JP2001042888A - Method of setting microphone, method for processing sound signal, and devices for inputting, recording and processing sound signal, using these methods and voice recognition processing device and recording medium recording program for processing sound signal - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the operation quantity for arranging phases of sound signals. SOLUTION: Concentric spheres whose respective radii change by λwhich is a moving distance of sound during a sampling period T by making a sound source S as the center are virtually set. Respective microphones M0, M1, M2 and M3 are mounted on the basis of peripheral surfaces of the spheres C0, C1, C2 and C3 selected from among the concentric spheres. The output signal of every microphone Mn is sampled with a fixed sampling period, thereafter, is delayed by the sampling number (n) corresponding to the difference n λof sphere radius between the sphere Cn which is the setting reference of the respective microphones and the sphere C0 which is the setting reference of the microphone M0 which is the most remote from the sound source S. Further, by summing the delayed sampling values and the sampling values at the present time of the microphone M0, the sound signal from the sound source S is extracted.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a technical field for extracting a sound generated from a predetermined spatial position by using a plurality of microphones, and more particularly, to the installation of each microphone and the installed microphone. Related to techniques for processing sound signals from a computer.

[0002]

2. Description of the Related Art To take a sound signal generated by a specific speaker into a computer and perform a predetermined recognition process or to digitize and record the sound signal, the collected sound signal is converted into noise in the surrounding environment. It is necessary to remove such a signal. Conventionally, for this kind of purpose, a microphone array in which a plurality of microphones are arranged in a row, in-phase and emphasizes speech from the direction where the speaker is, while canceling out noise from other directions. It has been proposed to reduce it.

In conventional sound signal processing using a microphone array, a convolution operation using a delay filter is performed in order to correct a time difference when sound emitted from a sound source at a predetermined time reaches each microphone. After that, the sum of each sampling is calculated to generate a signal in which the sound signal of the same phase from the sound source is strengthened.

FIG. 23 shows a schematic configuration in a case where sound signal processing by the microphone array is performed by hardware. Each microphone M _n (n = 0 to 3 in the figure) is equally spaced on a straight line separated by a predetermined distance from a sound source (not shown) virtually set as the position of the speaker's mouth. And installed. For each microphone _Mn , a microphone amplifier 50 (hereinafter simply referred to as “amplifier 50”), an A / D converter 51, and a sound signal storage unit 52, a filter coefficient storage unit 53, a convolution arithmetic unit 54, and the like are set. Is done.

The sound signal output from each microphone M _n is amplified by an amplifier 50 and then sampled by an A / D converter 51 at a predetermined sampling cycle. The sound signal storage unit 52 and the filter coefficient storage unit 53
Have a data capacity of NT (NT> 0), the sound signal storage unit 52 stores NT sampled values including the current time, and the filter coefficient storage unit 53 stores
The NT-order filter coefficient corresponding to each sampling value is
Each is stored. Each convolution calculator 54 calculates the sum of the multiplication value of each sampling value held in the sound signal storage unit and the filter coefficient at that time in accordance with the sampling period, and calculates the calculated value R _n as another value. Is output as correction data in which the phase shift from the sampling value of the microphone is corrected. Further correction data R _n for each microphone is averaged is input to the adder, and output as a sound signal O of the digital quantity for recognition processing (t).

The above-described convolution operation is actually performed by
This is obtained by executing the following equation (1) on the past NT sampling data for each microphone _Mn in the computer. (1)
In the equation, D (t) is a sampling value obtained by the t-th sampling. H (k) is a delay filter corresponding to the sampling value k times before the current time, and is calculated by Expression (2), where T is the sampling period and DT is the delay time.

[0007]

(Equation 1)

[0008]

(Equation 2)

FIG. 24A shows a flow of processing when the processing of FIG. 23 is executed by a computer. In the drawing, n indicates a counter for specifying the microphone to be processed, t indicates a counter for counting the number of times of sampling, and N indicates the number of microphones installed.
MAX is equivalent to the number of times of sampling performed within a predetermined period (for example, one second), and O (t) is a signal output at the time of the t-th sampling (hereinafter O (t) is output signal O (t) ).

Prior to the processing, the control unit calculates a time difference at which a sound signal from a predetermined direction reaches each microphone based on the installation position of each microphone, and calculates a shift amount DT and a sampling period T. (T = 1sec / MA
X) is applied to equation (2) to calculate a delay filter H (k) (ST2). Thereafter, the control unit resets each counter and the output signal O (t) to zero in ST2 to ST4,
In the loop of T5～8, for each microphone, while calculating correction data R _n by the convolution operation, by repeating the calculated addition operations of the correction data R _n, O
The value of (t) is updated.

In particular, in ST5, ST5- in FIG.
By executing the procedures of 1 to ST-9, the sampling values D (p) (t ≧ p ≧ t−
NT + 1) are sequentially read from the memory, and each sampling value D (p) is converted to a k-th filter coefficient H (k).
(0 ≦ k ≦ N−1), and the multiplied values are sequentially added to calculate the correction data R _n .

[0012] ST5~8 loop is completed, control unit has been represented as a sum of the correction data R _n O
(T) is divided by the number M of microphones, averaged, and output as an output signal at the time of the t-th sampling. In the same way, each time samples the sound signal, respectively, after correcting by the algorithm of FIG. 24 (2) the sampling values of each microphone, the correction data R _n is averaged signal an output signal O ( Output to the outside as t).

[0013]

According to the control described above [0008], for N microphones, other respective NT multiplication processing, update processing and the correction data R _n in ST5-6 ST5-7,5- Since the increment processing of the counters p and k at step 8 is performed, FIG.
By executing the above algorithm only once, a total of 4 * NT calculations are performed. In addition, addition processing of ST6 and ST7 in FIG. 5A is performed for each of the N microphones. Therefore, the division processing in ST9 and S
Including the increment processing of the counter t at T10, [N * (4 * NT + 2) +
2] calculations are executed.

In order to generate digital data necessary for voice recognition processing from sound signals from the microphones, each sound signal is sampled at a frequency several times that of a sound wave, and the number of taps NT is set to each of ten or more microphones. It is necessary to perform a convolution operation set to about 100. However, when performing calculations with such settings, the amount of calculation per second exceeds hundreds of millions of times, so that it cannot be realized at all by processing using a general-purpose CPU incorporated in a personal computer or the like.

Therefore, in order to perform this arithmetic processing, it is necessary to operate a plurality of CPUs in parallel or to perform processing using a DSP, but this time, the apparatus configuration becomes large and the cost increases. Problems arise. In particular, in recent years, the technical level of voice recognition processing is approaching a level at which processing such as data input to a personal computer and operation of a vending machine can be realized. It is necessary to be able to generate a sound signal for input at low cost and in a short time.

The present invention has been made in view of the above problems, and the amount of calculation for aligning the phases of the sound signals from the microphones is greatly reduced, and the configuration is simplified when processing by hardware is performed. It is another object of the present invention to enable a general-purpose CPU to sufficiently perform processing by software.

Further, according to the present invention, it is possible to greatly reduce the amount of calculation for aligning the phases of sound signals, and to install each microphone along a straight line or on an installation surface such as a front surface of a device. Accordingly, it is a second object to provide a practical device by simplifying the configuration relating to voice input.

[0018] In addition, according to the present invention, even if the installation position of the microphone with respect to the sound source is inaccurate, by adjusting the sampling frequency, the phase shift of the sound signal captured by the microphone is kept within an allowable value range. A third object is to stabilize the processing.

[0019]

Assuming that a sound is generated at a predetermined spatial position in a space where there is no object blocking the sound, the sound is the same from the sound source toward the surrounding directions. It will progress at the speed of. Therefore, the arrival position of the sound at a point in time when a predetermined time has elapsed since the sound was generated is represented by a peripheral surface of a sphere whose radius is the distance from the sound source to which the sound travels within the elapsed time. Similarly, the arrival position of the sound from the sound source is determined by a predetermined period T
When the sound is extracted every time, the arrival position of the sound every hour is represented by each peripheral surface of a concentric sphere whose radius increases by the distance λ where the sound travels during the period T around the sound source. become. In other words, the sound emitted from the sound source at a certain time arrives at each peripheral surface of the concentric sphere with a delay of the period T.

FIG. 1 shows an example in which a microphone is installed by applying the above principle. For a sound source S virtually set at a predetermined spatial position, a radius around the sound source S is set to the distance λ. Four changing concentric spheres C _{0 to} C ₃ are virtually set, and the microphones M ₀ to M ₀ are respectively set on the peripheral surface of these spheres, specifically, at the intersection of a straight line passing through each sphere and the sphere.
It is established the M _3. In the illustrated example, the largest sphere C ₀ of concentric spheres on the basis, said the sound from the sound source S is assumed to have reached the time t on the peripheral surface of the sphere C _0, each sphere C ₁ ~ inside against the peripheral surface of the C _3, the same sound at an earlier time by the period T respectively it will have been reached.

That is, the reference sphere C ₀ and each inner sphere C ₁
When the difference in radius between the -C ₃ put a nλ (n = 1,2,3), on the peripheral surface of each sphere C ₁ -C _3, sound from the sound source S to a point in time (t-nT) Has been reached. Therefore, the sound arriving at the peripheral surface of each ball C _{1 to} C ₃ from the sound source S is
By delaying each nT time, each ball C ₁
It becomes possible to take out align the phase of a sound signal that has reached the peripheral surface of the sound signal and the reference sphere C ₀ that has reached the peripheral surface of -C _3.

On the other hand, when a sound is generated at a spatial position other than the sound source S, the position at which the sound is generated and the concentric spheres C _{0 to} C _{3 are set.}
Does not have a regular relationship with the period T as described above, so even if the sound arriving at the peripheral surface of each of the spheres C _{0 to} C ₃ is delayed by nT time, the phase of the sound signal is Not aligned.

[0023] Thus, while the output signal from the microphone M ₀ ~M ₃ disposed on the peripheral surface of each sphere C ₀ -C ₃ sampled by the period T, the sampled value respectively, the microphone M ₀ ~M _The delay is delayed by the number of samplings n according to the difference nλ between the radius of the sphere and the reference criterion of ₃ (however, the delay amount for the microphone M ₀ is zero), and the sum of the delayed sampling values for each microphone is calculated. Then, while the sound emitted from the sound source S at the same time is emphasized, the sound emitted from another position is canceled, and the sound from the sound source can be accurately extracted.

The invention of claim 1 is based on the above-mentioned principle, and concentric spheres whose diameters change sequentially by a sound traveling distance corresponding to a fixed sampling period around a sound source are virtually assumed. After the setting, the microphones are set using the peripheral surfaces of a plurality of spheres selected from the concentric spheres as positioning criteria. In this case, as shown in FIG. 1, each microphone is installed not only on each of the peripheral surfaces of a plurality of spheres whose radius changes by λ,
An arbitrary sphere is selected from the concentric spheres, and is set on the peripheral surface of these spheres or at a position near the peripheral surface.

According to the second aspect of the present invention, as shown in FIG.
Each microphone is installed at a position separated by a predetermined distance from the sound source with reference to a position at which a straight line intersecting each ball and the peripheral surface of each ball intersects. According to the third aspect of the present invention, a microphone installation surface is virtually set instead of a straight line, and the microphone is installed with reference to a position where the installation surface intersects with the peripheral surface of each ball. The “installation surface” in claim 3 is not limited to a flat surface but may be a curved surface. The installation surface is desirably opposed to the target sound source, but is not necessarily limited to this.

According to a fourth aspect of the present invention, there is provided a method for receiving and processing audio output signals from a plurality of microphones installed toward a sound source at a predetermined spatial position, wherein each of the sphere diameters is centered on the sound source. After virtually setting concentric spheres that sequentially change by a traveling distance of a sound corresponding to a fixed sampling period, each microphone is set to a plurality of spheres selected from the concentric spheres based on their installation positions. And the output signal of each microphone is sampled at the sampling period, and the sampling value of each microphone is compared with the sphere corresponding to the microphone and the sphere corresponding to the microphone farthest from the sound source. The addition is performed by sequentially delaying by the number of samplings based on the difference between the ball diameters.

Further, according to the fifth aspect of the present invention, it is possible to cope with a case where a phase difference occurs between a sound signal from a sound source taken by a microphone and a sound signal taken on a peripheral surface of a sphere associated with the microphone. Therefore, when the microphone is associated with the peripheral surface of a predetermined sphere, a sampling period is set such that the phase shift falls within a predetermined error, and the output signal of each microphone is set according to the sampling period. Sampling and mapping to concentric spheres.

According to a sixth aspect of the present invention, there is provided a sound signal input device including a plurality of microphones for collecting sounds from a sound source virtually set at a predetermined spatial position at a plurality of locations, wherein each microphone is a sound source. Are installed using the method of claim 1 for the virtual position.

According to a seventh aspect of the present invention, there is provided an audio input device comprising a plurality of microphones having the same configuration as in the sixth aspect, and a signal processing device for generating and outputting an input signal from an output signal of each microphone. It is. The signal processing device,
Sampling means for sampling the output signal of each microphone at a sampling cycle when a concentric sphere as a reference for setting each microphone is set, and a sampling value of an output signal for each microphone, and a sphere serving as a reference for setting the microphone, and A memory for storing the number of times of sampling based on the difference between the diameter of the microphone and the sphere serving as the installation reference of the microphone located farthest from the virtual sound source, and at the time of each sampling, the oldest sampling value for each microphone from the memory. Reading means for calculating the sum of these sampling values and the current sampling value of the microphone located farthest from the virtual sound source; and a digital signal indicating the sum of the sampling values obtained by the calculating means. Output means for outputting the To.

On the other hand, in the voice input device according to the invention of claim 8, each microphone is installed in the same manner as in claim 7, and the signal processing device also includes the same sampling means, memory, and arithmetic means as in claim 7. To equip it. Further, the signal processing device is provided with output means for converting the sum of the sampling values obtained by the arithmetic means into an analog signal and outputting the converted signal to the outside.

According to a ninth aspect of the present invention, there are provided a plurality of microphones having the same structure as those of the sixth to eighth aspects, a signal processing device for taking in an output signal of each microphone and generating a sound signal for recording, And a storage processing device for storing the sound signal in a memory or a predetermined recording medium. The signal processing device includes the same sampling means, memory, and calculation means as in claims 7 and 8, and the storage processing device converts a sound signal corresponding to the sum of the sampling values obtained by the calculation means into a sound signal for recording. It is configured to write to the memory or the recording medium as a signal. As the recording sound signal, a digital sound signal obtained by the arithmetic means may be used as it is, but the present invention is not limited to this, and an analog-converted sound signal may be recorded.

[0032] The inventions of claims 10 and 11 relate to a sound signal processing device for taking in and processing output signals of a plurality of microphones for collecting sounds from a sound source at a plurality of locations. A sound signal processing apparatus according to a tenth aspect of the present invention is a sound signal processing apparatus for sampling an output signal of each microphone at a constant sampling cycle, and for sequentially storing a predetermined number of sampling values obtained by the sampling means for each microphone. A memory and a concentric sphere in which each sphere diameter is sequentially changed by a sound traveling distance corresponding to the sampling period around the virtual position of the sound source as a center is virtually set, and each microphone is set based on its installation position. An associating means for associating with the peripheral surface of a predetermined sphere of the concentric spheres, and at the time of each sampling, for each microphone, the sphere corresponding to the microphone and the microphone farthest from the virtual position of the sound source. The sampling value delayed by the number of samplings based on the difference between the sphere diameter and the sphere Serial reading from the memory, and a calculating means for calculating the sum of the sampling value read out.

According to the eleventh aspect of the present invention, in order to execute the method of the fifth aspect, a sampling period setting means and a control means are added to the apparatus configuration of the tenth aspect. The sampling cycle setting means sets a sampling cycle for each microphone such that the phase shift of the sound signal described in the fifth aspect falls within a predetermined error.
The control means controls each means so that the sampling means and the associating means operate in accordance with the respectively set sampling period.

According to a twelfth aspect of the present invention, for a sound source virtually set at a predetermined spatial position, a plurality of microphones installed by the method of the first aspect and output signals of these microphones are fetched and processed. A sound signal processing device including a control device. According to the twelfth aspect of the present invention, the control device includes the same sampling means as in the seventh to ninth aspects, a memory for sequentially storing a predetermined number of sampling values for each microphone, and a microphone for each sampling. For each, read from the memory a sampling value delayed by the number of samplings based on the difference between the ball diameter of the sphere as the microphone installation reference and the sphere as the microphone installation reference farthest from the virtual position of the sound source, and Calculating means for calculating the sum of the read sampling values.

Further, the sound signal processing device according to the present invention has a microphone installation surface which is disposed opposite to the virtual position of the sound source by a predetermined distance.
According to the thirteenth aspect of the present invention, each microphone is installed on the installation surface using the intersection of a peripheral surface of a selected one of the concentric spheres and a predetermined straight line on the installation surface as a reference for positioning. . On the other hand, in the invention of claim 14, each microphone is installed on the installation surface with reference to the intersection of the peripheral surface of the selected concentric sphere and the entire installation surface. In addition, the installation surface of the microphone can be set on one surface of the body of the body constituting the control device, or can be set at a position independent of the control device.

According to the fifteenth aspect of the present invention, a plurality of microphones are installed on the peripheral surface of the sphere selected as the microphone installation reference.

According to a sixteenth aspect of the present invention, in addition to the configuration of the twelfth aspect, in order to enable processing of a sound signal generated at a position different from the virtual position,
The apparatus includes sound source position estimating means for estimating the actual position of the sound source using output signals of the microphones, sampling period setting means, and control means. When the estimated position of the sound source by the sound source position estimating unit is different from the virtual position, the sampling period setting unit virtually sets each sphere diameter around the estimated position to sequentially change by a sound traveling distance corresponding to a fixed sampling period. Assuming that the concentric sphere is set in a predetermined manner, a sampling value is set such that a phase shift based on the same principle as in claims 5 and 11 falls within a predetermined error. The control device controls the sampling means so as to perform processing at the set sampling period, and sets a reading position of a sampling value for each microphone by the arithmetic means to a concentric sphere centered on the estimated position of the sound source. It changes according to the correspondence of microphones.

A seventeenth aspect of the present invention provides a control device for inputting a sound signal and performing a predetermined voice recognition process, a monitor for displaying a result of the voice recognition process by the control device, and a spatial position in front of the monitor. And a plurality of microphones for collecting sounds from a sound source virtually set as described above and inputting the collected sounds to the control device. 2. The microphone according to claim 1, wherein the microphone is located on a peripheral portion of a display screen of the monitor and at an intersection of a peripheral surface and a peripheral portion of a plurality of selected concentric spheres according to the method of claim 1 around the virtual position of the sound source. Is set as a reference for positioning. The control device includes a sampling unit, a memory, and a calculation unit similar to those of the twelfth aspect, and a recognition processing unit that processes time-series data based on a calculation value obtained by the calculation unit and performs a predetermined voice recognition process. I have it.

The speech recognition processing apparatus according to the eighteenth aspect of the present invention has a plurality of microphones installed on one surface of the machine, and a control device having the same configuration as that of the seventeenth aspect is incorporated in the machine. The installation surface of the microphone has a plurality of concentric spheres virtually set by the method of claim 1 centered on a sound source virtually set at a predetermined space position facing the installation surface. Each microphone is installed using the intersection of the peripheral surface and the installation surface as a reference for positioning.

Each of the nineteenth and twentieth aspects of the present invention is a recording medium in which a program for processing by a computer the output signals of a plurality of microphones for collecting sounds from a sound source at a plurality of locations is recorded. In the recording medium according to the nineteenth aspect, a procedure for sampling an output signal of each microphone at a fixed sampling cycle using an A / D conversion means built in a computer, and a predetermined number of sampling values for each microphone are stored. A step of sequentially storing the concentric spheres in the memory according to the method of claim 1, centering on a virtual position of the sound source, and selecting each of the microphones from among the concentric spheres based on the respective installation positions. Procedure for associating with the peripheral surface of a plurality of spheres, at each sampling point, for each microphone according to the difference in sphere diameter between the sphere corresponding to that microphone and the sphere corresponding to the microphone farthest from the virtual position of the sound source The sampling value delayed by the number of sampling times read out from the memory is read out, and each sampled value read out is read out. Procedure for calculating the sum of values, each step of is recorded.

Further, in the recording medium according to the twentieth aspect, there is provided a procedure for setting a sampling frequency such that a phase shift of a sound signal according to the same principle as in the fifth and eleventh aspects falls within a predetermined error. A program is recorded for executing a procedure of sampling the sound signal of each microphone and a procedure of controlling each microphone to correspond to a concentric sphere at a set sampling cycle.

[0042]

According to the first aspect of the present invention, assuming that a sampling period for each microphone is T and a distance at which sound travels within the sampling period T is λ, a concentric sphere whose radius changes by λ around the sound source is obtained. Each microphone is virtually set, and each microphone is set based on the peripheral surface of a plurality of spheres selected among them. Therefore, the sampling value for each microphone is calculated as the number of times of sampling according to the difference between the diameter of the sphere used as the reference of the microphone and the sphere used as the reference of the microphone farthest from the sound source, specifically, the radius of each sphere. (I is an arbitrary integer)
By delaying i times of sampling times corresponding to the above, it is possible to extract the sound signals reaching the respective microphones from the sound source in the same phase. Therefore, there is no need to perform an enormous amount of arithmetic processing as in the case where a conventional delay filter is used.

According to the sixth aspect of the present invention, the sound signals collected at a plurality of positions can be easily processed by arranging the microphones according to the above-mentioned method for the sound source virtually set at a predetermined spatial position. A possible sound signal input device is generated.

According to the second aspect of the present invention, as shown in FIG. 1, by placing each microphone at a position where a straight line and the peripheral surface of each sphere intersect, each microphone is arranged in a line. . According to the third aspect of the present invention, a microphone installation surface is virtually set instead of a straight line, and the microphone is installed at a position where the installation surface intersects with the peripheral surface of each sphere. It is installed with a two-dimensional spread such as a rectangular shape. Further, when the installation surface is a curved surface, each microphone can be installed three-dimensionally.

According to the fourth aspect of the present invention, when each microphone is installed according to the method of the first aspect, the sampling value of the output signal of each microphone is set to the sphere diameter of the sphere serving as the installation reference. By delaying by the corresponding number of times of sampling, the sound signal from the sound source that has reached each microphone can be extracted with the same phase.
Further, by calculating the sum of the sampling values including the signals having the same phase, a sound signal in which the sound signal from the sound source is emphasized is generated. Claim 10
According to the present invention, the sound signal processing device capable of executing the method of claim 4 is configured. By connecting the sound signal input device of claim 6 to this device, the output signal of each microphone can be processed. .

According to the seventh aspect of the present invention, the output signals of the microphones installed on the basis of the circumference of the sphere whose radius changes by λ around the virtual position of the sound source are sampled at the sampling period T. At the same time, the sampling values for i times based on the difference in the ball diameter between the sphere as the installation reference of each microphone and the reference sphere farthest from the sound source are stored in the memory. Then, by reading out the oldest sampling value for each microphone from the memory, a sampling value including the same sound signal from the sound source is extracted. Further, by calculating the sum of these sampling values and the current sampling value of the microphone located farthest from the virtual sound source, a sound signal of a digital amount in which the sound signal from the sound source is enhanced is generated, and speech recognition is performed. It is output to an external device such as a device for processing and recording of digital audio.

According to the eighth aspect of the present invention, the same signal processing as in the seventh aspect is executed in the sound signal input device for inputting the analog sound signal in which the sound from the sound source is emphasized.
According to the ninth aspect of the present invention, in a device for recording a sound signal in a digital or analog system, a sound signal for recording is created by the same processing as in the seventh and eighth aspects.

According to a twelfth aspect of the present invention, a plurality of microphones installed by the method of the first aspect and a control device for processing output signals of these microphones by the method of the fourth aspect allow input of a sound signal. A sound signal processing device having a function is configured.

According to a thirteenth aspect of the present invention, in the sound signal processing device of the twelfth aspect, a microphone installation surface is provided at a position facing the virtual position of the sound source, and the microphones are arranged in a line on the installation surface. Will be installed. According to the fourteenth aspect, the microphones are two-dimensionally or three-dimensionally arranged on the same installation surface. In addition, if this installation surface is made independent of the control device, the sound signal input section becomes a device configuration external to the control device, and it is possible to freely set the sampling position of the sound signal according to the position of the sound source. Become.

According to the fifteenth aspect of the present invention, a plurality of microphones are arranged on the peripheral surface of one sphere. Therefore, even if a large number of spheres are not set as the microphone installation standard, the installation standard can be maintained. By delaying the output signal of the microphone having the same sphere by the same number of times of sampling and calculating the sum of the sampling values, noise is removed and a sound signal in which the sound signal from the sound source is enhanced is extracted. Become like In addition, if the output signal is superimposed on each microphone having the same sphere as the installation reference at the stage of the analog signal, the number of sampling processes and the number of arithmetic processes are reduced.
Further decrease.

According to the seventeenth aspect of the present invention, on the periphery of the monitor for displaying the result of the speech recognition processing, a concentric sphere whose radius changes by λ around the virtual position of the sound source and the periphery of the monitor A microphone is installed based on the intersection position. Therefore, when the user utters the contents of the input data toward the display screen of the monitor, a digital amount of sound signal representing the user's voice is extracted from the sound signal captured by each microphone, and the input data is recognized by a recognition process for the signal. Will be recognized.

According to the eighteenth aspect of the present invention, one surface of the machine facing the virtually set sound source is a microphone installation surface, and the intersection position with the concentric sphere where the radius changes by λ around the sound source on this installation surface. When the microphone is installed on the basis of the reference, the user utters the contents of the input data toward the installation surface of each microphone, and the audio signal is captured to extract a digital sound signal representing the user's voice. You. Further, the content of the input data is recognized by the recognition process for the signal.

According to the nineteenth aspect of the present invention, the program recorded on the recording medium is introduced into a computer having A / D conversion means, so that the output signal of each microphone installed according to the method of the first aspect is obtained. An apparatus is created that processes and extracts sound signals from predetermined spatial locations.

According to the fifth, eleventh and twentieth aspects of the present invention, when the microphones are set off from the respective peripheral surfaces of the associated sphere, the microphones are located on the peripheral surface of the sphere at the ideal position. In this case, a sound signal having a phase shifted from that in the case where the sound signal is installed is taken in. However, in this case, the sampling frequency for determining the diameter of each concentric sphere is adjusted to a value such that the above-mentioned phase shift falls within a predetermined error. The effect is also kept to an acceptable level.

According to the sixteenth aspect of the present invention, when the estimated position of the sound source is different from the virtually set position, each microphone is placed in a concentric sphere when sampling is performed around the estimated position of the sound source. The signal processing is performed by associating with the peripheral surface. In this case, each microphone deviates from the peripheral surface of the newly associated sphere, but the sound signal captured by each microphone and the sound signal captured at an ideal position are similar to the inventions of claims 5 and 11. Is adjusted to a value that falls within a predetermined error, so that the effect of the microphone position shift on the final calculation result can be suppressed to an allowable range.

[0056]

FIG. 1 shows a microphone arrangement according to the present invention.
An example of the placement method is a method of processing the output signal of each microphone.
Shown together. S in the figure is a sound virtually set in the space
Source, and a fixed sample
The radius changes by the distance λ the sound travels in the gear cycle T
Such a concentric sphere is virtually set. A predetermined distance from the sound source S
h, a straight line P is virtually set at a position separated by h
Are set in order from a predetermined position toward the sound source S.
Four balls C₀~ C_ThreeAbout each sphere C ₀~ C_ThreeDirectly
Microphone M at the position where line P intersects₀~ M_ThreeIs installed
It is.

In FIG. 1 described above, the "hands for solving the problem"
According to the principle described in the column, “
Clohon M₀Sphere C corresponding to₀Each micro
Hong M _nThe sampling value for the output signal of
Microphone M_nBall C that serves as an installation standard_nAnd the group
Quasi ball C₀Number of samplings according to the difference in radius nλ
n, and sum the delayed sampled values
calculate. With this, a sound signal from the sound source S with a uniform phase
Generated at a position different from the sound source S while the
The signals will be canceled out and weakened,
Sound from the sound source S can be accurately extracted.
You.

In each of the specific examples shown in FIG. 1 and FIGS. 3 and 4 described later, the microphones are arranged in order on the peripheral surface of a plurality of spheres whose radius changes by λ. It is not necessary to set a microphone for each peripheral surface of the sphere. An arbitrary number of spheres at arbitrary positions are selected from concentric spheres extending infinitely around the sound source S, and the spheres of these selected spheres are selected. A microphone may be installed on each surface. As will be described later in detail, it is also possible to install two or more microphones on the peripheral surface of one sphere. If the sampling values for the microphones installed on the peripheral surface are processed with the same delay time, the microphones can be installed with a two-dimensional or three-dimensional spread. Moreover, according to such an apparatus, it is possible to superimpose an output signal for each microphone having the same installation reference, so that the amount of calculation can be reduced and noise can be accurately removed. Hereinafter, the present invention will be described in detail with reference to the embodiments to which the principle of FIG. 1 is applied.

(1) Signal Processing by Hardware FIG. 2 shows a schematic configuration of a microphone array 1 to which the microphone installation method of FIG. 1 is applied. Each of the microphones M _{0 to} M ₃ has a predetermined width. Fig. 1 on the installation surface
Are arranged in a line at intervals according to the principle of.

FIG. 3 shows a digital voice recorder using the microphone array 1 of FIG. In this apparatus, each microphone M ₀ ~M _3, microphone amplifier 2
(Hereinafter simply referred to as “amplifier 2”) and A / D converter 3 are set, and sphere C ₀ corresponding to microphone M ₀ farthest from sound source S (hereinafter referred to as “reference sphere C ₀ ”) for the microphones M ₁ ~M ₃ corresponding to the inside of the sphere C ₁ -C _3, FIFO memory 4A~4C for temporarily storing the sampled sampled value by the a / D converter 3, respectively, are deployed than.
Further, at the subsequent stage, an adder 5 and a storage device 6 are provided.

Each A / D converter 3 samples an output signal from each of the microphones M _{0 to} M ₃ at a constant sampling period T to generate a digital amount of audio signal. Each of the FIFO memories 4A to 4C stores a sampling number n according to a difference nλ between a radius of a sphere C _{n serving as} an installation reference of the corresponding microphone M _n (n = 1, 2, 3) and the reference sphere C _0. Data capacity.

In this embodiment, each of the FIFO memories 4A to 4A
C, the sampling values applied to the microphones M _{1 to} M ₃ are delayed by n samples (n = 1, 2, 3, 3), and then the delayed sampling values and the microphone M
The current sampling value from ₀ is input to the adder 5 to calculate the sum. The result of this calculation is
This corresponds to a signal obtained by in-phase and averaging sound signals reaching the microphones M _{0 to} M ₃ from the sound source S.

The storage device 6 is a memory for storing a digital amount of sound signal, or an FD, CD-R, MO
And the like, and a device for writing data to the recording medium. By outputting the calculation result of the adder for each sampling every hour to this storage device, the time series data obtained by digitizing the sound uttered by the sound source S is recorded.

FIG. 4 shows an analog voice input device using the microphone array 1 of FIG. In this device,
For each microphone M ₀ ~M _3, Figure 3 the same amplifier and 2, A / D converter 3, in addition to the FIFO memory 4A~4C is deployed, the adder 5 and the D / A converter 7 is provided. The adder 5, similar to the embodiment of FIG. 3, by calculating the sum of the sampling values at the present time from the sampling values and the microphone M ₀ which is delayed for the microphone M ₁ ~M _3, from the sound source S The sound signal having the same phase is extracted. The D / A converter 7 is connected to an output terminal of an analog sound signal (not shown), converts the sum of the sampling values obtained by the adder 5 to analog, and then outputs the converted sum to the outside via the output terminal.

The device having the configuration shown in FIG. 4 can be connected to a device having an input terminal for an analog sound signal, such as a personal computer, a recording device, and a telephone. Get higher. In addition, although the configuration is simple, the processing according to the principle shown in FIG. 1 is performed, so that noise generated at a spatial position other than the sound source S can be accurately removed. To produce a digital sound signal input device that operates on the same principle,
The output from the adder 5 may be directly output without passing through the D / A converter 7. Also, D in FIG.
By integrating the configuration up to the connection of the recording device to the / A converter 7, an analog voice recorder can be created.

(2) Installation of Microphone In the example of FIGS. 1 to 4, one microphone is installed on each of the virtually set spheres C _{0 to} C _3. As the microphone array above,
As shown in FIG. 5, two microphones M ₀ are provided for each of the spheres C _{0 to} C ₃ based on the intersection positions of the spheres C _{0 to} C ₃ and the straight line P.
~M _3, it is possible to place the _M'0 ~M' _3. In this case, the pair of microphones M _n and M ′ _n arranged on the peripheral surface of the same sphere C _n are arranged at symmetrical positions via the center point o of the array. _3, by determining either a distance w or microphone array across the array length l, between _M'3, using a distance λ going from the sound source S of the distance h and sound within one sampling period to the array surface , Each microphone M _{0 to} M ₃ ,
The positions of M ′ _{0 to} M ′ ₃ can be calculated.

FIGS. 6 and 7 show a method of determining the installation positions of the microphones, taking as an example a case where four microphones are installed on both sides of the center point o.
In each of the drawings, only the installation positions of the microphones M _{0 to} M ₃ on the right side of the center point o are shown, but each of the microphones M ′ _{0 to} M ′ ₃ on the left side is based on the center point o. It will be placed in the position of the microphone M ₀ ~M ₃ and symmetric.

FIG. 6 is a diagram for determining the distance r _n from the center point o of the array to the installation position of each microphone M _n using the total length 1 of the microphone array.
In this case, the distance d ₀ between the microphone M ₀ farthest from the sound source S and the sound source S can be calculated by Expression (3). Assuming that the microphones M _{0 to} M ₃ are sequentially arranged on the peripheral surface of the concentric spheres C _{0 to} C ₃ whose radius decreases by λ, each microphone M _n (n = 1,
2,3) the distance d _n between the sound source S will be denoted as d _n = d ₀ -nλ. Thus the distance r _n from the center point o of the array to the microphone M _n is calculated by the equation (4).

[0069]

(Equation 3)

[0070]

(Equation 4)

FIG. 7 shows a case in which the width w between a pair of microphones M _x and M ′ _x (x = 3 in the illustrated example) closest to the sound source S is determined in advance, and the width w is outside the microphones M _x and M ′ _x. Microphones M _{n and} M ′ _n (in the illustrated example, n =
For 0,1,2), in which to calculate the distance r _n from the center point o of the array to the installation position of the microphone respectively. In this case, the distance d'of the nearest microphone M _x and the sound source S to the sound source S, (5) because it is calculated by the equation, the microphones M ₀ ~M _x is concentric spheres C ₀ with a radius smaller by lambda ~ if it is arranged in order on the peripheral surface of the C _x, the distance between the other microphone M _n and the sound source S d
_n will be expressed as d _n = d ′ + (x−n) λ.
Thus the distance r _n from the center point o of the array to the microphone M _n is calculated by the equation (6).

[0072]

(Equation 5)

[0073]

(Equation 6)

FIG. 8 shows an example in which seven microphones M are arranged based on the method of FIG. Here, the microphone array by these microphones M is used as the sound source S
And that sampling is performed at a sampling frequency of 44.1 kHz. Under these conditions, the installation positions of the microphones were calculated so that the total length of the microphone array was about 40 cm. As a result, the position of the center point o of the array and 8.81 cm, 15.3 to the left and right of the center point o.
The symmetrical positions separated by 8 cm and 20.01 cm are determined as the installation positions of the microphone M, respectively.

FIG. 9 shows an example in which eight microphones M are arranged side by side according to the method of FIG. In this example,
The distance between the virtual position of the sound source S and the microphone array is 5
As a result of calculating the installation positions of the microphones so that the distance between the pair of microphones closest to the sound source S is 12 cm, with 0 cm and the sampling frequency being 44.1 kHz,
6.0 cm, 10.69 cm, left and right from the center o of the array,
Symmetrical positions separated by 16.56 cm and 20.96 cm are determined as microphone installation positions.

In FIGS. 8 and 9, the distance λ of the sound within one sampling period obtained from the sampling frequency is applied to the equations (4) and (6), and the theoretical installation position of each microphone M is determined. Is calculated. Further, assuming that the frequency band of the sound signal is in the range of 300 to 4000 Hz, the sampling frequency (600 to 8000 Hz) necessary for sampling the signal of this frequency band based on the sampling theorem is 8.5 to 8.5 Hz. In order to sample a signal having a wavelength of 113 cm, it is necessary to place the microphones at intervals of about 4.25 to 66.7 cm. Therefore, in each of the above embodiments, an installation position that satisfies the above conditions is selected from the calculated theoretical installation positions of the microphones, and each microphone is installed at a position that is close to the selected installation position. ing.

However, strictly speaking, this installation position is located at a position shifted by a predetermined distance from the peripheral surface of the concentric sphere shown in FIG. However, if the displacement is within an error range to be described later, or by adjusting the sampling frequency, it is possible to extract a sound signal with sufficient accuracy for recognition processing.

(3) Signal Processing by Software In the embodiments of FIGS.₀~
M_ThreeOutput signal from hardware
But not limited to this, each microphone M ₀~ M_Three
Input the output signal from the
It is also possible to perform processing by an embedded program.

FIG. 10 shows a configuration of a sound signal processing apparatus for processing output signals from the microphone array 1 having the arrangement shown in FIG. 5 by software. The sound signal processing device 10 includes a 4-channel amplifier 2A and an A / D
In addition to the converter 3A, a signal processing unit 8 using a microcomputer is provided. A pair of microphones M _n of the installation based on the peripheral surface of the same sphere C _{n, M'n,} after being superimposed by respectively mixing circuit 31 a to 31 d, are simultaneously input to the A / D converter 3A via the amplifier 2A . The A / D converter 3A sequentially samples each superimposed signal at a predetermined sampling period, and
As a result, every time sampling is performed, the signal processing unit 8 outputs, for each sphere C _n , 2 corresponding to the sphere C _n.
Four sampling values obtained by integrating the output signals of the microphones are input.

By using this method, the number of signals to be processed can be reduced before digital processing even when a large number of microphones are assigned to each sphere. Therefore A /
Efficient signal processing can be performed by reducing the number of signals to be D-converted and the amount of calculation.

FIG. 11 shows an example in which a rectangular frame 9 of a predetermined size is virtually set on a microphone installation surface (not shown), and 20 microphones are installed along the frame. In the illustrated example, it is assumed that the sound source S is located at a position of the center point O of the rectangular frame 9 on a straight line ls orthogonal to the installation surface at a distance of 50 cm. In this illustrated example, each microphone is indicated by a common symbol M,
The respective installation positions are indicated by numerals 1 to 20.

The rectangular frame 9 in the illustrated example has a width of 40 cm and a length of 30 cm.
With respect to the virtual position of the sound source S, each frame in the width direction (horizontal direction in the figure) is at a distance of 52.20 cm, and each frame in the length direction (vertical direction in the figure) is 53 At a distance of .85 cm,
Each is set to be located. And sound source S
From among the concentric spheres whose radius varies by the distance λ with the virtual position as the center, four spheres that intersect the rectangular frame are selected,
Twenty microphones are installed corresponding to the intersections between the selected spheres and the rectangular frame. Also in this case,
It is not necessary to make the microphone correspond to every intersection position, and the microphone may be set within a predetermined error range from the theoretical intersection position.

In the illustrated example, by setting the sampling frequency for each microphone to 44.1 KHZ, 8.04 cm, 15.17 cm, 2
A symmetrical position separated by 0 cm and a symmetrical position 7.42 cm, 11.81 cm, and 15 cm away from the center of the frame in the length direction are set as microphone installation positions, respectively, and specified by the set position and the rectangular frame. The microphones M are respectively installed at the 20 positions to be set. According to the above setting, in the illustrated example, when the microphones M are grouped based on the spheres serving as the installation reference, 1,
A set of microphones installed at each position of 6,15,20, a set of microphones installed at each position of 7,8,13,14, and a set of microphones 2,5,9,10,11,12,16, 1
9 and a set of microphones installed at each position
Four groups including the microphones set at the positions 17 and 18 are set.

FIG. 12 shows a configuration of the sound signal processing apparatus 10 using the microphones M installed as shown in FIG. This device 10 has the same configuration as that shown in FIG. 10, and the output signal of each microphone M is superimposed on each of four groups set by the mixing circuits 31A to 31D according to the sphere serving as the installation reference. After that, the signal is input to the A / D converter 3A via the amplifier 2A and is sampled at a predetermined sampling cycle. Therefore, for each sampling, four sampling values are generated and input to the signal processing unit 8.

In conventional signal processing of this type, if 20 microphones M are installed, it is necessary to perform convolution operation using a delay filter for each microphone M, so that an enormous amount of operation needs to be performed. Was. In contrast,
In this embodiment, since it is only necessary to calculate the sum of the four signals for each sampling period, the amount of calculation is greatly reduced. Moreover, since 20 microphones can collect sound signals at a plurality of positions surrounding the sound source S, noise generated at spatial positions other than the sound source S is accurately removed, and the sound signal from the sound source S is strengthened. Can be extracted.

FIG. 13 shows a specific configuration of the sound signal processing device 10 shown in FIGS. 10 and 12. The sound signal processing device 10 is a personal computer, and a CPU 11 serving as a control main body is connected to respective memories such as a ROM 12 and a RAM 13 and a hard disk device 14.
Is configured. Further, a CD-R
In addition to input / output units such as an OM drive 16, a keyboard 17, and a monitor 18, the four-channel amplifiers 2A and A /
Sound card 19 equipped with D converter 3A etc.
Is connected.

In the device 10 having the above-described configuration, a control program for executing the algorithm shown in FIG. 14, which will be described later, is incorporated in the hard disk device 14 in advance, and the CPU 11 executes the control program to execute the algorithm shown in FIG. The functions of the eleven signal processing units 8 are realized. The control program in the hard disk device 14 is stored by setting a CD-ROM on which the program is recorded in the CD-ROM drive 16 and performing a predetermined installation operation. However, the present invention is not limited to this, and a program recorded on a recording medium other than the CD-ROM or a program downloaded through communication with an external database can be stored in the hard disk device 14.

Further, in the RAM 13 or the hard disk device 14, a storage area for storing a sampling value of an audio signal taken from a sound card is set, and a sampling value integrated for each microphone M having the same installation standard is stored. , And up to a predetermined number in the sampling order. Regarding the storage of this sampling value,
Similar to the signal processing by the hardware configuration of FIGS. 3 and 4, the number of held data may be varied according to the spheres which are the microphone installation reference. However, in order to simplify the processing, It is also assumed that a fixed amount of sampling value considering the maximum amount of data required for processing is stored also for the input path of.

Here, for the sake of simplicity, it is assumed that N microphones are set in correspondence with the peripheral surfaces of different spheres. At this time, a predetermined number of samples a _N−1 is placed between the sphere C ₀ corresponding to the microphone M ₀ installed on the outermost side and the sphere C _N−1 corresponding to the microphone M _{N− 1} installed on the innermost side. Assuming that there is an interval of minutes, at least a _N-1 times of sampling is performed for each microphone, and the past a _N-1 times of the innermost microphone M _N-1 including the current sampling value are sampled. It is necessary to store the sampling value.

Therefore, in the embodiment shown in FIG.
For each of the microphones M _n (0 ≦ n ≦ N−1), a data sequence D _n based on a _N−1 sampling values per hour is stored in the RAM 13 (hereinafter, this data sequence D _n).
Is referred to as “sampling sequence D _n ”). Also, for each microphone _Mn , the number of samplings a corresponding to the distance a _n λ between the peripheral surface of the sphere C _n , which is the installation reference, and the peripheral surface of the sphere C ₀ corresponding to the microphone M ₀ furthest from the sound source S _n (an = 0 when _n = 0, an> 0 when _n > 0) is stored in the RAM 13 in association with each microphone _Mn .

FIG. 14 shows a procedure of signal processing by the sound signal processing device 10 set as described above. Note that ST in the figure means a processing step. The CPU 11
After resetting the counter t of the number of times of sampling to zero at T1, the processes of ST2 to ST9 are performed while sequentially incrementing the counter t until it reaches aN _-1 . Thus, each time a new audio signal is sampled from each microphone, output data O (t) is generated.

The CPU 11 first outputs the output data O in ST2.
In (t), the counter n for specifying the microphone of interest in ST3 is reset to zero, and then in ST4 and ST5, the reading position p for the sampling sequence D _n for the _n-th microphone M _n is set to t. set -a _n, reads the sampled value D _n (p) at that position. As a result, the sampling value at the present time is extracted for the outermost microphone M ₀ , and the sampling value obtained a _n times earlier is extracted for the microphone M _{n at} an arbitrary position inside the microphone M _0. become.

In the next ST6, the CPU 11 adds the read sampling value D _n (p) to the output data O (t). Thereafter, the processing of ST4 to ST6 is sequentially performed on N microphones while incrementing n in ST7, whereby the following equation (7) is executed, and the output data O in which the sound signal from the sound source S is strengthened is obtained.
(T) is obtained.

[0094]

(Equation 7)

The output data O (t) thus obtained is
In ST9, the volume is divided by the number of microphones N for volume adjustment, averaged, and externally output or stored in a memory as a digital sound signal for the t-th sampling process. In addition, as in the embodiments of FIGS.
When a plurality of microphones are installed in association with the peripheral surface of one sphere, the loop of ST4 to ST8 is not performed N times depending on the number of installed microphones, but is performed based on the number of spheres used as the microphone installation reference. It will be executed a corresponding number of times.

According to the algorithm shown in FIG.
Since the addition process is performed three times each time the loop of 4 to 8 is performed, and the addition process is performed twice after the loop is exited,
The amount of calculation within one sampling period is only 3N +
Two times. Therefore, assuming N = 20, 44.1K
When processing is performed at the HZ sampling cycle, the amount of calculation per second is about 273.42 million, which is a unit that can be sufficiently processed by a general-purpose CPU. Further, according to the above-described embodiments of FIGS. 10 and 12, since N = 4, the amount of calculation per second is only 617,400.

By the way, the speed at which the sound travels in the space is
It is known to fluctuate with temperature. Generally, between the sound speed v and the temperature TP, v = 331.5 + 0.60
Since there are relationship 7tp, temperature when increased by Δtp from TP _0, variation rate RT _v of sound for the sound velocity v ₀ when the temperature TP ₀ is expressed by equation (8) below.

[0098]

(Equation 8)

In each of the above embodiments, since each microphone is installed at a fixed position, if the sound speed v changes due to a change in temperature, the positional relationship of concentric spheres, which are the installation reference of each microphone, is not correct. As a result, the sound signals from the sound source S cannot be collected with the same phase. In order to solve this problem, it is necessary to appropriately detect the temperature in the installation space of the microphone, and to change the sampling cycle according to a change in the detected value. Specifically, if the change also RT _v times the sampling frequency f in response to the sound velocity v is RT _v times, the distance traveled by the sound within one sampling period λ (λ = v / f) is always constant , And a sound signal from the sound source S can be stably acquired.

FIG. 15 shows a schematic configuration of a sound signal processing device 10 having the above-described function of correcting the sampling period. In the figure, a signal processing unit 8 includes a frequency setting unit 20, a sound velocity calculation unit 21, a sampling data storage unit 23, a data processing unit 2
4. Each part of the data output unit 25 is set. These functions are realized by the control unit 15 of FIG. 13 similarly to the embodiments of FIGS. 10 and 12, and the sampling data storage unit 23 is set in the RAM 13 or the hard disk device 14. Other functions are realized by the control program installed in the hard disk device 14 and the CPU 11. Further, as a hardware configuration, a temperature sensor 26 for measuring an external temperature is installed in addition to the sound card 19.

FIG. 16 shows a processing procedure in the apparatus shown in FIG. Here, each procedure is indicated by “st” to distinguish it from the procedure of FIG. The sound speed calculation unit 21 reads the temperature TP detected by the temperature sensor 26 at a predetermined timing and calculates the sound speed v (st1, 2).
Then the frequency setting unit 20, the sound speed v calculated value and a predetermined time before the calculated value v ₀ Metropolitan from sound velocity change rate RT _v of prompts, variation sets the sampling frequency f of the sound signal according to the calculated value (St3).

Next, at st4, the A / D converter on the sound card 19 samples the audio signal at the sampling frequency set as above.
The sampling data storage unit 23 stores a _N−1 sampling values for each microphone. This gives
At t5, a series of data processing is performed by the data processing unit 24 to generate an output signal O (t), which is output from the data output unit 25 (st6). Note that the processing of st5 and st6 is in accordance with the above-described procedure of ST1 to ST9 in FIG. 14, and a detailed description of the procedure will be omitted.

Similarly, every time sampling is performed, the sound source S is obtained from the sampling train D _n for each microphone.
A sampling value in which the phases of the sound signals from are extracted is extracted, and an output signal O (t) based on a digital signal obtained by averaging the extracted values is output. When the sampling of aN _-1 times is completed, ST7 becomes "YES", and the process ends.

According to the above-described embodiment, each time the sampling of a _N−1 times is completed, the change of the temperature TP is checked and the sampling frequency f is set to fluctuate. However, the present invention is not limited to this. The sampling frequency f may be reset every arbitrary time corresponding to a predetermined multiple of _aN-1 times.

(4) Specific Example of Speech Recognition Processing Apparatus FIG. 17 shows the appearance of the apparatus when the present invention is applied to a personal computer 30 (hereinafter abbreviated as “PC 30”) in which the function of speech recognition processing is incorporated. 18 shows a circuit configuration showing the connection state between each microphone M of this device and the PC 30. Note that the PC 30 includes the same control unit 15 and peripheral devices as those in FIG. 13 described above. Therefore, in FIG. 18, the detailed internal configuration is omitted, and in FIGS. Are denoted by the same reference numerals.

The PC 30 incorporates various applications such as a word processor and a spreadsheet, in addition to the sound signal processing and voice recognition functions. The control unit receives a voice input from each of the microphone arrays 1A to 1D and recognizes the input content when a data input is required under execution of these applications, and performs a process according to the input data. Do. The monitor 18 displays input data based on the recognized voice, a processing result based on the input data, and the like.

In this embodiment, the display screen 18 of the monitor 18
a four microphone arrays 1A, 1B, 1
C and 1D are arranged and these microphone arrays 1A to
The sound signal captured by each microphone M on the 1D is processed to generate input data. The sound source S to be input to the microphone arrays 1A to 1D
Is virtually set at a position ahead of the center point of the monitor 18 by a predetermined distance.

In each of the microphone arrays 1A to 1D, four microphones M are provided around two spheres having a radius of iλ (i is a natural number corresponding to a predetermined number of samplings) and a center of the sound source S. In the position corresponding to the surface,
Will be installed. In the figure, a microphone M with an alphabetic symbol P (a microphone installed at an outer position on each substrate) corresponds to a virtual sphere having a larger diameter, and a microphone M with a Q symbol (an inner position on each substrate). Microphone) corresponds to the virtual sphere having the smaller diameter. The output signal of each microphone is superimposed for each group of P and Q by a mixing circuit 31 arranged in the preceding stage of the sound card 19, and then input to the sound card 19, and is sampled at a constant sampling period. .

In the case of this embodiment, sampling processing is performed after superimposing sound signals of eight microphones M installed along an arc generated by the intersection of the virtual sphere and the front surface of the monitor 18 in each of the groups P and Q. Therefore, considerable noise from each direction is removed at the stage of the signal superimposition process. The control unit in the PC 30 adds the current sampling value of the P group and the sampling value of the i group of the Q group before by the same algorithm as in FIG. The sound signal is emphasized and extracted.

Therefore, when the user's mouth is aligned with the center of the display screen 18a of the monitor 18 and data is input by voice, the sound signal generated by the voice is taken into each microphone M of each group P and Q. After that, the above-described signal processing is performed to extract a sound signal representing a sound from the sound source S from which noise has been removed. Further, the control unit recognizes the content of the input data by executing a predetermined voice recognition process while storing the time-series data based on the final sound signal in a RAM or the like, and executes a process according to the content. Will do. When data input by voice is accepted, a pointer indicating the ideal position of the user's mouth may be displayed at a position corresponding to the position of the virtual sound source S on the display screen 18a.

(5) Specific Example of Speech Recognition Processing Apparatus FIG. 19 shows an external view of the apparatus when the present invention is applied to a ticket vending machine 35 installed in a station, and FIG.
Shown respectively. The ticket vending machine 35 of this embodiment has an operation surface 36 on the front surface of which a plurality of operation keys 36a related to a ticket issuing process are arranged, an information display unit 37 such as a liquid crystal panel, a currency insertion unit 38, a ticket issuing unit 39, a change. It has a known configuration provided with a return port 40 and the like. Further, microphone arrays 1E and 1F for voice input are provided at the upper end and the lower end of the operation surface 36, respectively.

The upper and lower microphone arrays 1
E and 1F correspond to sound sources virtually set at two positions having different heights, and the intersection between the operation surface 26 and the peripheral surfaces of three spheres having different radii around the virtual position of the sound source. A total of six microphones M are provided, two for each ball, based on the position. In FIG. 20,
As in FIG. 18, the microphones M having the same sphere as the installation reference are grouped by the symbols P, Q, R, P ', Q', and R '.

Inside the machine, each microphone array 1
Amplifier 2B, 2B, A / D converter 3 for each E, 1F
B, the voice processing device 41 and the central control device 42 are incorporated. Further, the central control device 42 includes, as an input / output unit, not only the operation keys 36a and the information display unit 37, but also the
Ticketing processing unit 43 for issuing a ticket designated from 9
Also, a deposit and withdrawal processing unit 44 for processing deposits and changes is connected.

A three-channel amplifier is used for the amplifiers 2B and 2B, and a six-channel amplifier is used for the A / D converter 3B. Each microphone M
Is supplied to a mixing circuit 31E, 3 in the preceding stage of the amplifier.
After being superimposed on each set at 1F, the signals are input to the A / D converter 3B via the amplifiers 2B and 2B. The A / D converter 3B samples the input six sound signals sequentially and sequentially, and outputs the sampled values to the sound processing device 41.
Output to The sound processing device 41 processes these sampling values for each of the upper and lower microphone arrays 1E and 1F in the same procedure as in FIG. 14 to extract a sound signal for recognition processing.

The central processing unit 42 individually processes the time-series data of the sound signals obtained for each of the microphone arrays 1E and 1F and recognizes the contents of the input data. At this time, when the voice instructing the ticket issuing process is recognized, the central processing unit 42
Controls the operations of the ticketing processing unit 43 and the deposit / withdrawal processing unit 44, and issues a ticket according to the instruction content. In this case,
The accuracy with which the voices instructing the ticket issuing process by the microphone arrays 1E and 1F can be extracted varies depending on the height of the user. Therefore, for example, each microphone array 1
The final recognition is performed by integrating the voice recognition results for each of the microphones E and 1F, and the microphone array closer to the mouth position of the user is selected from the initial voice recognition results for each of the microphone arrays 1E and 1F. Thereafter, a method such as performing recognition processing using a sound signal from the selected microphone array is adopted.

Although the operation surface 36 in FIG. 19 is an inclined surface whose lower end protrudes forward, the surface on which the microphone M is to be installed is not limited to such a surface, but may be a flat vertical surface or a flat surface. It may be a curved surface. Further, as in the embodiment of FIG. 17, a plurality of microphones M may be installed along the periphery of the operation surface.

(6) Processing for Stabilizing Signal Processing In each of the above-described embodiments, the sound advances within one sampling period around the virtual position of the sound source S based on the principle shown in FIG. Of the concentric spheres whose radius varies by the distance λ,
It is assumed that microphones are installed on the peripheral surface of a predetermined number of selected spheres. However, in actual device design, it may be difficult to install the microphone as theoretically due to the size of the installation surface.

Therefore, the applicant has determined how much an allowable value of the deviation of the actual microphone installation position from the ideal microphone installation position shown in FIG. 1 is considered to have little effect on the speech recognition processing. I tried to calculate.

First, in order to simplify the explanation, the frequency 1H
Z, a sine wave w ₁ having an amplitude of 1 and a phase θ more than this sine wave w ₁
(0 ≦ θ <1) Sine wave w ₂ delayed is added. Here, the sine wave w ₁ is expressed as w ₁ (t) = sin (2π
t), a waveform w obtained by adding the sine waves w ₁ and w _2
a is represented by the following equation (9).

[0120]

(Equation 9)

[0121] Therefore, when calculating the power P (theta) per one period of the waveform w _a, the following (10)
It looks like an expression.

[0122]

(Equation 10)

Therefore, when θ = 0, that is, both sine waves w
_1, when there is no phase difference between w _2, P (θ) is maximum, when the maximum value P _max, the P _max = 2. Here, the reduction of the power P (θ) by the equation (10) is expressed by P
Assuming that a range satisfying (θ) ≧ 0.8P _max is allowed, the phase difference θ becomes maximum when P (θ) = 0.8P _max . That is, from 1 + cos (2πθ) = 1.6, a phase shift up to θ = 0.15 is recognized.

Next, the above conditions are applied to the sound source S
Sphere C set in relation to FIG. _nReach the circumference of
Sound signal and this sphere C_nMicro installed based on
Hong M _nTo the relationship with the sound signal arriving at
View. Temporarily microphone M_nIs the ball C_nLocated on the circumference of
If the sound signal from the sound source S is_nLap of
Surface and microphone M_nArrive at the same time as
Oh, ball C_nPeripheral surface and microphone M_nThe sound signal
The signals are in phase, and the signals obtained by adding these signals
The power value of the signal is P_maxBecomes In contrast, the microphone
Lohon M_nIs the ball C_nPosition shifted inward or outward from the circumference of
Microphone M_nThe sphere
C_nSignal out of phase with respect to the sound signal
No. will arrive. However, this phase shift
If the quantity e is up to 0.15, the microphone M_nTo
Ball C_nEven if it is regarded as being on the circumference of
That is, there is no difference.

Here, the sound speed is v and the sampling frequency is f
Then, the distance λ that the sound travels during one sampling period at this sampling frequency f is λ = v / f. Further, assuming that the upper limit value of the frequency of the sound signal representing a human voice is f _max , the distance λ _{max at} which the sound of this frequency f _max advances in one cycle is:
_λmax = v / _fmax . Further assuming that sets the sampling frequency f to the minimum value required to sample the sound frequency f _max, the f = 2f _max by the sampling theorem.

[0126] When the microphone under these conditions it is assumed that offset by αλ from the circumferential surface of the sphere C _n is the installation position of the ideal, when sampling sound signal of the frequency f _max, synchronized with the sound signal The phase difference e of the sound signal at the position shifted from the position by αλ is as shown in the following equation (11).

[0127]

[Equation 11]

Accordingly, from α / 2 <0.15, the deviation of the microphone installation position from the ideal position is 0.3λ.
Within this range, when the ideal sampling value and the actual sampling value of the sound signal are added, a power value of 0.8 P _max or more as the allowable range can be obtained. When performing the summation operation, even if an actual sampling value is used instead of the ideal sampling value, an error generated in the calculation result can be kept within an allowable range.

FIG. 21 shows that the sampling frequency f = 2f
_This shows the allowable range of the installation position of the microphone when it is _max . According to the above-described principle, each direction in and out from the peripheral surface of an arbitrary sphere C _n having a radius of (r ₀ −nλ) (where r ₀ is the radius of a reference sphere C ₀ corresponding to the microphone farthest from the sound source) The position at which the microphone _Mn is located at a distance of 0.3λ (the boundary position indicated by the broken line in the figure) is the installation allowable range of the microphone _Mn . In other words, by using the sampling value for the microphone _Mn set within the allowable range with a delay of n samples for use in the equation (7), a sound signal with sufficient accuracy for the recognition processing can be obtained.

[0130] Also further if the sampling frequency f was set larger than 2f _max, principles may be applied in the same manner as mentioned above. In this case, if the sampling frequency f is set such that the amount of positional deviation of the microphone from the ideal installation position can be allowed up to 0.5λ, the microphone installed at an arbitrary position can be located on the peripheral surface of the nearest sphere. And it can be processed.

That is, when the sampling frequency f is β * f
_max , the condition for keeping the phase difference e within 0.15 is the same as in the equation (11), e = 0.
From 5λ / λ _max = 0.5 / β ≦ 0.15, β ≧ 3.3
It becomes 3.

Therefore, as shown in FIG. 22, two spheres C _n and C _n-1 whose radii are different from each other by λ around the sound source S are
Assuming that the microphone M is installed at a predetermined position between the peripheral surfaces of the respective spheres, the peripheral surface of the sphere closer to the microphone M (in the illustrated example, the peripheral surface of the sphere C _n-1 ) is mounted on the microphone M. Will be considered a position. In this case, the sound signal from the microphone M, by sampling at a frequency of 3.33 * f _max, the phase difference with respect to a sound signal when sampled at the peripheral surface of a sphere C _n-1, with the allowable value It can be kept within a certain 0.15.

The maximum frequency f _max of human voice is about 10 KH
Since at Z, by setting the sampling frequency f of about 34 kHz, the microphone in the arbitrary position, of concentric spheres centered at the sound source S, it can be associated to the closest sphere C _n circumferential surface. That is, the sampling value of this microphone is delayed by n sampling times corresponding to the radius r ₀ −nλ of the sphere C _n and used for the calculation of the above equation (7), whereby the sound with sufficient accuracy for the recognition processing is obtained. A signal can be obtained. There are A / D converters currently available on the market that can set the sampling frequency up to 48 KHz. Therefore, this A / D converter is combined with a general-purpose microcomputer, and each microphone is associated with the peripheral surface closest to the microphone among the respective peripheral surfaces of the virtual concentric sphere based on the sampling frequency of the A / D converter. This makes it possible to perform signal processing with sufficient accuracy for generating a digital voice signal for recognition processing from human voice.

The allowable value of the power value is 0.8 P
Not only _max but an arbitrary value can be set, and according to the set value, an allowable value αλ of the microphone shift amount,
The allowable value of the sampling frequency for associating the microphone with the peripheral surface of one of the spheres also varies.

According to the above principle, each microphone M
Is not set correctly on the circumference of a concentric sphere whose radius changes by λ around the sound source S, sampling is performed by setting a sufficient sampling frequency, and each microphone M is set on the closest circumference. That is, it is only necessary to process each sampled value assuming that the sampled value is satisfied. In particular, in the case of performing signal processing by a computer as shown in FIG. 13, prior to the processing, an input of a virtual position of the sound source S and an installation position of each microphone is received, and an ideal installation position of each microphone is automatically calculated. , The delay time of each sampling value can be determined and processed.

For each of a plurality of sampling frequencies, the position data of the peripheral surface of each virtual sphere based on the sampling frequency is tabulated and stored in the memory, and this table is referred to by the installation position of each microphone. If the correspondence is determined by using the above method, the process of setting the delay time of the sampling value for each microphone can be simplified and speeded up.

Also, even if each microphone is installed in accordance with the principle of FIG. 1 with respect to the virtual position of the sound source S, the speaker's mouth is placed at a position deviated from this virtual position, and utterance is performed. In this case, each microphone is placed at a position shifted from the peripheral surface of the concentric sphere centered on the true sound source position. However, after estimating the true sound source position from the phase and intensity of the sound signal obtained by each microphone,
If signal processing is performed by associating each microphone with the peripheral surface of the virtual sphere centered on the true sound source position estimated using the above method, the accuracy of sound signal extraction can be stabilized. .

That is, each microphone is associated with each peripheral surface of a concentric sphere whose radius varies by λ around the estimated sound source position, and the sampling value of each microphone is determined according to the sphere diameter of the newly associated sphere. By executing the equation (7) with a delay corresponding to the number of times of sampling, it is possible to cope with a change in the position of the sound source. Of course, to map each microphone to the nearest sphere circumference,
As described above, it is necessary to set a sampling frequency of 34 KHZ or more. Therefore, if the initial sampling frequency is lower than 34 KHZ, the sampling frequency is set to 3 KHz.
It is necessary to perform the above-mentioned associating process after changing to 4 KHZ or more.

[0139]

According to the first and sixth aspects of the present invention, the diameters of the respective spheres are sequentially changed by a sound traveling distance corresponding to a fixed sampling period around a sound source virtually set at a predetermined spatial position. A peripheral surface of a concentric sphere is virtually set, and a plurality of microphones are installed based on the peripheral surfaces of a plurality of spheres selected from the concentric spheres. Therefore, by sequentially adding the sampling values for each microphone by sequentially delaying the number of samplings according to the difference in the ball diameter between the sphere that is the reference for setting the microphone and the sphere that is the reference for setting the microphone farthest from the sound source. Since the sound signal from the sound source can be strengthened and extracted, it is possible to provide an input sound signal that can be processed with a significantly reduced amount of calculation.

According to the second aspect of the present invention, each microphone is
By installing the microphone array at a position where the straight line and the peripheral surface of each sphere intersect, it is possible to provide a microphone array that can easily perform signal processing. According to the third aspect of the present invention, the microphone is installed at a position where the installation surface of the microphone and the peripheral surface of each sphere intersect, so that the microphone has a predetermined shape, for example, on the front or upper surface of a device having a voice recognition function. And can efficiently process sound signals to be recognized.

According to the fourth and tenth aspects of the present invention, when a plurality of microphones are installed according to the method of the first aspect, the sampling value for each microphone is determined in accordance with the sphere diameter of a sphere which is a reference for installing the microphone. By adding the signals while delaying them by the number of sampling times, a sound signal from the sound source can be extracted at high speed and with high accuracy. In addition, when the arithmetic processing is softwareized, the arithmetic processing can be performed by a general-purpose CPU, and when the arithmetic processing is performed by hardware, the configuration is simplified, so that the apparatus cost is reduced, and various types of equipment are reduced. Easy to incorporate.

According to the seventh aspect of the present invention, the signal processing speed in the sound signal input device for inputting a digital sound signal is increased, and the structure of the device is simplified to generate a high-precision input signal. can do. According to the invention of claim 8, a sound signal input device for inputting a sound signal of an analog amount is similarly produced, so that the sound signal input device is connected to a conventional device having an analog input terminal, such as a personal computer or a cassette recorder. A sound signal input device that can be used immediately and has high versatility can be provided.

According to the ninth aspect of the present invention, a device for recording a sound signal in a digital or analog system can be easily created by introducing the same configuration as in the seventh and eighth aspects, so that the performance is improved with a simple configuration. An audio recording device can be provided.

According to the twelfth aspect of the present invention, a sound signal input function is provided by a plurality of microphones installed by the method of the first aspect and a control device for processing output signals of these microphones by the method of the fourth aspect. It is possible to provide a sound signal processing device provided with a simplified signal processing and configuration.

According to the thirteenth aspect of the present invention, in the above-described sound signal processing apparatus, the microphones can be arranged in a line on a flat or curved installation surface to accept a user's voice input. According to the fourteenth aspect of the present invention, since the microphones are two-dimensionally or three-dimensionally mounted on the same installation surface, noise from each direction can be accurately removed. Further, according to the invention of these claims, by setting the installation surface on the front surface or the upper surface of the control device, it is possible to provide a practical signal processing device with an integrated voice input unit. Further, when the installation surface is made independent of the control device, an effect is obtained in which the sampling position of the sound signal can be freely set according to the position of the sound source by the external audio input unit.

According to the fifteenth aspect of the present invention, a plurality of microphones are arranged on the basis of the peripheral surface of one sphere. By delaying the output signal of a microphone having the same sphere as the installation reference by the same number of times of sampling, it is possible to accurately extract a target sound signal. If the sampling process is performed after superimposing the output signals of the microphones having the same sphere as the installation reference, the number of samplings and the amount of calculation can be significantly reduced.

According to the seventeenth aspect, the user's voice is simply and accurately extracted from the sound signal captured by the microphone installed at the periphery of the display screen of the monitor.
According to the eighteenth aspect, a user's voice is similarly extracted from a sound signal captured by a microphone installed on one side of the machine. Further, since the contents of the data included in the extracted voice are recognized and the predetermined processing is performed, a voice input function is introduced into a personal computer, a vending machine, and other general-purpose electronic devices, thereby achieving high precision. Processing can be executed.

According to the nineteenth aspect of the present invention, the program recorded on the recording medium is introduced into a computer having A / D conversion means to process the output signal of each microphone installed according to the method of the first aspect. Thus, it is possible to extract a sound signal from a predetermined spatial position, so that the sound signal processing function according to the present invention can be easily introduced into various computers.

According to the fifth, eleventh and twentieth aspects of the present invention, even when the microphone is installed at a position deviated from the ideal installation position, the effect of the position shift on the sound signal captured by the microphone is within a predetermined error. Since the sampling frequency is adjusted so as to fall within the range, the effect of the microphone displacement on the final calculation result can be suppressed to an allowable range, and the accuracy of sound signal extraction can be stabilized.

Further, according to the sixteenth aspect, even when the actual position of the sound source is different from the virtually set position, the error caused by the misalignment of the positional relationship between the sound source and the microphone is allowed by adjusting the sampling frequency. Since the voice can be kept within the range, when voice is input, each voice can be stably extracted even if the position of the mouth differs depending on the user.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram showing the principle of the present invention.

FIG. 2 is an explanatory diagram showing a configuration of a microphone array to which the microphone installation method of FIG. 1 is applied.

FIG. 3 is a block diagram showing a configuration of a voice recorder using the microphone array of FIG. 2;

FIG. 4 is a block diagram showing a configuration of a voice input device using the microphone array of FIG. 2;

FIG. 5 is an explanatory diagram showing a specific configuration of a microphone array.

FIG. 6 is an explanatory diagram showing a method for determining an installation position of each microphone.

FIG. 7 is an explanatory diagram showing a method for determining an installation position of each microphone.

FIG. 8 is an explanatory diagram showing a specific installation example of a microphone according to the method of FIG. 6;

FIG. 9 is an explanatory diagram showing a specific installation example of a microphone according to the method of FIG. 7;

10 is a block diagram illustrating a configuration of a sound signal processing device that processes output signals of the microphone array according to the configuration of FIG. 5;

FIG. 11 is an explanatory diagram showing a specific example in which microphones are two-dimensionally installed.

12 is a block diagram illustrating a configuration of a sound signal processing device that processes an output signal of the microphone array according to the configuration of FIG. 11;

FIG. 13 is a block diagram illustrating a configuration of a computer that realizes the sound signal processing device of FIGS. 10 and 12.

FIG. 14 is a flowchart showing a procedure of sound signal processing by the control unit in FIG. 13;

FIG. 15 is a block diagram illustrating an example in which a function of changing a sampling frequency according to a temperature change is incorporated in the sound signal processing device.

16 is a flowchart showing a processing procedure in the sound signal processing device of FIG.

FIG. 17 is a perspective view illustrating an external appearance of a voice recognition device using a personal computer.

FIG. 18 is a block diagram showing an electrical configuration of the device of FIG.

FIG. 19 is a perspective view showing the appearance of a ticket vending machine having a voice recognition function.

FIG. 20 is a block diagram showing an electric configuration of the ticket vending machine shown in FIG. 19;

FIG. 21 is an explanatory diagram showing an allowable range of an installation position of a microphone.

FIG. 22 is an explanatory diagram showing a specific example of associating a microphone with a peripheral surface of a sphere.

FIG. 23 is a block diagram showing a configuration of a conventional sound signal processing device.

FIG. 24 is a flowchart showing a procedure when performing conventional sound signal processing by software.

[Explanation of symbols]

M _n microphone C _n (virtual) sphere 1 microphone array 3 A / D converter 4A, 4B, 4C FIFO memory 5 adder 6 storage device 7 D / A converter 10 sound signal processing device 11 CPU 15 control unit 18 monitor 30 personal Computer 35 Ticket vending machine 36 Operation surface 41 Voice processing device 42 Central control device

Continuation of the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat II (reference) H04R 27/00 (54) [Title of the invention] Microphone installation method, sound signal processing method, and sound using these methods Signal input device, recording device, processing device, voice recognition processing device, recording medium on which program for sound signal processing is recorded

Claims (20)

[Claims] 1. A method of installing a plurality of microphones for collecting sounds from a sound source at a plurality of places, wherein each of the spheres has a diameter corresponding to a fixed sampling period with respect to the sound source. A microphone installation method, comprising: setting virtually concentric spheres that change sequentially, and then installing each microphone with reference to positioning of a plurality of spheres selected from the concentric spheres. 2. A method of installing a plurality of microphones for collecting sounds from a sound source at a plurality of places, wherein each of the spheres has a diameter corresponding to a fixed sampling period with respect to the sound source. A concentric sphere that sequentially changes is virtually set, and a straight line that intersects a plurality of selected spheres of the concentric sphere is virtually set at a spatial position separated by a predetermined distance from the sound source. rear,
A microphone installation method, wherein each microphone is installed using an intersection of the peripheral surface of each of the selected spheres and the straight line as a reference for positioning. 3. A method for installing microphones for collecting sounds from a sound source at a plurality of places, wherein each of the sphere diameters is sequentially changed by a sound traveling distance corresponding to a fixed sampling period around the sound source. A virtual concentric sphere was virtually set, and a microphone installation surface intersecting a plurality of spheres selected from the concentric spheres was virtually set at a spatial position separated by a predetermined distance from the sound source. Thereafter, a microphone is installed using the intersection of the selected peripheral surface of each ball and the installation surface as a reference for positioning. 4. A method for capturing and processing audio output signals from a plurality of microphones installed toward a sound source in a predetermined spatial position, wherein each of the sphere diameters around the sound source has a constant sampling period. After virtually setting concentric spheres that sequentially change by the traveling distance of the sound corresponding to, each microphone corresponds to the peripheral surface of a plurality of spheres selected from the concentric spheres based on their installation positions. In addition, while sampling the output signal of each microphone at the sampling period, the sampling value of each microphone is calculated as the difference between the sphere diameter of the sphere corresponding to the microphone and the sphere corresponding to the microphone farthest from the sound source. A sound signal processing method in which the signals are sequentially delayed by the number of samplings and added. 5. The method according to claim 4, wherein each microphone is associated with a sound signal from the sound source when the microphone is associated with a peripheral surface of a predetermined sphere. The sampling period is set so that the phase shift of the sound signal occurring between the case where the sound signal is captured on the peripheral surface of the attached sphere is within a predetermined error. A sound signal processing method that performs sampling and association with concentric spheres. 6. A sound signal input device comprising a plurality of microphones for collecting sound from a sound source virtually set at a predetermined spatial position at a plurality of locations, wherein each microphone determines a virtual position of the sound source. Of the concentric spheres virtually set so that the diameter of each sphere changes sequentially by the sound traveling distance corresponding to a fixed sampling period as the center, the peripheral surface of a plurality of selected spheres is set as a reference for positioning. Sound signal input device. 7. A plurality of microphones for collecting sound from a sound source virtually set at a predetermined spatial position at a plurality of locations, and a signal processing device for generating and outputting an input signal from an output signal of each microphone. Wherein each microphone is virtually set so that each sphere diameter is sequentially changed by a traveling distance of a sound corresponding to a fixed sampling period around the virtual position of the sound source. Among the concentric spheres, the peripheral surfaces of a plurality of selected spheres are installed as a reference for positioning, the signal processing device includes a sampling unit that samples an output signal of each microphone at the sampling period, and an output for each microphone. The sampling value of the signal is set to the sphere that is the reference of the microphone and the microphone located farthest from the virtual sound source. A memory for storing the number of times of sampling based on the difference between the ball diameter and the installation reference ball, and at the time of each sampling, reading out the oldest sampling value for each microphone from the memory, these sampling values, Calculating means for calculating the sum of the current sampling value for the microphone located farthest from the virtual sound source, and output means for outputting a digital signal indicating the sum of the sampling values obtained by the calculating means to the outside A sound signal input device comprising: 8. A plurality of microphones for collecting sound from a sound source virtually set at a predetermined spatial position at a plurality of locations, and signal processing for generating and outputting an input sound signal from an output signal of each microphone. A microphone, wherein each microphone is virtually set so that each sphere diameter is sequentially changed by a sound traveling distance corresponding to a fixed sampling period around the virtual position of the sound source. Among the concentric spheres, the peripheral surfaces of a plurality of selected spheres are installed as a reference for positioning, the signal processing device comprises: sampling means for sampling an output signal of each microphone at the sampling period; The sampling value of the output signal is used as the reference sphere for the microphone and the microphone farthest from the virtual sound source. A memory for storing the number of times of sampling based on the difference between the ball diameter and the sphere serving as the installation reference, and at the time of each sampling, the oldest sampled value for each microphone is read from the memory, and these sampled values and A calculating means for calculating the sum of the sampling value at the present time for the microphone located farthest from the virtual sound source, and an output means for converting the sum of the sampling values obtained by the calculating means into an analog signal and outputting the converted signal to the outside. A sound signal input device comprising: 9. A plurality of microphones for collecting sound from a sound source virtually set at a predetermined spatial position at a plurality of locations, and a signal processing device for taking in an output signal of each microphone and generating a sound signal for recording. And a storage processing device for storing the recording sound signal in a memory or a predetermined recording medium, wherein each microphone has a sphere diameter around a virtual position of the sound source. Of the concentric spheres virtually set so as to sequentially change by the traveling distance of the sound corresponding to a fixed sampling period, the peripheral surfaces of a plurality of selected spheres are installed as a reference for positioning, and the signal processing device Is a sampling means for sampling the output signal of each microphone at the sampling period, and a sampling value of the output signal for each microphone. A memory for storing the number of times of sampling based on the difference between the sphere as the microphone installation reference sphere and the sphere as the microphone installation reference farthest position from the virtual sound source, at the time of each sampling, Computing means for reading out the oldest sampled value for each microphone from the memory and calculating the sum of these sampled values and the current sampled value for the microphone furthest from the virtual sound source; The sound signal recording device writes the sound signal according to the sum of the sampling values obtained by the arithmetic means in the memory or the recording medium as the sound signal for recording. 10. A device for capturing and processing output signals of a plurality of microphones for collecting sounds from a sound source at a plurality of locations, wherein sampling means for sampling the output signals of each microphone at a constant sampling period; A memory for sequentially storing a predetermined number of sampling values obtained by the sampling means for each microphone; and a sphere diameter corresponding to the sounding distance corresponding to the sampling period, with each sphere diameter sequentially centered on the virtual position of the sound source. Correlation means that virtually sets a concentric sphere that changes, and associates each microphone with the peripheral surface of a predetermined sphere among the concentric spheres based on its installation position, at the time of each sampling, For each microphone, the ball corresponding to that microphone and the microphone furthest from the virtual position of the sound source A signal processing unit that reads out a sampling value delayed by the number of times of sampling based on the difference between the ball diameter and the ball corresponding to the crophone from the memory and calculates the sum of the read out sampling values. apparatus. 11. The signal processing apparatus according to claim 10, wherein a sound signal from the sound source is taken into each microphone when the microphone is associated with a peripheral surface of a predetermined sphere. Sampling period setting means for setting a sampling period such that a phase shift of a sound signal that occurs between the case where the sound signal is captured and the case where the sound signal is captured on the peripheral surface of the sphere is within a predetermined error. A signal processing device comprising a control means for controlling the means and the associating means to operate in accordance with the set sampling period. 12. A sound signal processing apparatus comprising: a plurality of microphones for collecting sound from a sound source virtually set at a predetermined spatial position at a plurality of locations; and a control device for receiving and processing output signals of the microphones. An apparatus, wherein each microphone is selected from among concentric spheres that are virtually set so that their respective sphere diameters sequentially change by a traveling distance of a sound corresponding to a fixed sampling period around the virtual position of the sound source. The control device is provided with sampling means for sampling an output signal of each microphone at the sampling period, and for each microphone, the control means is provided by the sampling means. A memory for sequentially storing a predetermined number of sampling values, and a microphone for each microphone at each sampling time. A sampling value delayed by the number of times of sampling based on the difference between the sphere as the microphone installation reference sphere and the sphere as the microphone installation reference farthest from the virtual position of the sound source is read from the memory, and read. A sound signal processing device comprising: a calculating means for calculating a sum of the respective sampled values output. 13. The sound signal processing device according to claim 12, further comprising: a microphone installation surface that is disposed opposite to a virtual position of the sound source by a predetermined distance, and each microphone is provided with the concentric sphere. A sound signal processing device installed on the installation surface using an intersection of a peripheral surface of the selected sphere and a predetermined straight line on the installation surface as a reference for positioning. 14. The sound signal processing device according to claim 12, further comprising: a microphone installation surface that is disposed opposite to a virtual position of the sound source by a predetermined distance, and wherein each microphone has the concentric sphere. As a reference for positioning the intersection position between the peripheral surface of the selected sphere and the installation surface,
A sound signal processing device installed on the installation surface. 15. The sound signal processing device according to claim 12, wherein a plurality of peripheral surfaces of a selected one of the concentric spheres are used as a reference for positioning. A sound signal processing device equipped with a microphone. 16. The sound signal processing device according to claim 12, wherein: a sound source position estimating unit that estimates an actual position of the sound source using an output signal of each microphone; Is estimated to be at a position different from the virtual position, concentrically set so that the respective sphere diameters are sequentially changed by the sound traveling distance corresponding to a fixed sampling period around the estimated position of the sound source. For a sphere, for each of the microphones,
When the microphone is associated with the peripheral surface of a predetermined sphere based on the installation position, when the sound signal from the estimated position of the sound source is captured by the microphone and when it is captured on the peripheral surface of the associated sphere A sampling cycle setting means for setting a sampling cycle such that a phase shift of a sound signal occurring between the sampling signal and the sound signal falls within a predetermined error, and processing the sound signal of each microphone according to the set sampling cycle. Control means for controlling the sampling means, and changing the reading position of the sampling value for each microphone by the arithmetic means in accordance with the correspondence of each microphone to a concentric sphere centered on the estimated position of the sound source. Sound signal processing device. 17. A control device for inputting a sound signal to perform a predetermined voice recognition process, a monitor for displaying a result of the voice recognition process by the control device, and a virtual position set in a spatial position in front of the monitor. A speech recognition processing device comprising: a plurality of microphones for collecting sounds from a sound source and inputting the sounds to the control device, wherein each microphone is provided on the periphery of a display screen of the monitor, and Of the concentric spheres virtually set so that each sphere diameter is sequentially changed by the traveling distance of the sound corresponding to a certain sampling period with the center as the center, the peripheral surface of the plurality of spheres and the peripheral portion are selected. The control device is provided with a sampling unit for sampling an output signal of each microphone at the sampling period, and a microphone. A memory for sequentially storing a predetermined number of sampling values obtained by the sampling means, and at each sampling point, a sphere serving as an installation reference of the microphone and a virtual position of the sound source. Calculating means for reading from the memory sampling values delayed by the number of times of sampling based on the difference between the ball diameter and the ball that is the installation reference of the farthest microphone, and calculating the sum of the read sampling values; A speech recognition processing device comprising: recognition processing means for processing time-series data based on the calculated values obtained by the means and executing predetermined speech recognition processing. 18. A plurality of microphones for capturing a sound signal to be recognized is installed on one surface of a machine, and a control device for performing a predetermined recognition process using an output signal of the microphone is incorporated in the machine. Wherein each sphere has a constant sampling period around a sound source virtually set at a predetermined space position facing the installation surface on the installation surface of the microphone. Of the concentric spheres that are virtually set so as to sequentially change by the travel distance of the corresponding sound, a crossing position between the peripheral surface of the selected plurality of spheres and the installation surface is installed as a reference for positioning, The control device includes a sampling unit that samples an output signal of each microphone at the sampling period, and the sampler for each microphone. A memory for sequentially storing a predetermined number of sampling values obtained by the sampling means, and at each sampling time, for each microphone, a sphere as a reference for setting the microphone and a microphone farthest from the virtual position of the sound source. An arithmetic unit that reads a sampling value delayed by the number of times of sampling based on the difference in ball diameter from the ball serving as an installation reference from the memory, and calculates the sum of the read sampled values; A speech recognition processing device comprising: a recognition processing unit that processes time-series data based on a calculation value obtained by the arithmetic unit in the meantime and executes a predetermined speech recognition process. 19. A recording medium storing a program for processing by a computer the output signals of a plurality of microphones for collecting sounds from a sound source at a plurality of places, wherein the A / B is built in the computer. A procedure for sampling the output signal of each microphone at a constant sampling cycle using the D conversion means, a procedure for sequentially storing a predetermined number of sampling values obtained by the sampling means in a memory for each microphone, Concentric spheres whose respective sphere diameters sequentially change by the traveling distance of the sound corresponding to the sampling period with the virtual position as the center are virtually set, and each microphone is placed on the basis of the installation position of the concentric sphere. Procedure for associating multiple selected spheres with the circumference, at each sampling point For each microphone, a sampling value delayed by the number of times of sampling corresponding to the difference in ball diameter between the sphere corresponding to the microphone and the sphere corresponding to the microphone farthest from the virtual position of the sound source is read from the memory, and read. A recording medium for recording a program for processing a sound signal, the program recording a program for executing each of the steps of calculating the sum of each of the output sampling values. 20. The recording medium according to claim 19, wherein a sound signal from the sound source is taken into the microphone when the microphone is associated with a peripheral surface of a predetermined sphere. A procedure for setting a sampling period such that a phase shift of a sound signal generated between the case where the sound signal is captured on the peripheral surface of the sphere corresponding to the microphone falls within a predetermined error, and a step of sampling the sound signal of each microphone. And a procedure for controlling each microphone to correspond to a concentric sphere, and a procedure for controlling so as to be performed at the set sampling period. Recording media.