JP2000341658A - Speaker direction detecting system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a speaker direction detecting system for miniaturizing the voice inputting part of a television conference device, and for improving stability by reducing the detection error of a speaker direction even when a signal arriving from a direction other than the direction of a speaker is superimposed on a speaker voice signal. SOLUTION: An echo canceller 8 estimates the transfer function of arrival delay from the output signals of two microphones 1 and 2 arranged in a horizontal direction. A maximum tap detecting circuit 10 retrieves a tap having the maximum value from an H register for holding the estimated transmission function, and supplies the tap number to a tap number/speaker direction converting circuit 12. In the same way, an echo cancel 9 estimates the transfer function of the arrival delay from the output signals of two microphones 2 and 3 arranged in a vertical direction. A maximum tap detecting circuit 11 retrieves the tap having the maximum value from the H register for holding the estimated transfer function, and supplies the tap number to the tap number/ speaker direction converting circuit 12. The tap number/speaker direction converting circuit 12 converts the two maximum tap numbers into a horizontal angle and a vertical angle for obtaining a speaker directional signal.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speaker direction detecting system, and more particularly to a speaker direction detecting system for pointing a video camera for image input to a microphone for voice input of a speaker, such as a television conference apparatus. is there.

[0002]

2. Description of the Related Art Conventionally, as a technique for receiving a voice signal of a subject and detecting the direction of the voice signal in order to control the imaging angle of a camera of an imaging apparatus, there is a technique disclosed in Japanese Patent Application Laid-Open No. Hei 7-123331. That is, a unidirectional microphone and a bidirectional microphone are used as audio input sources,
The output signals of the two microphones are synchronized, the phase difference is calculated through the sensitivity adjustment circuit of the microphones, and the direction of the subject is detected.

[0003]

In order to reduce the size of the television conference apparatus, it is conceivable to house a microphone inside the apparatus. In this case, the unidirectionality disclosed in the above-mentioned publication is disclosed. To achieve the ideal unidirectional pattern characteristics of a microphone, it is necessary to make a large hole for sound collection in the device housing around the microphone. It is disadvantageous to realize.

In a situation where a television conference apparatus is used, the number of speakers is not always one, but rather a plurality of persons are often present around the apparatus. The audio signal input to the video conference device has most of the voice component of one speaker in a normal conversation, but there is a possibility that another speaker starts to interrupt and talk during the conversation. A non-stationary non-speech signal (noise signal) such as a sound of hitting a desk with a pencil or the like or turning a document to look through a distributed document may well be superimposed.

[0005] Therefore, when an attempt is made to control the imaging angle of a video camera using the output signal of the speaker direction detecting device,
If a speaker direction detection error occurs, a problem arises in that the camera is directed in a direction other than the speaker, and there is a problem that a great inconvenience occurs for the user of the television conference apparatus.

SUMMARY OF THE INVENTION It is an object of the present invention to reduce the size of a voice input unit of a television conference apparatus, and to reduce detection errors in a speaker direction even when a signal arriving from a direction other than the speaker is superimposed on a speaker voice signal. An object of the present invention is to provide a speaker direction detection system capable of improving stability.

[0007]

According to the present invention, a delay time estimating means for estimating a delay time between signals output from first and second microphones, and a speaker direction is calculated based on the estimated delay time. And a speaker direction calculating means.

The delay time estimating means has an echo canceller configuration in which a signal output from the first microphone is used as a main signal input, and a signal output from the second microphone is used as a reference signal input. Wherein the echo canceller comprises: an adaptive FIR filter that receives the reference signal as input; a subtractor that subtracts the output of the adaptive FIR filter from the main signal; Control means for controlling a learning operation of the adaptive FIR filter.

The main signal is delayed by a predetermined time and input to the echo canceller, and the control means controls the adaptive FIR filter so that the subtraction output becomes substantially zero. The tap coefficient is controlled to be updated, and the delay time estimating means has means for detecting a maximum value of the tap coefficient of the adaptive FIR filter, and the speaker direction calculating means determines the maximum value of the tap coefficient. The apparatus further comprises means for calculating the speaker direction according to a corresponding tap number.

Further, the invention is characterized in that it comprises means for calculating the amount of erasure of the echo canceller, and means for stopping updating of the calculation of the speaker direction calculating means when the calculated amount is less than a predetermined threshold value.

Furthermore, the first and second microphones are arranged at regular intervals in a horizontal (or vertical) direction,
The delay time estimating means and the speaker direction calculating means may calculate a horizontal (or vertical) direction angle of the speaker. In addition, the first microphone and the third microphone are arranged at regular intervals in the vertical (or horizontal) direction, and a delay time between signals output by these first and third microphones is estimated. It is characterized by further including second delay time estimating means and second speaker direction calculating means for calculating the vertical (or horizontal) angle of the speaker based on the estimated delay time.

A speaker direction detection system according to the present invention includes a part for estimating a arrival delay time existing between two audio signals obtained from two omnidirectional microphones arranged at a fixed interval from each other. An echo canceller is provided. The echo canceller performs an operation of estimating the transfer function of the arrival delay, and the estimated transfer function is held in the H register that holds the tap coefficient of the FIR filter constituting the echo canceller. A maximum tap detecting circuit for detecting a tap number having the maximum amplitude in the estimated transfer function is provided. This maximum tap detection circuit operates to consider the position of the maximum amplitude part of the transfer function as the arrival delay time. Therefore, the arrival delay time is obtained from the estimation result of the transfer function, converted into the speaker direction, and output.

[0013]

Embodiments of the present invention will be described in detail with reference to the accompanying drawings. FIG. 1 shows a speaker direction detecting circuit 13 according to an embodiment of the present invention. The speaker direction detection circuit 13 has omnidirectional microphones 1 to 3 for inputting audio signals. The outputs of these microphones 1 to 3 are supplied to A / D (analog-digital) converters 4 to 6 and converted into digital signals. The output signal of the A / D converter 5 is a delay circuit 7
And output with a time delay compared with the output signals of the A / D converters 4 and 6.

FIG. 2 shows an example of arrangement of these microphones 1-3.
Is shown in The microphones 2 and 1 are arranged at a fixed interval in the horizontal direction of the apparatus, and the two microphones 2 and 1 detect the horizontal angle θ in the speaker direction. The microphones 2 and 3 are arranged at a certain interval in the vertical direction of the apparatus, and the two microphones 2 and 3 detect a vertical angle φ in the speaker direction.

An echo canceller 8 is provided for obtaining a transfer function of a horizontal delay between both outputs of the microphones 1 and 2. Echo canceller 8 is A / D converter 4
Is input as a reference signal, and the output signal of the delay circuit 7 is input as a main signal. By performing the learning operation of the echo canceller 8, the transfer function of the arrival delay of the audio signals input to the microphones 1 and 2 is estimated, and the estimation result is obtained by the FIR (Finite) which constitutes the echo canceller 8.
Impulse Response) is stored in the H register inside the digital filter 17 (see FIG. 3).

Further, a maximum tap detecting circuit 10 is provided. The maximum tap detecting circuit 10 receives the content (tap coefficient) of the H register supplied from the echo canceller 8 as an input, and selects a tap having a maximum value from the input. And supplies the tap number to the tap number-speaker direction conversion circuit 12.

Similarly, an echo canceller 9 is provided for obtaining a transfer function of a vertical delay. The echo canceller 9 inputs the output signal of the A / D converter 6 as a reference signal, and inputs the output signal of the delay circuit 7 as a main signal. By performing the learning operation of the echo canceller 9, the transfer function of the arrival delay of the audio signal input to the microphones 3 and 2 is estimated, and the estimation result is stored in the H register inside the echo canceller 9.

The maximum tap detecting circuit 11 receives the content (tap coefficient) of the H register supplied from the echo canceller 9, searches for a tap having the maximum value from the input, and determines the tap number as a tap number-speaker direction. It is supplied to the conversion circuit 12. The tap number / speaker direction conversion circuit 12 converts the tap numbers supplied from the maximum tap detection circuits 10 and 11 into a horizontal angle θ and a vertical angle φ in the speaker direction with respect to the front direction of the device, respectively, and outputs them. .

FIG. 3 shows the echo canceller 8 shown in FIG.
9 is a diagram illustrating a configuration of FIG. Each of echo cancellers 8 and 9 includes leak integrators 14 and 15 and coefficient update control circuit 16
, An adaptive FIR filter 17, a subtractor 18, and an H register correction amount calculation circuit 19. Each of the echo cancellers 8 and 9 is supplied with a main signal y and a reference signal x.

The main signal y and the output signal y # of the FIR filter 17 are supplied to a subtracter 18, and the signal y # is converted from the signal y.
Is subtracted to generate a residual signal e. The H register correction amount calculation circuit 19 calculates the correction amount Δh of the tap coefficient of the adaptive FIR filter 17 so as to minimize the residual signal e and supplies the same to the adaptive FIR filter 17.

FIG. 4 is a diagram showing a configuration of the adaptive FIR filter 17. The FIR filter 17 has an X register 20, a multiplier 21, an H register 22, and an adder 23. The input reference signal x has one sample delays 1001 to 100N-1 (N
Is a positive integer representing the tap length). The reference signal data supplied to the one-sample delay 1001 is transferred to an adjacent one-sample delay every clock. That is, the X register is a tapped delay line.

The reference signal x is supplied to a multiplier 2000 and is multiplied by a tap coefficient supplied from the H register 22. Output signals of the one-sample delay units 1001 to 100N-1 are supplied to multipliers 2001 to 200N-1, respectively, and are multiplied by tap coefficients supplied from the H register 22. The output signals of the multipliers 2000 to 200N-1 are supplied to the adder 23. The addition circuit 23 includes multipliers 2000 to
Calculate the sum of the output signals received from 200N-1 and calculate the FIR
It is assumed that the output signal of the filter 17 is y ＾.

The leak integrator 14 is provided as a means for easily obtaining the power of the main signal y, and receives the main signal y as input and outputs the power Ly. Leak integrator 15
Is provided as a means for easily obtaining the power of the reference signal x, and receives the reference signal x and outputs its power Lx. These leak integrators have the configuration shown in FIG. 5, but are well known to those skilled in the art, and thus detailed description thereof will be omitted.

The coefficient update control circuit 16 is provided as means for generating a signal instructing permission or stop of coefficient update of the adaptive FIR filter 17. The operation of the coefficient updating circuit 16 will be described with reference to FIG.

The coefficient update control circuit 16 operates in one sampling cycle. The output signal Ly of the leak integrator 14 and the output signal L of the leak integrator 15 are supplied to the coefficient update control circuit 16.
x is supplied (step 601). Next, in order to detect the presence or absence of an audio signal based on the magnitude of power, Lx and threshold
The value of n is compared with the value of n (step 602). If Lx is equal to or smaller than the threshold Ln, an update stop signal is output (step 6).
05) End the operation. If Lx is greater than the threshold value Ln, the process proceeds to the next step 603. In step 603, the magnitudes of Ly and the threshold Ln are similarly compared.
If Ly is equal to or smaller than the threshold value Ln, an update stop signal is output (step 605), and the operation ends. Lx is the threshold L
If it is larger than n, a coefficient update permission signal is output (step 604), and the operation ends.

The echo canceller performs a convolution operation on a reference signal x stored in an X register and a tap coefficient stored in an H register in an adaptive FIR filter which is a constituent element of the echo canceller, thereby obtaining a signal y ＾. , And subtracting the signal y か ら from the main signal y to generate a residual signal e. Therefore, this residual signal e is

[0027]

(Equation 1) Becomes

At this time, if the adaptive FIR filter converges and the residual signal e can be regarded as zero, the above equation becomes:

[0029]

(Equation 2) In the echo canceller 8, the signal delay, which is the difference between the main signal y and the reference signal x, is represented by a transfer function h. That is, h is a transfer function that is the difference between the transfer function of the acoustic space from the sound source to the microphone 1 and the transfer function obtained by adding the delay of the delay circuit 7 to the transfer function of the acoustic space from the sound source to the microphone 2.

If the distance from the sound source to the microphone 1 and the distance from the sound source to the microphone 2 are equal, there is no difference between the transfer functions of the two, and the tap position having the maximum value in the transfer function h is the delay. It is uniquely determined by the delay by the circuit 7. If the distance from the sound source to the microphone 1 is not equal to the distance from the sound source to the microphone 2, the tap position having the maximum value of the transfer function h is proportional to the difference between the distances. Deviated from the position determined by Therefore, it is possible to specify the sound source direction by calculating backward from the displacement of the maximum tap position. The same applies to the case of the echo canceller 9.

Even when the magnitude of the reference signal and the main signal input to the echo canceller is different, the amplitude of the entire transfer function only increases or decreases in accordance with the ratio between the magnitude of the reference signal and the magnitude of the main signal. The position of the tap with is not affected. That is, the position of the tap having the maximum value is not affected by the magnitude of the sensitivity of the microphone that is the audio signal input means.

There are various types of algorithms for correcting the H register coefficient of the adaptive FIR filter, which is a component of the echo canceller. For example, when the learning identification method is used, the correction of the H register is performed according to the following equation. However, in the following equation, the subscript k is the tap number of the adaptive FIR filter, Δh is the correction amount of the H register, and μ is a coefficient that determines the speed and stability of adaptation.

[0033]

(Equation 3) The correction algorithm of the echo canceller is not limited to the learning identification method described above, but may be an LMS method (Least Mea
n Square) and other algorithms can of course be used.

The operation of the maximum tap detection circuit 10 will be described with reference to FIG. Where N is the adaptive FIR
A constant indicating the tap length of the filter, k is a variable indicating the tap number, and m is a variable for storing the maximum value. First, the maximum tap detection circuit 10 inputs an H register from the echo canceller circuit 8 (step 501). Next, each of the variables i, k, and m is initialized by substituting zero (step 502). Then, the variable i is compared with N (step 503). If i is smaller than N, the process proceeds to the next step 504. If i is equal to or larger than N, the tap number k
Is output (step 507), and the processing ends.

In step 504, the value of the i-th tap h (i) of the H register is compared with the variable m. If h (i) is greater than the variable m, the process proceeds to step 505, where h
If (i) is equal to or less than the variable m, the process proceeds to step 506. Step 505 is a process in the case where a tap having a value larger than the value of the already searched tap is found,
The variable m is substituted for h (i), and the variable k is substituted for the variable i. In step 506, 1 is added to the value of the variable i, and the process returns to step 503. With the above operation, the tap number having the maximum value can be detected from the H register. The operation of the maximum tap detection circuit 11 is the same. The tap having the maximum value is searched from the H register of the echo canceller circuit 9 and the tap number is output.

The operation of the tap number / speaker direction conversion circuit 12 will be described. First, referring to FIG.
There is a delay circuit 7 in the signal transmission path from the input to the main signal inputs of the echo cancellers 8 and 9. This delay circuit 7
It is provided for the purpose of satisfying that the echo cancellers 8 and 9 can reliably converge regardless of the direction in which the speaker is located with respect to the apparatus. This is because, as is generally known, in order for the echo canceller to converge, one of the conditions is that it does not violate causality, and that the main signal is input to the echo canceller after being delayed from the reference signal. It is a necessary condition.

Therefore, for example, the delay amount of the delay circuit 7 is 2
The distance between the two microphones should be equal to or longer than the time obtained by dividing the distance by the speed of sound in the air. That is, the transfer functions estimated by the echo cancellers 8 and 9 include the delay amount of the delay circuit 7.

A method for converting the speaker direction in the horizontal direction will be described with reference to FIG. Assuming that the distance between the microphones is L, the sound velocity in the air is v, the delay added by the delay circuit 7 is Dc, and the maximum tap number supplied from the maximum tap detection circuit 10 is kh,

[0039]

(Equation 4) Becomes Therefore, the angle θ with respect to the apparatus front direction is

[0040]

(Equation 5) Is required.

Similarly, a method for converting the speaker direction in the vertical direction will be described with reference to FIG. Assuming that the distance between the microphones is L, the sound velocity in the air is v, the delay added by the delay circuit 7 is Dc, and the maximum tap number supplied from the maximum tap detection circuit 11 is kv.

[0042]

(Equation 6) Becomes Therefore, the angle φ with respect to the front direction of the device is

[0043]

Equation 7 Is required.

Next, another embodiment of the present invention will be described.
The basic configuration of the other embodiment is the same as that of the first embodiment, except that the tap number / speaker direction conversion circuit 1
The stability of the horizontal angle signal and the vertical angle signal which are the output signals of No. 2 is further improved. 10 and 11 are diagrams showing another embodiment of the present invention, and the same parts as those in FIGS. 1 and 3 are denoted by the same reference numerals. As shown in FIG.
In addition to the configuration of the echo canceller 13 described above, erasure amount calculation circuits 30 and 31 and an update control circuit 32 are provided for the purpose of improving the stability of the horizontal and vertical angle output signals in the speaker direction.

Referring to FIG. 12, erase amount calculation circuit 3
The configuration of 0, 31 will be described. The erasure amount calculation circuit includes a division circuit 35, a logarithm calculation circuit 36, and a leak integration circuit 37, and performs an operation of calculating the erasure amount of the echo canceller. The output signal Ly of the leak integrator 14 and the output signal Le of the leak integrator 34 (see FIG. 11) are supplied to the elimination amount calculation circuit. The leak integrator 34 calculates the power Le of the residual signal e, which is the output of the echo canceller.

First, the signal Ly and the signal Le are divided by the division circuit 3
5, the division circuit 35 divides Le by Ly, finds the ratio of Le and Ly, and supplies the ratio to the logarithmic calculation circuit 36. The logarithmic calculation circuit 36 calculates the common logarithm of the ratio of Le and Ly,
It is supplied to the leak integrator 37. The leak integrator 37 smoothes the common logarithm and outputs it. The amount of erasure of the echo canceller obtained in this manner is stored in the update control circuit 32.
Supplied to

The update control circuit 32 is provided for determining whether or not the echo canceller has converged, and for verifying the estimated transfer function. The configuration and operation of the update control circuit 32 will be described with reference to FIG. The update control circuit 32 includes size comparators 38 and 39, switches 40 and 41, and memories 42 and 43. The magnitude comparator 38 is supplied with the amount of elimination of the echo canceller 8 that has estimated the transfer function in the horizontal direction, and the threshold L
Compare with ht and output the result to switch 40. The switch 40 is supplied with a horizontal angle signal from the tap number / speaker direction conversion circuit. If the erasure amount is larger than the threshold Lht, the switch is closed and the horizontal angle signal is supplied to the memory 42. If the erase amount is equal to or less than the threshold Lht, the switch is opened to stop supplying the horizontal angle signal. The memory 42 always outputs the held content as a horizontal angle signal.

Similarly, the magnitude comparator 39 is supplied with the amount of erasure of the echo canceller 9 that has estimated the transfer function in the vertical direction, compares it with the threshold value Lvt, and outputs the result to the switch 41. The switch 41 is supplied with a vertical angle signal from the tap number / speaker direction conversion circuit. When the erasure amount is larger than the threshold Lvt, the switch is closed and the vertical angle signal is supplied to the memory 43. If the erasure amount is equal to or less than the threshold Lvt,
Open the switch to stop supplying the vertical angle signal. The memory 43 always outputs the held content as a vertical angle signal.

With the operation described above, a horizontal angle output signal and a vertical angle output signal with improved stability can be obtained. As described above, in the present embodiment, it is possible to switch the updating or holding of the output of the speaker direction signal while monitoring the amount of elimination of the echo canceller, and a person in another direction can pencil the desk with the speaker voice. Even if the noise as if struck by the user is superimposed, or a speaker in another direction interrupts the conversation, and multiple speaker voices are added, the effect of retaining the speaker direction signal detected until immediately before is obtained. Can be Therefore, an effect of further improving the accuracy of the detected speaker direction is obtained, and the object of the present invention is achieved.

[0050]

As described above, according to the present invention,
The following effects can be obtained. First, it is possible to eliminate the need to adjust the sensitivity of the microphone as the audio signal input means. Secondly, even when a noise signal other than the speaker's voice signal is input while being superimposed on the voice signal, detection errors in the speaker direction can be reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention.

FIG. 2 shows the arrangement of microphones used in the present invention,
FIG. 4 is a diagram illustrating a relationship between a horizontal angle and a vertical angle of an arrival direction of a speaker voice signal with respect to a front direction of the apparatus.

FIG. 3 is a block diagram showing a configuration of an echo canceller used in one embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of an adaptive FIR filter used in one embodiment of the present invention.

FIG. 5 is a block diagram showing a configuration of a leak integrator used in one embodiment of the present invention.

FIG. 6 is a flowchart showing the operation of a coefficient update control circuit used in one embodiment of the present invention.

FIG. 7 is a flowchart illustrating an operation of a maximum tap detection circuit.

FIG. 8 illustrates tap numbers used in an embodiment of the present invention.
FIG. 6 is a diagram illustrating calculation for obtaining a horizontal angle in a speaker direction conversion circuit.

FIG. 9 illustrates tap numbers used in an embodiment of the present invention.
It is a figure explaining calculation which calculates a perpendicular angle in a speaker direction conversion circuit.

FIG. 10 is a block diagram of another embodiment of the present invention.

FIG. 11 is a diagram showing a configuration of an echo canceller used in another embodiment of the present invention.

FIG. 12 is a block diagram showing a configuration of an erasure amount calculation circuit used in another embodiment of the present invention.

FIG. 13 is a block diagram showing a configuration of an update control circuit used in another embodiment of the present invention.

[Explanation of symbols]

 1-3 Microphone 4-6 A / D converter 7 Delay circuit 8,9 Echo canceller 10,11 Maximum tap detection circuit 12 Tap number-speaker direction conversion circuit 13 Speaker direction detection circuit 14,15,34,37 Leakage Integrator 16 Coefficient update control circuit 17 Adaptive FIR filter circuit 18 Subtractor 19 H register correction amount calculation circuit 20 X register 21 Multiplication circuit 22 H register 23 Addition circuit 24, 27 Multiplier 25 Adder 26 One sample delay 30, 31 Erasure amount calculation circuit 32 Update control circuit 35 Division circuit 36 Logarithm calculation circuit 38,39 Large / Small comparator 40,41 Switch 42,43 Memory

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) H04R 3/00 320 H04N 5/232 Z 5J083 // H04N 5/232 G10L 3/00 511 5K046 F term (reference) 5C022 AA12 AB62 5C064 AA02 AC09 5D015 DD02 5D020 BB00 BB09 5J023 DA05 DB03 DC07 5J083 AA05 AC17 AC18 AD02 AE08 AF01 BE20 BE53 BE57 CA10 5K046 HH01 HH53 HH79

Claims (9)

[Claims] 1. A system comprising: a delay time estimating means for estimating a delay time between signals output by first and second microphones; and a speaker direction calculating means for calculating a speaker direction based on the estimated delay time. A speaker direction detection system characterized by the following. 2. The delay time estimating means has an echo canceller configuration in which a signal output from the first microphone is used as a main signal input, and a signal output from the second microphone is used as a reference signal input. The speaker direction detection system according to claim 1, wherein: 3. The echo canceller according to claim 1, wherein the adaptive FIR filter receives the reference signal, a subtractor subtracts the output of the adaptive FIR filter from the main signal, and the output of the subtractor becomes substantially zero. 3. A speaker direction detecting system according to claim 2, further comprising control means for controlling a learning operation of said adaptive FIR filter. 4. The speaker direction detection system according to claim 2, wherein the main signal is delayed by a predetermined time and input to the echo canceller. 5. The speaker direction detecting system according to claim 3, wherein said control means updates and controls tap coefficients of said adaptive FIR filter so that said subtraction output becomes substantially zero. . 6. The adaptive FIR according to claim 6, wherein
Means for detecting the maximum value of the tap coefficient of the filter,
6. The speaker direction detection system according to claim 3, wherein said speaker direction calculation means includes means for calculating said speaker direction according to a tap number corresponding to said maximum value. 7. The apparatus according to claim 1, further comprising: means for calculating an erasure amount of said echo canceller; and means for stopping updating of said speaker direction calculation means when said calculated amount is less than a predetermined threshold value. The speaker direction detection system according to claim 2. 8. The first and second microphones are arranged at regular intervals in the horizontal (or vertical) direction, and the delay time estimating means and the speaker direction calculating means determine the horizontal (or vertical) of the speaker. 8. The speaker direction detecting circuit according to claim 1, wherein the direction angle is calculated. 9. The first microphone and the third microphone are arranged at regular intervals in the vertical (or horizontal) direction, and a delay time between signals output by the first and third microphones is set. The apparatus further includes: second delay time estimating means for estimating; and second speaker direction calculating means for calculating an angle of the speaker in the vertical (or horizontal) direction based on the estimated delay time. A speaker direction detection system according to any one of Items 1 to 8.