JP3451022B2 - Method and apparatus for improving clarity of loud sound - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to control of a loud sound generated in a sound field when performing loud sound using electro-acoustic equipment, and more particularly to a method and apparatus for improving the clarity of the loud sound in the sound field.

[0002]

Conventionally, as a method for improving the intelligibility of the loudspeaker is the amplified sound with the electro-acoustic equipment, there has been known one described in the literature (IEICE EA94 over 38). First, with reference to FIG. 4, a conventional method for improving the intelligibility of a loud sound in a sound field will be described. FIG. 4 is a diagram showing the configuration of a conventional method and apparatus for improving the intelligibility of a loud sound in a sound field. In FIG. 4, 201 is an input and 202 is a FI
R filter, 203 is amplifier, 207 is sound field, 204 is speaker, 205 is transfer function, 206 is control target point, 2
Reference numeral 08 is a transfer function setter, 209 is a target characteristic, 210 is a divergence prevention coefficient, 211 is a coefficient calculator, and 212 is a coefficient.

Next, referring to FIG. 4, the structure of the conventional method and apparatus for improving the intelligibility of a loud sound will be described. Sound field 20
In FIG. 7, N speakers 204 are placed and M control target points 206 are set. The transfer function 205 is the speaker 2
04 is a transfer function between the control target points (1 to M) 206 and is obtained by a measuring or predicting means. The coefficient calculator 211 uses the target characteristic 20 for which a waveform with little reverberation is set.
9. Based on the transfer function from the transfer function setter 208 that sets the transfer function 205, the divergence prevention coefficient 210, and the like,
Coefficient 2 for the FIR filter 202 by the least squares method
Calculate twelve. The coefficient 212 is used as the FIR filter 20.
By setting to 2 and connecting to the speaker 204 via the amplifier 203, the audio signal of the input 201 can be clearly transmitted to the control target point 106. In this way, even in the conventional method for improving the intelligibility of the loud sound, the intelligibility can be improved for the time being.

[0004]

However, the above-mentioned conventional method for improving the intelligibility of the loud sound has a problem that the amount of calculation of the control filter required becomes large when the reverberation time is long. In addition, if the number of control target points is increased, the number of transfer functions that need to be measured also increases, and if the sound field changes with time due to temperature or the like, intelligibility may decrease. There was a problem.

The present invention has been made to solve the above-mentioned conventional problems, and reduces the calculation amount of the control filter,
The purpose is to reduce the number of transfer functions that need to be measured, and to always obtain a clear loud sound even when the sound field changes with time.

[0006]

In order to solve the above problems, the present invention divides a sound field control filter into a low-pass control filter and a high-pass control filter, and sets the number of low-pass control filters. Less than the number of high-frequency control filters, the low-frequency components divided from the audio input signal are processed by the low-frequency control filter, and the low-frequency signal passed through the low-frequency control filter is output from the speaker. The transfer function of the sound field is calculated to calculate the coefficients for the target characteristics of the low frequency control filter and the high frequency control filter.

According to the present invention, the sound field control filter is divided into a low frequency control filter and a high frequency control filter, and only the low frequency control filter is used to control the low frequency control filter and the high frequency control filter. By calculating the coefficient of the above, it is possible to obtain a method and a device for improving the intelligibility of a loud sound that can reduce the amount of calculation for the entire control filter and improve the intelligibility of a wide area.

Further, according to the present invention, when there are a plurality of control target points, the unknown control target point and the control speaker are connected based on the transfer function between the known control target point and the control speaker. Is provided with a means for estimating the transfer function of. INDUSTRIAL APPLICABILITY The present invention provides a method and apparatus for improving the intelligibility of a loud sound, which can reduce the amount of calculation for the entire filter and can reduce the number of transfer functions required for calculating the filter coefficient.

Further, according to the present invention, in addition to a filter whose coefficient is time invariant, a filter whose coefficient is time varying is provided by adaptive signal processing. Thereby, the present invention
(EN) A method and a device for improving the intelligibility of a loud sound that can always obtain an optimum filter coefficient with respect to a change in a sound field.

Expansion in the invention of claim 1 of the [0010] present invention
A method of improving the clarity of a voice sound is to add a voice input signal to multiple frequencies.
It is divided into bands and the divided audio input signal is reduced
For each control filter through at least one control filter
When outputting from the provided loud sound presentation speaker,
By performing the process of 3 for each frequency band,
It is a method of improving the clarity of loud sound that determines the
In the first step, each loud sound presentation speaker and control target point
The transfer function between the
The process is based on the impulse response based on the band-limited transfer function.
The direct sound part of the answer is the target characteristic, and the third step is the bandwidth control.
Based on the limited transfer function and target characteristics,
It is characterized by comprising each step of determining the coefficient.

Expansion in the invention of claim 2 of the [0011] present invention
The method of improving the intelligibility of voice sound is to reduce the voice input signal to low and high frequencies.
At least each of the divided audio input signals
One low-pass control filter and one high-pass control filter
Loudspeaker presentation speaker provided for each control filter
Each output is controlled by performing steps 1-5.
A method for improving the intelligibility of a loud sound that determines the filter coefficient.
Then, the first step is to make a loud sound presentation speaker that outputs low frequencies.
The transfer function between the power and the controlled point is band-limited in the low range,
The second step is the impal based on the band-limited transfer function.
The direct sound part of the response is the first target characteristic, and the third step
Is based on the transfer function band-limited in the low range and the first target characteristic.
Then, the coefficient of the low-pass control filter is determined based on the fourth step.
Is a loudspeaker presenting speaker that outputs high frequencies and control points
Between the first target characteristic and the transfer function between
From the impulse response difference used in the second step, the second target
The fifth step is to obtain the characteristics, and the transfer function is band-limited at high frequencies.
Of the high-pass control filter based on the number and the second target characteristic
It is characterized by comprising each step of determining.

[0012] An extension of the invention according to claim 3 of the present invention.
The device for improving the intelligibility of the voice sound has a low-pass filter and a high-pass filter.
Overfilter and at least one low-pass control filter
And at least one high-frequency control filter.
A device for improving the clarity of a voice sound, wherein the low-pass filter is
Input audio signal is processed and output to each low-pass control filter
The high-pass filter processes the input audio signal and
Output to the low frequency control filter, and each low frequency control filter
The filter processing is performed by the coefficient determined in the following 1-3 steps.
Output to the loudspeaker presentation speaker, and the high-frequency control filters
Filter is determined by the coefficient determined in the following 1-3 steps.
It is filtered and output to the loudspeaker presentation speaker.
Is the transmission between each loudspeaker presenting speaker and the control target point.
The function is band-limited by its frequency band, and the second step is
Direct part of impulse response based on band-limited transfer function
Is the target characteristic, and the third step is the bandwidth-limited transfer function.
Determine the coefficient of each control filter based on the number and target characteristics
It is characterized by having each function.

[0013] An expansion of the invention according to claim 4 of the present invention.
The device for improving the intelligibility of the voice sound has a low-pass filter and a high-pass filter.
Overfilter and at least one low-pass control filter
And at least one high-frequency control filter.
A device for improving the clarity of a voice sound, wherein the low-pass filter is
Input audio signal is processed and output to each low-pass control filter
The high-pass filter processes the input audio signal and
Output to the low frequency control filter, and each low frequency control filter
The filter processing is performed by the coefficient determined in the following 1-3 steps.
Output to the loudspeaker presentation speaker, and the high-frequency control filters
Filter is determined by the coefficient determined in the following 1-5 step.
It is filtered and output to the loudspeaker presentation speaker.
Is the loudspeaker presenting speaker that outputs low frequencies and the control target point.
The transfer function between and is band-limited in the low range, and the second step is
Impulse response direct sound based on band-limited transfer function
The part is the first target characteristic, and the third step is the low frequency band.
Low frequency control based on the limited transfer function and the first target characteristic
The filter coefficient is determined, and the fourth step outputs the high frequency band.
The transfer function between the loudspeaker presenting speaker and the control point
Band-limited in the high range and used in the first target characteristic and the second step
The second target characteristic is obtained from the difference in impulse response
The process consists of a transfer function that is band-limited at high frequencies and a second target characteristic.
Functions that determine the coefficient of the high-pass control filter based on the
It is characterized by having.

Expansion in the invention of claim 5 of the [0014] present invention
Intelligibility enhancement apparatus vocal when the varying of the filter, when a plurality of
An intelligibility improving device for a loud sound including an invariant filter.
Therefore, the time-varying filter filters the input audio signal by a coefficient.
A microphone that filters and sets the sound field
The coefficient is constantly updated with the sound collection signal from
The signal from the time-varying filter is converted into
Presentation of loud sound by filtering with a coefficient determined by
The first step is to output to each speaker
The transfer function between the power source and the controlled point is band-limited, and the second
Process is impulse response based on band-limited transfer function
The target characteristic is the direct sound part of the
Time-invariant filter based on the transfer function and target characteristics
It is characterized by having each function of determining the coefficient of.

[0015]

[0016]

[0017]

[0018]

[0019]

[0020]

[0021]

[0022]

[0023]

[0024]

[0025]

Hereinafter, based on the attached drawings and FIGS. 1 to 3,
Embodiments of the present invention will be described in detail. (Embodiment 1) First, with reference to FIG. 1, a method and apparatus for improving the clarity of a loud sound according to Embodiment 1 of the present invention will be described. FIG. 1 is a diagram showing a configuration of a method and apparatus for improving the intelligibility of a loud sound according to Embodiment 1 of the present invention.
In FIG. 1, 101 is an input, 102 is a switch, and 10
3 is a low-pass filter, 104 is a high-pass filter, 1
Reference numeral 05 is a delay device, 106 is a low frequency control filter, 107 is a high frequency control filter, 108 is an amplifier, 109 is a speaker, 110 is a transfer function, 111 is a transfer function, 112 is a control target point, 113 is a sound field, 114 is a low-pass filter,
115 is a target characteristic setting device, 116 is a divergence prevention coefficient, 11
Reference numeral 7 is a coefficient calculator, 118 is a coefficient, 119 is a high-pass filter, 120 is a target characteristic setting device, 121 is a divergence prevention coefficient, 122 is a coefficient calculator, and 123 is a coefficient.

Next, with reference to FIG. 1, the structure of the method and apparatus for improving the clarity of the loud sound according to the first embodiment will be described in detail. The input 101 is connected to a switch 102, a low pass filter 103, and a delay device 105. Switch 102
Is connected to a high-pass control filter 107 through a high-pass filter 104, the output of which is combined with the outputs of the delay device 105 and the low-pass control filter 106, and the amplifier 108
And is connected to the speaker 109 in the sound field 113. In the sound field 113, the transfer function 110 between the N speakers 109 and the M controlled points is defined, and the transfer function 111 between the 1 and the M controlled points of the speaker 109 is defined. Is defined. Low pass filter 11
4, the target characteristic setting unit 115 and the divergence prevention coefficient 116 are connected to the coefficient calculator 117, and the output thereof becomes the coefficient 118. Further, the high-pass filter 119, the target characteristic setting device 120 and the divergence prevention coefficient 121 are connected to the coefficient calculator 122, and the output thereof becomes the coefficient 123.

Next, with reference to FIG. 1, an operation procedure of the method and apparatus for improving the intelligibility of the loud sound in the first embodiment will be described. In the method of improving the intelligibility of the loud sound in the first embodiment, first, the switch 102 is set to the A side to use the low-frequency input signal to output the loud sound from the loud sound presentation speaker through the low-frequency control filter. , The coefficients of the low-pass and high-pass control filters are calculated so that the best sound can be obtained at the control target point, and are set in the respective control filters. After setting the coefficient, the switch 102 is tilted to the B side to use the low-pass and high-pass control filters and all the speakers.
Hereinafter, a method of calculating the coefficient of the low-pass and high-pass control filters that can obtain the best clarity of the loud sound according to the first embodiment will be described.

First, in the sound field 113, the speaker 10
N × M transfer functions 110 between 9 and M control target points 112 are obtained by prediction means such as measurement or simulation. In this example, the speaker 109 1 is a loud sound presentation speaker. Next, the target characteristic setting device 115
Is M between 1 and M control target points 112 of the loud sound presentation speaker 109 that has passed through the low-pass filter 114.
The target characteristic is calculated using the individual transfer functions 111. The cut-off frequency fc of the low-pass filter 114 is within the frequency bandwidth of the audio signal, may be about 4 kHz, or may be determined by (Equation 1). fc = c / L (Formula 1)

However, c is the speed of sound, and L is the size of the controlled area. The target characteristic setting device 115 first calculates an initial response. The initial response is calculated by representing each transfer function in the time domain and cutting out only an initial part including the direct sound. The direct sound detection method is based on time t0 when the square amplitude of each transfer function in the time domain takes the maximum value.
A method of setting it in the range of 0-t1 to t0 + t1 [ms] is also possible. t1 is 5 [ms] or less, and t2 is 10 [m]
s] is desirable, generally t1 is 1 [ms] and t2 is 5
[Ms] will do. From the M initial responses set as described above, M transfer functions 111 between 1 of the corresponding loudspeaker presentation speaker 109 and M control target points 112 are subtracted, and then a delay is added. The target characteristic is the one. The length of the delay time must be equal to or larger than the maximum value of the length of the duration of the N × M transfer functions 110 expressed in the time domain.

Next, the divergence prevention coefficient 116 is set.
The larger the divergence prevention coefficient 116, the smaller the required coefficient length, but the control result at the controlled point departs from the target characteristic. Generally, the power spectrum of the transfer function 110 is calculated, the maximum value and the minimum value on the frequency axis are calculated,
If the ratio of the two is within 40 dB, the divergence prevention coefficient 116 may be 0. However, if the ratio of the both exceeds 40 dB, a value obtained by subtracting 40 dB from the maximum value on the frequency axis may be used as the divergence prevention coefficient.

The coefficient calculator 117 obtains the coefficient 118 set in the low-pass control filter 106 by the method of least squares. The determinant can be solved in time series, but it is better to solve it in the frequency domain because the memory required for calculation increases as the response length increases. The solution in the frequency domain is described below. Target characteristic setter 11 for each frequency
5 is D (M, 1), the result of applying the low-pass filter 114 to the transfer function 110 (low-pass component of the transfer function 110) is A (M, 1), and the coefficient 118
Is H (1,1), and the divergence prevention coefficient 116 is β. However, D, A, and H are complex matrices, where the subscripts (X,
Y) means X rows and Y columns. At this time, the least squares solution H is obtained by solving (Equation 2). H = (A'D) / [A'A + βI] (Equation 2)

However, A'is a Hermitian matrix of A, and I is a 1-by-1 complex identity matrix. (Equation 2) is a one-dimensional simultaneous linear equation, which can be solved by a known numerical solution method, and in particular, the QL method or LU decomposition can be used. First, the cutoff frequency f of the low pass filter 114
After obtaining H for each frequency equal to or lower than c, one coefficient 118 is obtained by converting into H and then into the time domain. One coefficient 118 thus obtained is set in the low-pass control filter 106. In addition, the delay unit 10 of the loud sound system
In 5, the delay time previously added to the target characteristic is set.

The difference between the response and the target characteristic at each control target point used in the low-pass control filter 106 calculated as described above is obtained and used as the target characteristic 120 of the high-pass control filter. As a method for obtaining the difference between the impulse response and the target characteristic at each control target point, the low-pass control filter 106 is used.
It is also possible to use a method based on computer simulation using the transfer function 110 and the transfer function 110, or a method based on measurement. In the case of measurement, the switch 102 is set to the A side and the impulse response at each control target point 112 is measured. At this time, the cutoff frequency of the low pass filter 103 is set to be the same as the cutoff frequency fc of the low pass filter 114. Next, in the target characteristic setting device 120, each control target point 1 when the low-frequency control filter 106 is used
The difference between the response in 12 and the target characteristic set in the target characteristic setting unit 115 is calculated to be the target characteristic for high frequency control.

Next, the divergence prevention coefficient 121 is calculated by the same method as the divergence prevention coefficient 116, and the transfer function 110 is input to the coefficient calculator 122 via the high pass filter 119. The cutoff frequency fc of the high pass filter 119
Is the same as the cutoff frequency fc of the low pass filter 114. The coefficient calculator 122 uses a matrix to obtain a coefficient that is close to the target characteristic by the least square method. The determinant can be solved in time series, but it is better to solve it in the frequency domain because the memory required for calculation increases as the response length increases. The solution in the frequency domain is described below.
The target characteristic set by the target characteristic setting unit 120 for each frequency is D (M, 1), and the transfer function 110 is A (M,
N), the coefficient 123 is H (N, 1), the divergence prevention coefficient 121
Be β. However, D, A, and H are complex matrices, and the subscripts (X, Y) mean X rows and Y columns. At this time, the least squares solution H is obtained by solving (Equation 2). (Equation 2) is an N-ary linear simultaneous equation and can be solved by a known numerical solution method. In particular, the QL method or LU decomposition can be used.

Cutoff frequency fc of high-pass filter 119
After H is obtained for each of the above frequencies, N coefficients 123 are obtained by converting the H into the time domain. The N coefficients 123 thus obtained are set in the N filters 107, respectively. The cutoff frequency of the high pass filter 104 is the cutoff frequency fc of the high pass filter 119.
Set the same as. In the above description, the coefficients 118 and 123 of the low-frequency control filter 106 and the high-frequency control filter 107 are obtained by using only the low-frequency band of the input signal. The loud sound presentation speaker 10
1 of 9 can be used like other speakers in the normal use state after the coefficient calculation.

Next, by switching the switch 102 to the B side, the audio signal to the input 101 is delayed by the delay device 105,
Low-pass control filter 106 and high-pass control filter 1
After being amplified by the amplifier 108 via 07 and output from the speaker 109, a clear audio signal can be heard at the control target point 112. Further, since the bandwidth of the high-frequency control filter 107 is limited to the cutoff frequency fc or more, the amount of calculation of the entire filter is reduced as compared with the conventional case. In this way, it is possible to improve the intelligibility in the loud sound system while reducing the calculation amount of the control filter.

(Embodiment 2) Next, referring to FIG.
A method and apparatus for improving the intelligibility of a loud sound according to Embodiment 2 of the present invention will be described. FIG. 2 is a diagram showing a configuration of a method and apparatus for improving the intelligibility of a loud sound according to the second embodiment of the present invention. In FIG. 2, 301 is an input, 302 is a control filter, 303 is an amplifier, 304 is a speaker, 3
05 is a transfer function, 306 is a control target point, 307 is a sound field,
308 is a transfer function estimator, 309 is a band limiter, and 310
Is a target characteristic setter, 311 is a divergence prevention coefficient (β), 31
2 is a coefficient calculator, and 313 is a coefficient. Note that the speaker 304 is used as a control speaker during coefficient calculation.

Next, with reference to FIG. 2, the structure of the method and device for improving the clarity of the loud sound according to the second embodiment will be described in detail. In FIG. 2, an input 301 is connected to a control filter 302 for controlling a sound field, and is connected to a speaker 304 provided in a sound field 307 via an amplifier 303.
The transfer function 305 is a speaker 304 and three control points 3
And the transfer function between 06 and 06. The transfer function estimator 308 estimates an as yet unmeasured transfer function between the speaker 304 and the three controlled points 306. Band limiter 30
9 and output of target characteristic setting device 310 and divergence prevention coefficient 31
1 is an input to the coefficient calculator 312, and the output of the coefficient calculator 312 is a coefficient 313.

Next, with reference to FIG. 2, an operation procedure of the method and apparatus for improving the intelligibility of the loud sound in the second embodiment will be described. In the sound field 307, of the transfer functions 305 between the speaker 304 and the three control target points 306, the transfer function A1 between the speaker 304 and 1 of the control target points 306.
It is assumed that only the transfer function AM between the speaker 304 and M of the control target point 306 is obtained by the measurement.
The transfer function estimator 308 estimates a transfer function A2 between the speaker 304 and 2 of the control target point 306, which has not been obtained by measurement, using A1 and AM which have already been obtained. As an estimation method, the transfer functions A1 and AM are converted into the frequency domain, and interpolation is performed for each of the real number part and the imaginary number part. In particular, a known interpolation method can be used as the interpolation method, and spline interpolation is particularly preferable.

The frequency band of the three transfer functions 305 including A2 thus obtained is limited by the band limiter 309. The frequency at which the band is limited has a large contribution to the intelligibility of the voice, and may be narrower than the range of 200 Hz to 8 kHz, particularly 400 Hz to 4 kHz.
And The output of the band limiter 309 is the target characteristic setting device 31.
0 and the coefficient calculator 312 are input.

The target characteristic setting unit 310 first calculates the initial response. The initial response is calculated by representing each transfer function 305 in the time domain and cutting out only an initial part including the direct sound. In order to maximize the clarity of the loud sound in the sound field 307, only the direct sound portion is used as the initial response. The direct sound can be detected by a visual method, and the time t at which the squared amplitude of each transfer function takes the maximum value in the time domain is obtained.
It is also possible to adopt a method in which the range is from t0-t1 to t0 + t1 [ms] with 0 as a reference. t1 is 5 [ms] or less, t2
Is preferably 10 [ms], and generally t1 is 1 [m
s] and t2 may be 5 [ms]. From the three initial responses set as described above, three transfer functions (A1, A2, A3) between the corresponding speaker 304 and the three control target points 306 are set.
AM) is subtracted and then a delay is added to obtain the target characteristic. The length of the delay time must be equal to or larger than the maximum value of the length of the duration of the three transfer functions 305 expressed in the time domain.

Next, the divergence prevention coefficient 311 is set.
As the divergence prevention coefficient 311 increases, the required coefficient length decreases, but the control result at the control target point departs from the target characteristic. Generally, the power spectrum of the transfer function 305 is calculated, the maximum value and the minimum value on the frequency axis are calculated, and if the ratio of both is within 40 dB, the divergence prevention coefficient (β) 31
1 may be 0, but when the ratio of the two exceeds 40 dB, a value obtained by subtracting 40 dB from the maximum value on the frequency axis may be used as the divergence prevention coefficient β.

The coefficient calculator 312 obtains the coefficient 313 by the method of least squares. The determinant can be solved in time series, but it is better to solve it in the frequency domain because the memory required for calculation increases as the response length increases. The solution in the frequency domain is described below. For each frequency, the target characteristic set by the target characteristic setting unit 310 is set to D (3,
1), transfer function 305 is A (3,1), and coefficient 313 is H
(1, 1), the divergence prevention coefficient 311 is β. However,
D, A and H are complex matrices, and subscripts (X, Y) mean X rows and Y columns. At this time, the least squares solution H is obtained by solving (Equation 2).

The above (formula 2) is a one-dimensional linear equation,
Although it can be solved by a known numerical solution, QL method or LU decomposition can be used in particular. After H is obtained for each frequency within the pass frequency of the band limiting filter 309, it is converted into the time domain to obtain one coefficient 313.
Is required. One coefficient 313 obtained in this way
Is set in the control filter 302.

When the audio signal input to the input 301 is amplified by the amplifier 303 through the control filter 302 and output from the speaker 304 as described above, the control target point 3
In 06, a clear voice signal can be heard, and even if the number of transfer functions to be measured is reduced by estimating a part of the transfer functions with other transfer functions, the intelligibility in a loudspeaker system can be improved. it can.

In this embodiment, the number of control speakers is one and the number of control target points is three.
Even if the number of control speeches is two or more and the number of control target points is four or more, it is obvious that the clarity can be improved by the same method.

(Third Embodiment) Next, referring to FIG.
A method and apparatus for improving the intelligibility of a loud sound according to the third embodiment of the present invention will be described. FIG. 3 is a diagram showing a configuration of a method and apparatus for improving the intelligibility of a loud sound according to the third embodiment of the present invention. In FIG. 3, 401 is an input, 402 is a variable coefficient filter as a time-varying filter, 403 is a filter as a time-invariant filter, 404 is an amplifier, 40
5 is a speaker, 406 is a transfer function, 407 is a control target point, 408 is a sound field, 408 is a microphone, 410 is a microphone, 411 is a band limiter, 412 is a target characteristic,
413 is a divergence prevention coefficient (β), 414 is a coefficient calculator, 4
15 is a coefficient, 416 is a filter, 417 is a target characteristic, 4
18 is an LMS calculator, 419 is an adder, 420 is a switch, and 421 is a filter.

Next, with reference to FIG. 3, the configuration of the method and apparatus for improving the intelligibility of the loud sound in the third embodiment will be described in detail. In FIG. 3, an input 401 is connected to N time-invariant filters 403 for controlling a sound field through a variable coefficient filter 402 for controlling a sound field. N filters 40 each
3 are respectively connected to N speakers 405 through N amplifiers 404. In the sound field 408, N speakers 405 are provided and M control target points 407 are set. N speakers 405 and M control target points 407
Between N and M, N × M transfer functions 406 are set.

The sound field 408 has a microphone 409.
And a microphone 410 is provided. Coefficient calculator 4
14 inputs the output of the band limiter 411, the target characteristic 412 and the divergence prevention coefficient (β) 413, and outputs the coefficient 415.
Is output. The transfer function 406 is set in the filter 416, and the same target characteristic as the target characteristic 412 is set in the target characteristic 417. The LMS calculator 418 is a variable coefficient filter 4
The coefficient W (N) of 02 is calculated. The switch 420 switches between the two microphones 409 and 410, and the adder 41
Reference numeral 9 subtracts the output of the switch 420 from the output of the target characteristic 417 and outputs the residual E (M) to the LMS calculator 418.

Next, with reference to FIG. 3, an operation procedure of the method and apparatus for improving the intelligibility of the loud sound in the third embodiment will be described. In the sound field 408, N speakers 405 and M
N × M transfer functions 40 between the control target points 407
6 is obtained by measurement or simulation. Here, 1 of the speaker 405 is a speaker for presenting a loud sound. Next, the target characteristic 412 is determined by using M transfer functions between 1 of the loud sound presentation speaker 405 and the M control target points 407. Represent each transfer function in the time domain,
Only the initial part including the direct sound is set as the target characteristic 412. For the clearest, only the direct sound part is set as the target characteristic 412.

The direct sound detection method is based on the time t0 when the squared amplitude of each transfer function in the time domain takes the maximum value.
It is possible to use a method in the range of 0-t1 to t0 + t1 [ms]. t1 is 5 [ms] or less, t2 is 10 [ms]
Is desirable, generally, t1 is 1 [ms] and t2 is 5 [m
s]. After subtracting M transfer functions 406 between 1 of the corresponding loudspeaker presentation speaker 405 and M control target points 407 from the M target characteristics set as described above, the delay is totally delayed. The final target characteristics are those with added time. The length of the delay time must be equal to or greater than the maximum value of the length of the duration of the N × M transfer functions 406 expressed in the time domain.

Further, N × M transfer functions 412 between N speakers 405 and M control target points 407 are input to the band limiter 411. The band limiter has a transfer function 406
And the frequency bandwidth of coefficient 415 is limited to the frequency bandwidth capable of transmitting voice information. Generally from 200Hz to 8
The frequency may be about kHz, and if it is desired to further reduce the frequency bandwidth, it may be 400 Hz to 4 kHz. Next, the divergence prevention coefficient 413 is set. The divergence prevention coefficient 413 is set for each frequency, and is set to a value 20 dB to 60 dB smaller than the maximum value of the power spectrum of the N × M transfer functions 406. In general, a value that is 40 dB smaller than the maximum value of the power spectrum of the N × M transfer functions 406 may be used as the divergence prevention coefficient 413. The coefficient calculator 414 uses a matrix to find a coefficient close to the target characteristic by the least square method.

The determinant can be solved in time series, but it is better to solve in the frequency domain because the memory required for calculation increases as the response length increases. The solution method in the frequency domain will be described below. Target characteristics 4 for each frequency
12 is D (M, 1), the transfer function 406 is A (M, N),
The coefficient 415 is H (N, 1), and the divergence prevention coefficient 413 is β. However, D, A, and H are complex matrices, and the subscripts (X, Y) mean X rows and Y columns. At this time, the least squares solution H is obtained by solving (Equation 1). (Equation 1) is an N-element linear simultaneous equation and can be solved by a known numerical solution method, but in particular, the QL method or LU decomposition can be used. After obtaining H for each frequency limited by the band limiter 411, N coefficients 415 are obtained by converting into H and then converting into H. The N coefficients 415 thus obtained are set in the N filters 403, respectively.

Next, a method of performing adaptive signal processing will be described. A plurality of known methods can be used as the adaptive signal processing method, but here, a known Filtered-XL method is used.
The method by MS will be described. For the target characteristic 417, the same target characteristic as the target characteristic 412 is set. Further, the sound signal of the sound field 408 is picked up by the microphone. Ideally, like the microphone 410 shown in FIG.
It is desirable to install in the vicinity of 7. In this case, the switch 420 is set to the A side. When it cannot be installed near the control target point 407 like the microphone 410, it is installed in a place away from the control target point 407 as shown by the microphone 409. At this time, the filter 421 may be set with a coefficient such that the points A and B in the switch 420 are equal to each other. When using the microphone 409, the switch 420 is set to the B side.

The adder 419 subtracts the output of the switch 420 from the output of the target characteristic 417, and the result is the residual E
Output as (M). Note that M is a microphone 409.
Or the number is 410. A transfer function 406 is set in the filter 416 and is set to A (M, N). The LMS calculator 418 uses the residual E (M) and the transfer function A (M, N) to output the coefficient W (N) of the coefficient variable filter 402 according to a known (formula 3) and update it. Wn + 1 = Wn-αAE (Equation 3) Note that α represents the size of the coefficient update step, the subscript n represents a discretized certain time, and n + 1 represents the next time of n.

As described above, the audio signal to the input 401 passes through the variable coefficient filter 402 and the filter 403, and then the amplifier 4
When the signal is amplified in 04 and output from the speaker 405, a clear audio signal can be heard at the control target point 407.
In addition, even if the sound field 408 changes, the variable coefficient filter 40
Since 2 is updated, the clear sound output from the speaker 405 can always be heard at the control target point 407. In this way, intelligibility in a loudspeaker system can be improved.

[0058]

The present invention is configured as described above, and in particular, the sound field control filter is divided into a plurality of control filters, for example, a low-frequency control filter and a high-frequency control filter for low-frequency control. By calculating the coefficients of the low-pass control filter and the high-pass control filter using only the filter, it is possible to reduce the calculation amount of the control filter as a whole and improve the clarity of a wide area. it can.

Further, the present invention is configured as described above, and in particular, the transfer function between the control speaker and another control target point is calculated by using the known transfer function between the control speaker and the control target point. Is estimated and the coefficients of the control filter are obtained using all transfer functions including the estimated transfer function, which reduces the number of transfer functions required to calculate the control filter coefficient and makes clear a wide area. The degree can be improved.

Further, the present invention is configured as described above, and particularly uses a time-varying filter and a time-invariant filter as a control filter for a voice input signal, and outputs the output of the control filter from a speaker. By obtaining the transfer function between the controlled point and calculating the coefficient of the time-invariant filter and the initial value of the coefficient of the time-variant filter, the sound field may change with time. Even in such a case, the optimum filter coefficient can always be obtained with respect to the change in the sound field, and the intelligibility of a wide area can be improved.

Further, the present invention is configured as described above, and in particular, the coefficient of the time-varying filter is constantly updated based on the acoustic signal picked up in the sound field after setting the initial value. Even when the sound field changes with time, the optimum filter coefficient can always be obtained with respect to the change of the sound field, and the clarity of a wide area can be improved.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of a method and a device for improving intelligibility of a loud sound according to Embodiment 1 of the present invention;

FIG. 2 is a diagram showing a configuration of a method and a device for improving intelligibility of a loud sound according to Embodiment 2 of the present invention;

FIG. 3 is a diagram showing a configuration of a method and a device for improving intelligibility of a loud sound according to Embodiment 3 of the present invention;

FIG. 4 is a diagram showing a configuration of a conventional method and apparatus for improving intelligibility of a loud sound in a sound field;

[Explanation of sign]

101, 201, 301, 401 input 102, 420 switch 103, 114 Low pass filter 104,119 High-pass filter 105 delay device 106 Low-pass control filter 107 High frequency control filter 108, 203, 303, 404 Amplifier 109, 204, 304, 405 speakers 110, 205, 305 Transfer function 111, 406 Transfer function 112, 206, 306, 407 Control target point 113, 207, 307, 408 sound field 115, 310 Target characteristic setting device 116, 210, 311, 413 divergence prevention coefficient 117, 211, 312, 414 Coefficient calculator 118, 212, 313, 415 Coefficient 120 target characteristic setting device 121 Divergence prevention coefficient 122 coefficient calculator 123 coefficient 202 FIR filter 208 Transfer function setter 209, 412, 417 Target characteristics 302 Control filter 308 Transfer function estimator 309 Band limiter 402 Coefficient variable filter 403, 416, 421 filters 409 and 410 microphones 411 Band limiter 418 LMS calculator 419 adder

Claims (5)

(57) [Claims] 1. When a voice input signal is divided into a plurality of frequency bands, and the divided voice input signals are output from a loud sound presentation speaker provided for each control filter through at least one control filter, A method for improving the intelligibility of a loud sound by determining the coefficient of each control filter by performing the step 3 for each frequency band. The first step is between the loud sound presenting speaker and the control target point. The transfer function is band-limited in its frequency band, the second step is the direct sound part of the impulse response based on the band-limited transfer function, and the third step is the band-limited transfer function and target characteristic. A method for improving the intelligibility of a loud sound by determining the coefficient of each control filter based on it. 2. A loud sound provided for each control filter by dividing a voice input signal into a low band and a high band, and passing the divided voice input signal through at least one low band control filter and high band control filter, respectively. A method for improving the intelligibility of a loud sound by determining the coefficient of each control filter by performing the first to fifth steps when outputting from the presentation speaker. The first step is to present the loud sound by outputting a low frequency band. The transfer function between the loudspeaker and the control target point is band-limited in the low frequency band, and the second step uses the direct sound part of the impulse response based on the band-limited transfer function as the first target characteristic, and the third step. Determines the coefficient of the low-frequency control filter based on the transfer function band-limited in the low frequency range and the first target characteristic. The fourth step is to determine the loudspeaker presentation speaker that outputs high frequency and the control target point. Bandwidth control of transfer function between The second target characteristic is obtained from the difference between the first target characteristic and the impulse response used in the second step, and the fifth step is to obtain a transfer function band-limited in the high frequency range and a second target characteristic. A method for improving the intelligibility of a loud sound by deciding the coefficient of a high-frequency control filter based on it. 3. A loudness sound intelligibility improving device comprising a low-pass filter, a high-pass filter, at least one low-pass control filter, and at least one high-pass control filter. The high-pass filter processes the input audio signal and outputs it to each low-pass control filter.The high-pass filter processes the input audio signal and outputs it to each high-pass control filter, which outputs each low-pass control filter. The filter performs a filtering process using the coefficient determined in the following 1-3 steps and outputs it to the loud sound presentation speaker, and each high frequency control filter uses the coefficients determined in the 1-3 steps below. The processed sound is output to the loudspeaker presenting speaker, and the first step band-limits the transfer function between each loudspeaker presenting speaker and the control target point in the frequency band, and the second step band-limits it. A based on the transfer function Direct sound part of the pulse response and target characteristics, the third step, intelligibility enhancement apparatus of amplified sound to determine the coefficients of the control filters on the basis of the transfer function and the target characteristics has been band-limited. 4. A loudness sound intelligibility improving device comprising a low-pass filter, a high-pass filter, at least one low-pass control filter, and at least one high-pass control filter. The high-pass filter processes the input audio signal and outputs it to each low-pass control filter.The high-pass filter processes the input audio signal and outputs it to each high-pass control filter, which outputs each low-pass control filter. The filter performs a filtering process using the coefficient determined in the following 1-3 steps and outputs it to the loud sound presentation speaker, and each high frequency control filter uses the coefficients determined in the 1-5 steps below. The processed sound is output to the loudspeaker presenting speaker, and the first step is to band-limit the transfer function between the loudspeaker presenting speaker outputting the low frequency band and the control target point in the low frequency band, Based on band-limited transfer function The direct sound part of the impulse response is set as the first target characteristic, and the third step is to determine the coefficient of the low-pass control filter based on the transfer function band-limited in the low band and the first target characteristic. In the step, the transfer function between the loud sound presenting speaker that outputs a high frequency band and the control target point is band-limited in the high frequency band, and the second target characteristic and the impulse response used in the second step are used for the second And a fifth step is an intelligibility improving apparatus for loud sound, which determines a coefficient of a high-frequency control filter based on a transfer function band-limited in a high frequency band and a second target characteristic. 5. A loudness sound intelligibility improving device comprising a time-varying filter and a plurality of time-invariant filters, wherein the time-varying filter filters an input voice signal by a coefficient and a sound field. The coefficient is constantly updated with the sound collection signal from the microphone set to, and each time-invariant filter filters the signal from the time-varying filter with the coefficient determined in the steps 1-3 below. Output to a loudspeaker presentation speaker, the first step band-limits the transfer function between each loudspeaker presentation speaker and the control target point, and the second step is the impulse response based on the band-limited transfer function. An apparatus for improving the intelligibility of a loud sound, in which the direct sound portion is used as a target characteristic, and the third step is to determine the coefficient of the time-invariant filter based on the band-limited transfer function and target characteristic.