hi, by partially differentiating it with respect to hi and solving it by setting it to 0, it is possible to find the weighting coefficient 〓 that gives the minimum value of the evaluation amount E as shown below. ∂E／∂〓＝∂／∂〓｜′(− ₀ )−〓 ^T XI()
| ² = ∂/∂〓(′(− ₀ ) ² −2〓 ^T ′(− ₀ )XI()+〓 ^T 〓x〓)=−
2'(- ₀ )XI()+2〓x〓=0(4) However, In this way, using the correlation matrix 〓x of each xi(t) (i=1, . . . , L), the weight addition unit 4 performs weight addition using the weighting coefficient 〓 expressed by equation (5). For example, it is possible to obtain an output y(t) in which noise components are suppressed. Here, if it is possible to use exactly the same signal as the target signal actually received as the virtual target signal, the obtained weighting coefficient will be the optimal coefficient in the sense of minimizing the square error for the actual target signal. It's summery. However, in such a case, the optimal solution would be to output the virtual target signal itself, which is unrealistic. Furthermore, if a signal similar to the actual target signal, such as a pseudo-voice for voice, is used as the virtual target signal, it is possible to obtain coefficients close to the optimal solution in the sense of minimizing the squared error. However, if the frequency power spectrum of the actual target signal is not flat, optimization in the sense of minimizing the squared error using a virtual target signal with a similar power spectrum is mainly performed on frequency components with large power. You will be killed. As a result, the frequency sensitivity of this method to the target signal is flat in the frequency band where the power of the target signal is large;
It does not level out in frequency bands with small power. Therefore, as a first method to improve this point, a signal (for example, white noise) having a flat power spectrum within a desired frequency band is used as the virtual target signal. As a result, uniform optimization is performed for each frequency component, resulting in a more even frequency sensitivity characteristic of the target signal. A second method is to use a signal (for example, colored noise) having a power spectrum similar to the noise to be suppressed as a virtual target signal. This method aims to achieve uniform optimization by keeping the ratio of the virtual target signal and noise constant for each frequency component, and is effective when the temporal fluctuation of the power spectrum of the noise is small. The frequency sensitivity characteristics can be flattened. By selecting the virtual target signal as described above and solving equation (5), it is possible to obtain a preferable weighting coefficient, but 〓
If the correlation matrix 〓 x of (t) is not regular, the inverse matrix 〓 ⁻¹ _x does not exist, and Equation (5) cannot be calculated. Therefore, as a first method to solve this problem, it is possible to use the generalized inverse matrix 〓 ⁺ _x instead of 〓 ^-1 _x . Here, the generalized inverse matrix 〓 ⁺ _x is the orthogonal matrix that orthogonally decomposes the symmetric matrix 〓 x,
The column vector is 〓i(i=1,...L)(〓=
(〓 ₁ , ...〓 _L )), When expressed as This is expressed as Further, as a second method for the case where 〓x is not regular, a method can be considered in which uncorrelated noise is added to each XI(t) to make 〓x regular. This method is 〓 ^-1 instead of _x This corresponds to using q, and when q is sufficiently small, it corresponds to providing an approximate optimal solution. Furthermore, as a method for determining the weighting coefficient, it is also possible to use the SN ratio as the evaluation quantity and based on the principle of minimizing this. The output XI(t) of the delay circuit section 7 is calculated using the virtual target signal component 〓(t) and the noise component 〓(t), x(t)=〓(t)+〓(t) However, 〓(t )=(S ₁ (t), . . . , S _L (t)) ^T 〓(t)=(n ₁ (t), . . . , n _L (t)) ^T. The power of the signal obtained by adding the weight with the coefficient 〓 is, assuming that the noise and the virtual target signal are uncorrelated, However, it can be expressed as (〓x _s )ij=()・() (〓x _N )ij=()・(). Here is the SN after adding the load.
The ratio, When defined, 〓 that maximizes the S/N ratio can be found as follows. First of all, 〓xs, 〓x _N
Since both are symmetric matrices, there exists a regular matrix 〓 that satisfies the following relationship. Therefore, by performing variable conversion of the coefficients as shown below 〓＝〓 ^-1 〓 (12), the SN ratio becomes It can be expressed as At this time α _M αi (i=
1,...L), then the g that maximizes the SN ratio is clearly g = (g ₁ , g ₂ , ..., g _L ) ^T gi = 1 i = M 0 i≠M (14), From this, the load coefficient that maximizes the SN ratio is
can be obtained using the relationship 〓=〓〓 (15). As a first means of realizing the invention described above, there is a method of configuring the entire system as a digital system. Figure 4 shows its configuration. In Figure 4, the first
Parts corresponding to those in the figure are given the same reference numerals. Each output of the microphone elements _{1 1} . The output of the load adder 4 is converted into an analog signal by the D/A converter 14 and output. An embodiment of the delay circuits 2 and 7 in this configuration is shown in FIG. The delay unit ₁₁₁ consists of an M-stage shift register 15, and an output is derived from each shift stage. The other delay units are similarly configured. As shown in FIG. 6, the load adder 4 multiplies the outputs x〓 ₁ (n ₎ , . . . x〓 _L (n) of the delay circuit 2 by coefficients h ₁ . is,
The results are added by an adder 17. In FIG. 4, the weighting factor determination unit 8 is a processor having calculation ability, and obtains the weighting factor 〓 by directly calculating the above-mentioned equation (5) or (15). In addition to the method described above, various successive approximation methods can be used to determine the weighting coefficient. When using the successive approximation method, special attention must be paid to the convergence of the algorithm, but on the other hand, the above equation (5) or (15)
It is possible to obtain the weighting coefficient 〓 with a smaller amount of calculation and memory than directly calculating the formula. The configuration in this case is shown in FIG. In FIG. 7, parts corresponding to those in FIG. 4 are given the same reference numerals. The output of the delay circuit 7 is input to a load adder 18 which is the same as the load adder 4, and is also input to a sequential load coefficient determiner 19. The output of the weight adder 18 is added to the output of the delay element 9 by an adder 21, and the result is input to the weight factor determiner 19. The load coefficients of the load adders 4 and 18 are controlled by the determined load coefficient. When using a method known as the steepest descent method or LMS algorithm as a successive approximation method that uses the root mean square error as an evaluation quantity, the weighting coefficient 〓(n) at each time (where n is a parameter representing time) is Shape load coefficient determining section 1
9, the following formula〓(n)=〓 _(o-1) +k _s XI _(o-1)・l _(o-1) (16) However, l _(o-1) =S′ _(o- 〓 _{0 )} −〓 Calculated by ^T _(o-1)・XI _(o-1) . As a second means for realizing the present invention, there is a method in which the entire system is constructed from an analog system. An example of the configuration in this method is shown in FIG. 1, and an example of the element configuration of each part is as follows. First, the configurations of the delay circuits 2 and 7 are shown in FIG. 2, and each delay element is realized by an analog delay element such as a BBD or a CCD. The load adder 4 has the same configuration as the digital system shown in FIG. 6, except that the multiplier 16 in FIG. 6 is replaced with an analog multiplier and the adder 17 is replaced with an analog adder. Also, when determining the load coefficient in an analog system,
It is difficult to perform operations such as calculating the inverse matrix. Therefore, the load coefficients are determined by a successive approximation method, and the load coefficient determining section 8 in FIG. 1 is realized, for example, by the circuit configuration shown in FIG. 8. That is, the output XI(t) of the delay circuit 7 is supplied to L analog multipliers 23 and L analog correlators 24. The outputs of the L analog multipliers 23 are added by an analog adder 25, and the outputs thereof and the output of the delay element 9 are added by an adder 26 and supplied to each correlator 24. Each correlator 24
The outputs of L are passed through analog multipliers 27 respectively.
is supplied to two integrators 28. These integrators 28
A weighting coefficient is then obtained and supplied to the multiplier 23. This circuit satisfies the following equation dhi(t)/dt= _s ()·(), which is a successive approximation equation for the weighting coefficient in a continuous system when the root mean square error is used as the evaluation quantity. Moreover, as a third means to realize this invention,
A configuration combining digital circuits and analog circuits is conceivable. As an example of this method, for example, the first
A method may be considered in which the portion consisting of the delay circuit 2 and the load adder 4 in the figure is configured with an analog circuit, and the other portions related to determining the weight coefficient are configured with a digital system. Now, as described above, in order for the device of the present invention to perform the desired operation, the following two assumptions are made: The arrival time difference of the target signal between each microphone element is known. (Assumption-) There is a silent section in the target signal, and only the noise to be suppressed is received at that time. need to be satisfied. However, if the level of the noise is lower than the level of the target signal and it can be considered that only the target signal is being received, or if there is a silent period in the noise, (Assumption -) is satisfied by adding means for detecting the arrival time difference of the target signal, an example of which will be described below. First, each microphone element output signal u ₁ (t),...
Compute the cross-correlation function between u _N (t). and,
Cross-correlation function φsij(τ) between ui(t) and uj(t)
If the value τ _M ij of the value τ that takes the maximum value is found, it can be regarded as the arrival time difference of the target signal between the microphone element i and the microphone element j. The detection of this arrival time difference τ _M ij is performed using the digitized signal ui
(n), uj(n) (n=..., -1, 0, 1,...)
When determining τ _M ij with high accuracy, the sampling frequency of the signal must be made high enough, or
Alternatively, it is necessary to apply an interpolation method to the cross-correlation function obtained from a low sampling frequency and then find the arrival time difference. This result (assumption-)
is satisfied. An embodiment of the present invention based on the above concept is shown in FIG. In FIG. 9, a target signal arrival time difference detection section 29 is provided in the configuration shown in FIG. 4, and the output of the A/D converter 13 is branched into this. Each delay amount of the delay circuit section 6 is set based on the detection result of the detection section 29. As shown in FIG. 10, the target signal arrival time difference detection section 29 inputs each output of the A/D converter 13 to a cross-correlation function calculation section 31 and calculates the difference between the microphone outputs ui(n) and uj(n). The cross-correlation function φsij (τ) is calculated, and the maximum value detection unit 32 detects the value τ _M ij of τ at which φsij (τ) takes the maximum value. Next, in the virtual target signal delay amount determination unit 34, τ _M ij (i=1, 2,...N, j=
Using a value sufficiently larger than 1, 2,...N),
The virtual target signal delay amount 〓=(τ ₁ , . . . , τ _N ) ^T is determined by the following formula τi=〓−τ _M ij (j is an arbitrary value of 1jN). Next, a case will be described in which the operation of the apparatus of the present invention is completely automated. One way to make the operation of this device automatic is to add means for determining whether substantially only noise is being received or whether substantially only a target signal is being received. As a result of the determination, in a state in which only the target signal is substantially being received, the arrival time difference of the target signal is known by, for example, a method using the above-mentioned cross-correlation function, and the assumption - is satisfied. Also,
It is clear that the assumption - is satisfied at a time when it is determined that substantially only noise is being received. FIGS. 11 and 12 show an embodiment of the present invention in which it is determined whether only noise is being received or if only a target signal is being received. FIG. 11 shows an embodiment of the present invention in a conference telephone system, and the output of the multipoint sound receiving section 35 is inputted to the multipoint sound receiving system section 36 according to the present invention described above. The speaker 38 is driven by the signal on the reception signal path 37,
The output of the multipoint sound receiving system section 36 is transmitted through the transmitting signal path 39.
It is output from FIG. 12 shows an embodiment of the present invention applied to a loudspeaker telephone, in which the multipoint sound receiving section 35 is outputted to the transmission signal path 39 through the multipoint sound receiving system section 36 according to the present invention. A speaker 38 is driven by a signal from the reception signal path 37. A dial 41 is provided. In these two examples, the sound emitted from the speaker 38 is received by the multi-point sound receiving section 35, and the transmission signal path 3 receives the sound emitted from the speaker 38.
9, various problems will occur, such as howling and deterioration of call quality. Therefore, in these two examples, the noise is the voice generated from the speaker 38, and the target signal is the voice uttered by the speaker. Therefore, since there is always a silent section in speech, there are states in which only noise exists and states in which only the target signal exists. By monitoring the input level (received signal level) to the speaker 38, for example, even if the received signal level is 0, the microphone (multipoint sound receiving section 3
5) When the output level increases, it means that only the target signal is present, the received signal level is above a certain level, and the microphone output level is equal to the received signal level, the electroacoustic conversion amount of the speaker 38, and the speaker 38. - If the amount of acoustic coupling between the microphones is less than a value determined by the microphone sensitivity, it is possible to determine that only noise exists. As a result, this device can perform desired operations, suppress noise, and selectively receive target signals. Further, if the position of the target signal sound source or the position of the noise source can be considered to be substantially fixed, one or two of the above assumptions can be satisfied by prior learning. For example, in the integrated loudspeaker telephone shown in FIG. 12, the noise is the received voice emitted from the speaker 38, the vibration of the body that accompanies it, and the vibration that the sound pressure inside the body gives to the microphone element. The positional relationship of this noise source with respect to each microphone element is substantially fixed. Therefore, a test signal (for example, white noise, colored noise, etc.) is generated from the speaker 38 in advance and received.
If it is stored in this system section 36 (assuming -
) will be satisfied. Therefore, during a call, if only the difference in arrival time of the target signal is detected when the received signal level is substantially zero,
This device is capable of desired operation. Furthermore, the following means can be considered as an embodiment in which the load coefficient is semi-fixed. First, assuming that the position of the noise source is substantially fixed, only noise is generated and stored in the system in advance. Next, the expected position of the target signal source is determined, a speaker is installed at that position, an acoustic signal is generated, the sound is received at multiple points, and a weighting coefficient is determined based on this and the previous noise. Next, a speaker is installed at another position of the expected target signal, and the load coefficient is determined in the same manner. In this way, K expected target signal source positions
p ₁ , p ₂ , ...K weighting coefficients 〓 ₁ , corresponding to p _K
〓 ₂ ,...〓 _K is determined in advance. If the system section 36 operates using the weighting coefficient 〓i, it is guaranteed that when the target signal source is at position pi, noise will be suppressed and extraction of the target signal will be maximized. Therefore, if the number of expected positions of the target signal source is made sufficiently large, the position of the target signal will be approximately in the vicinity of one of the positions pj. Therefore, by creating K outputs y ₁ =〓 ₁ ^T XI,...y _K =〓 ^T _K Desired operation becomes possible by outputting a component that is expected to contain a large amount of components. This method involves preparing K directional microphones that are less sensitive to noise but more sensitive to the target signal at position pi, and then selectively using these K microphones. corresponds to Also, as the output of this method, if we set y= _K 〓 ⁱ⁼¹ yi , then the output is low sensitivity to noise and high sensitivity to any position within a certain range of the expected target signal. A sound receiving system is configured. Next, we will discuss the simulation results and experimental results conducted to confirm the effectiveness of this method. First, the sound field in the simulation was assumed to be a two-dimensional sound field, and the noise and target signals were each plane waves. As shown in Fig. 13, the directions of arrival of the noise and the target signal are two directions (θ
= 60°, 100°), the target signal is in one direction (θ = -60°)
It was assumed that it would arrive from In addition, the band of the signal to be considered is 300 to 3000 Hz, and the two noises are independent noises that have a white spectrum in that band. As shown in Fig. 14A, B, and C, the components of this device are five microphone elements, three microphone elements arranged at equal angles on a circle with a radius of 8.5 cm on a plane, and microphone elements with a radius of 8.5 cm. We considered two pieces arranged at intervals. The configuration of the device is based on the configuration of a digital system as shown in FIG.
The sampling frequency of the A converter 14 was 8KHz. The delay time of each delay element in the delay circuit 2 was a sampling period (125 μsec), and the number of delay elements for each microphone input was eight. Further, the virtual target signal generator 5 generated a digital virtual target signal having a white spectrum in a band of 300 to 3000 Hz. Assuming that the above-mentioned (assumption-) and (assumption-) are satisfied, the operation of this device with various microphone arrangements shown in FIG. 14 is simulated to calculate the load coefficient
I did the calculations. The directional sensitivity pattern of this multipoint sound receiving device when using the load coefficient 〓 obtained by arrangement A is
It is shown in Figure 5. As shown in Fig. 13, although the direction of maximum directional sensitivity and the direction of arrival of the target signal are not necessarily the same, the present invention automatically creates a directional sensitivity pattern in which the sensitivity of the direction of arrival of the two noises is sufficiently reduced. It can be seen that it is formed. Further, FIG. 16 shows the frequency sensitivity characteristics of this device for each direction of arrival when using the weighting coefficients obtained by the arrangement of each microphone element. 16th
Figure A shows the frequency sensitivity characteristics when arrangement A is used, and it can be seen that the noise component is suppressed by 20 dB or more compared to the target signal component, confirming the effectiveness of the present invention. Next, when arrangement B was used, when the directions of arrival of noise were set to two directions as shown in FIG. 13, almost no noise component suppression effect was obtained. This is because this device performs noise suppression by utilizing the difference in the direction of arrival of the noise and the target signal, and the direction is symmetrical to the line connecting the two microphone elements (θ
This shows that when the noise and the target signal arrive from angles of θ=60° and θ=−60°, the directions of arrival cannot be distinguished. If the microphone elements are arranged in a straight line, this problem cannot be solved even if the number of microphone elements is increased as long as the arrangement is one-dimensional. Therefore, it is desirable to arrange the microphone elements in a multi-dimensional manner, unless this poses a practical problem, in order to improve separation and detection of the direction of arrival of the noise and the target signal. However, if the arrival directions of the noise and the target signal are not symmetrical with respect to the line connecting the two microphone elements, a sufficient noise suppression effect can be obtained even with two microphone elements. FIG. 11B shows the result when arrangement B is used and the direction of arrival of noise is set to only one direction (θ=110°), and this can be confirmed. Next, the results when the number of microphones is three as shown in arrangement C are shown in FIG. 16C. 1st
As can be seen by comparing FIG. 6C with FIG. 16A (in the case of five microphone elements), in the case of FIG. 16C, the frequency sensitivity characteristics with respect to the target signal are not flat. As these two results show, in general, increasing the number of microphone elements allows for more detailed control and improves the performance of the device. However, in practice, it is necessary to determine the number of microphones corresponding to the size of the arrangement, taking into consideration the degree of performance improvement, possibility of implementation, cost, etc. Next, we will discuss the experimental results. The experiment was conducted in a room with a reverberation time of 0.6 seconds. The experimental conditions are the first
7, a multi-point sound receiving unit 1 having four microphone elements is installed on a desk 44, and a speaker 4
5 and speaker 46 were placed at a distance of 60 cm and 50 cm from the center of the four microphone elements, respectively. The four microphone elements were arranged on the circumference of a circle with a radius of 8.5 cm, three at equal angles, and one at the center of the circle. From the speaker 45 as a target signal
A voice with a band limit of 300 to 3000 Hz was emitted, and a white noise with a band limit of 300 to 3000 Hz was emitted as noise from the speaker 46. In determining the weighting coefficient, the outputs of the speakers 45 and 46 were temporarily stopped, and only noise and only the target signal were received, respectively. The resulting load coefficient〓
Speaker 45 and speaker 4 of this device using
The sensitivity to 6 is shown in FIG. From this figure, the sensitivity to noise (speaker 46) is 20 dB lower than the sensitivity to the target signal (speaker 45) (curve 48), as shown by curve 47, and 7 to 7 dB even in higher frequencies.
It was found that the noise was reduced by 8dB, suppressing the noise,
It was confirmed that the effect of this invention in extracting the target signal was exhibited. However, as can be seen from FIGS. 16 and 18, the frequency sensitivity characteristics of this device to the target signal component are not necessarily flat. In order to improve this point, FIG. 19 shows an embodiment of a method for correcting the frequency sensitivity characteristic by providing an equalizer at the final stage of this device. The difference in FIG. 19 from FIG. 1 is that an equalizer 49 is provided, and the frequency sensitivity characteristics for the target signal component are expressed as, for example, virtual target signal delay amount = (τ ₁ , τ ₂ , ...τ _N )
It is calculated from the specified load coefficient 〓, and its inverse characteristic is assumed. In the above, the delay circuits 2 and 7 may be used in common, and one may be used by switching. It is better that the number of delay elements 11 in the delay circuits 2 and 7 is larger, and the delay time of each one is shorter. In any case, the delay amount of the series delay elements as a whole is made larger than the sound wave propagation time between the farthest elements among the microphone elements 1 ₁ . . . 1 _N. As explained above, the device of this invention obtains an output by adding the weight after passing the signals received at multiple points through a delay circuit. In this method, only the arrival time difference between the signals and the noise received when the target signal is in a silent section are used as information. Therefore, even if the arrival direction of the noise or the properties of the target signal are unknown, or even if the target signal source or noise source moves, the weighting coefficients can be adaptively changed to suppress the noise component and extract the target signal. It is possible to do this. Another advantage is that there is no need to assume the plane wave nature of sound waves, which is required in conventional array microphone design theory, and sufficient noise suppression effects can be obtained as long as the actual microphone arrangement is a dozen or more centimeters in size. It has been experimentally confirmed that there is.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing one embodiment of the device of the present invention, FIG. 2 is a diagram showing an example of the delay circuit 2, and FIG.
The figure is an explanatory diagram of the arrival time difference of the target signal, Figure 4 is a block diagram showing an embodiment of this invention device realized in a digital system, and Figure 5 is a delay circuit 2 when this invention device is realized in a digital system. FIG. 6 is a diagram showing an embodiment of the load adding section 4 when the device of the present invention is implemented in a digital system.
Fig. 7 is a block diagram showing an example of a digital system implementing the method of determining the load coefficient in the device of this invention by the successive approximation method, and Fig. 8 shows the method for determining the load coefficient in the device of the invention by the successive approximation method. FIG. 9 is a block diagram showing an embodiment of the weighting coefficient determining section 8 for realizing the method of determining the signal in an analog system. FIG. 10 is a block diagram showing an example of the means 29 for detecting the arrival time difference of the target signal, and FIG. 11 is a schematic diagram showing an embodiment of the present invention in a conference telephone system.
Fig. 2 is a schematic diagram showing an embodiment of the present invention in a public address telephone; Fig. 13 is a diagram showing the direction of arrival of noise and a target signal in a simulation experiment; and Fig. 14.
The figure is a diagram of the microphone arrangement in the simulation experiment, Figure 15 is a diagram showing the directional sensitivity pattern of this device as a result of the simulation, Figure 16 is a diagram showing the frequency sensitivity characteristic of the simulation result, and Figure 17 is a diagram showing the experimental conditions. FIG. 18 is a frequency sensitivity characteristic diagram of experimental results, and FIG. 19 is a block diagram showing an example in which an equalizer is added to the device of the present invention. 1: Multi-point sound receiving section consisting of multiple microphones,
2, 7: Delay circuit, 4, 18: Load addition section, 5:
Virtual target signal generator, 6: Virtual target signal delay circuit, 8: Loading coefficient determination unit, 9: Delay element, 13:
A/D converter, 14: D/A converter, 19: Sequential weighting coefficient determination section, 29: Target signal arrival time difference detection section.

Claims (1)

[Scope of Claims] 1. A plurality of microphone elements that receive acoustic signals; a first delay means that outputs a plurality of signals by delaying the output signals of these microphone elements by mutually different times; and a plurality of outputs of the plurality of microphone elements. means for extracting an acoustic signal that needs to be extracted by suppressing unnecessary acoustic signals from among the acoustic signals received by the plurality of microphone elements by adding the weights of the signals; a second delay means for outputting a plurality of signals obtained by adding an appropriate time delay to the virtual target signal generated from the generation means; a plurality of delayed signals output from the second delay means; means for adding signals received by a plurality of microphone elements, and outputting a plurality of signals in which a delay similar to that of the first delay means for the output signals of the plurality of microphone elements is added to the output signal of the addition means. find the correlation between the third delay means, the output signal of the delay means, and the virtual target signal, and set the coefficient of the weight addition means based on the correlation so that a predetermined evaluation amount is minimized. A multi-point sound receiving device comprising means for.