JPH08182100A - Method and device for sound image localization - Google Patents

Abstract

PURPOSE: To obtain a stable sound image localization output by regarding a microphone output for one ear as an output of an unknown system, identifying the unknown system and calculating sound image localization of a sound signal divided into plural frequency bands through the use of an identification system and synthesizing outputs by bands. CONSTITUTION: Let a signal series of a left ear by yL1 , let a signal series of a right ear be yR1 , and let an original white noise series be (e), then let a spatial transfer function from a virtual sound source to the left ear microphone be TM and let a spatial transfer function from a virtual sound source to the right ear microphone be TC and let a space ratio transfer function FS be TC/TM, then the relation of yR1 =(TC/TM)×yL1 =FS.yL1 is obtained. Then a speaker is located to an actual speaker position and the same white noise (e) outputted from the speaker is used. In this case, let a signal series of a left ear be yL2 , let a signal series of a right ear be yR2 , let a spatial transfer function from a sound source to the left ear microphone be SM and let a spatial transfer function from a virtual sound source to the right ear microphone be SC then identification is made to the system of yR2 =(SC/SM)×yL2 =FS.yL2 , then a space ratio transfer function FC=SC/SM is obtained.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a sound image localization method and a sound image localization apparatus for locating a sound image at an arbitrary position.

[0002]

2. Description of the Related Art Conventionally, there have been two methods for localizing a sound image.
There was a general stereo method using speakers and headphones. In the general 2-channel stereo method, left and right 2
The sound image can be localized only between the two speakers, and cannot be localized except between the two speakers, that is, around the listener. In addition, in headphones, there is a phenomenon called sound localization in the head, in which a sound image is localized in the head. Therefore, the impression was far different from the actual sound field.

Therefore, by processing the normal sound source signal,
A method has been proposed in which a sound image is localized around the listener even with two speakers (hereinafter referred to as stereoscopic sound field processing), and out-of-head localization is realized with headphones (hereinafter referred to as out-of-head localization processing). First, before explaining the present invention, the basic principle of the stereoscopic sound field processing and the out-of-head localization processing will be described below.

First, the principle of three-dimensional sound field processing will be described with reference to FIG.
explain. The sound source signal u is emitted at the virtual sound source position O.
And it reaches the left and right ears EL and ER through the space.
Shall be. At this time, the acoustic signal from the left ear EL
y_L1, Y the acoustic signal at the right ear ER _R1And In this state
In addition, the transmission from the virtual sound source position O to the left ear EL and the right ear ER
The transfer function can be represented by a transfer function. This transfer function will be
We will call it the spatial transfer function.
Sometimes called a function, but strictly speaking
Since it may be something other than a number, we call it here.
To). The spatial transmission function from the virtual sound source position O to the left ear EL
Number T_L, T to the spatial transfer function to the right ear ER_RIs defined.
Here, the path from the virtual sound source position O to EL or ER is short.
The longer one is called the main path and the longer one is called the crosstalk path.
And then T_M, T_CAnd From now on
Limits the virtual sound source to the left of the listener.
I will proceed with the story. When the virtual sound source is on the right
However, the outline is the same.

The relationship between the acoustic signal at the sound source position and the acoustic signal at the ear is as follows: y _L1 = T _L · u = T _M · u (1-1) y _R1 = T _R · u = T _C · u ( It is represented by 1-2). Next, acoustic signals at the left speaker SL and the right speaker SR that are actually installed are respectively set as x _L and x _R. Here, it is assumed that the two speakers are placed symmetrically with respect to the listener. The sound radiated from these two speakers arrives as an acoustic signal of y _L2 at the listener's left ear EL and y _R2 at the right ear ER. At this time, the transfer path from SL to EL and SR to ER is the main path, and this spatial transfer function is S _M. Also, S
The transmission path from L to ER and SR to EL is a crosstalk path, and this spatial transfer function is S _C. The relationship at this time is y _L2 = S _M · x _L + S _C · x _R (2-1) y _R2 = S _C · x _L + S _M · x _R (2-2) Here, although the sound is actually emitted from the two speakers SL and SR, in order to make it heard from the point O, y _L1 = y _L2 (3-1) y _R1 = y _R2 (3-2 ). Therefore, S _M · x _L + S _C · x _R = T _L · u (4-1) S _C · x _L + S _M · x _R = T _R · u (4-2), and x _L and x _{R solving, x L = (1-F} C 2) -1 (F L -F C · F R) u (5-1) x R = (1-F C 2) -1 (F R -F C · _FL ) u (5-2). However, F _C = S _C / S _M , F _L = T _L / S _M , F _R = T _R / S _M (5-3). Furthermore _{rewrite, x L = (1-F} C 2) -1 (1-F C · F S) F M · u (6-1) x R = (1-F C 2) -1 (F S - It can be expressed as F _C ) F _M · u (6-2). However, F _S = T _C / T _M and F _M = T _M / S _M (6-3). That is, if the signal processing represented by the above equation is applied to the acoustic signal u, the two speakers SL and SR sound as if there is a sound image at the point O. Where filter F
_When the gain of _C ² is sufficiently smaller than 1, this part is omitted and x _L = (F _L −F _C · F _R ) u (7-1) = (1−F _C · F _S ). F _M · u (8-1) x _R = (F _R −F _C · F _L ) u (7-2) = (F _S −F _C ) F _M · u (8-2) .

A circuit representing this process is shown in FIG.
Is. Here, the input channel is expanded to a plurality. Here, the direction localization means is shown in FIG. 4 (a) or (b), and the crosstalk canceling means is shown in FIG. 4 (c).

The above is the principle of stereophonic sound field processing by two speakers, but if two speakers are replaced by headphones, it can be used for out-of-head localization processing. In this case, since it can be assumed that the gain of the crosstalk path through the headphones is sufficiently smaller than that of the main path, the term regarding F _C can be regarded as 0. Transfer function from headphone to ear is H
If _{M, F L = T L /} H M, F R = T R / H M, is replaced with _{F M = T M / H M} (9), it can be realized out-of-head localization. This circuit is shown in FIG.

The principle described so far has already been announced in the 1960s, and many documents (as a representative example, "Spatial sound" by Brauert, Morimoto, and Goto, published by Kashima Press) are listed. Has become a known fact.

[0009]

From the above description of the principles of the three-dimensional sound field processing and the out-of-head localization processing, it is possible to identify the spatial transfer functions (T _M , T _C , S _M , S _C , H _M ). It turns out that you have an important key. Various methods have been published for this identification, but most of them use a fast Fourier transform or the like to obtain a single spatial transfer function.

According to this principle, the basic units of three-dimensional sound field processing and out-of-head localization processing are F _M , F _S , and F _C , and convolution of one spatial transfer function with the inverse of another spatial transfer function. It can be regarded as a calculated transfer function. Hereinafter, such a transfer function that can be represented by a calculation form of two spatial transfer functions will be referred to as a spatial ratio transfer function for convenience.

Here, when the spatial transfer function is obtained by an ordinary FIR filter or IIR filter, in general, all zeros do not always fall within the unit circle. Rather, there are more zeros outside the unit circle. Therefore, when the inverse function of the spatial transfer function is taken, a pole outside the unit circle is generated, and the spatial ratio transfer function becomes unstable. As a countermeasure for this, it is necessary to perform a stabilizing process such as folding back the poles generated outside the unit circle to the unit circle and rearranging them.

Further, the spatial ratio transfer functions F _M , F _S , and F _C are
To cover a wide range of frequencies from bass to treble,
The filter order has to be quite large. However, as the order increases, the number of parameters increases and the poles are concentrated near the unit circle, causing stability problems such as easy oscillation. Therefore, there is a trade-off relationship between the effect of sound quality and sound image localization and stability.

The present invention has been made in view of the above problems, and achieves both a sound image localization method capable of stably, easily and highly accurately obtaining a spatial ratio transfer function, and an effect and stability of sound quality / sound image localization. It is an object of the present invention to provide a sound image localization device.

[0014]

As means for solving these problems, the present invention relates to a first spatial transfer function from a sound source position in an actual sound field to a specific position of a listener or the left ear of a pseudo-head, A second spatial transfer function to a specific location on the right ear,
It includes a calculation of a third spatial transfer function from the virtual sound source position to the specific location of the left ear and a fourth spatial transfer function to the specific location of the right ear, and the sound image is localized at an arbitrary position from the calculation result. In a sound image localization method for a sound image localization apparatus, a microphone is provided at the specific location of the left and right ears, an acoustic signal satisfying a continuous excitation condition is generated from a sound source position or a virtual sound source position, and a microphone output signal is recorded. To the fourth inverse spatial transfer function of any of the above and the remaining spatial transfer function of the unknown transfer function, the microphone output signal as the input and output of the unknown transfer function. It is characterized in that it is obtained by using the considered system identification means.

Further, according to the present invention, the output data is relatively delayed by the input / output of the unknown transfer function, and the present output is relatively performed, and the linear operation of the past output and the present and past / future inputs is performed. The unknown transfer function is obtained by using the model represented by.

Further, according to the present invention, the arrival time difference of the sound between the two ears is measured, and the data of the earlier arrival time is shifted by the obtained arrival time difference, so that among the unknown transfer functions, It is characterized in that the arrival time difference and the element other than the arrival time difference are obtained separately.

Further, according to the present invention, a microphone is provided at the specific places of the left and right ears, a first acoustic signal satisfying a continuous excitation condition is generated from a sound source position or a virtual sound source position, and a microphone output signal is recorded, The difference between the propagation time of the sound from the sound source position or the virtual sound source position to the specific location of the left ear and the propagation time of the sound to the specific location of the right ear is estimated, and the spatial transfer function of any one of the first to fourth Is estimated in the form of an autoregressive model, a convolution operation of the inverse function of the spatial transfer function is performed on a signal satisfying the continuous excitation condition, and the second sound output is again performed from the sound source position or the virtual sound source position. The microphone output signal at this time is recorded, and the unknown transfer function represented by the product of the inverse function of any of the first to fourth spatial transfer functions and the remaining spatial transfer function From Kurohon output signal, or including the first acoustic signal, and obtains using the system identification means regarded as input and output of the unknown transfer function.

Further, according to the present invention, a microphone is provided at the specific location of the left and right ears, an acoustic signal satisfying a continuous excitation condition is generated from a sound source position or a virtual sound source position, the microphone signal is recorded, and the sound source position or the virtual sound source position is recorded. The difference between the propagation time of the sound from the sound source position to the specific location of the left ear and the propagation time of the sound to the specific location of the right ear is estimated, and the first to fourth spatial transfer functions are formed in the form of an autoregressive model. Estimate,
An unknown transfer function represented by the product of the inverse function of any one of the first to fourth spatial transfer functions and any of the remaining spatial transfer functions can be expressed by division of two autoregressive models I
It is characterized in that it is obtained in the form of an IR filter.

Further, according to the present invention, a signal sequence satisfying a continuous excitation condition is input to the obtained transfer function, and the output sequence and the input signal sequence are identified again with a lower order than the transfer function, It is characterized in that the transfer function of order is obtained.

Further, the present invention is characterized in that a signal obtained by filtering a recorded microphone signal is replaced with the microphone signal and used as an input / output of the system identification means.

Further, according to the present invention, the recorded acoustic signal is divided into a plurality of frequency bands by the band dividing means, and the acoustic signal band-divided for each band is regarded as an input / output of an unknown transfer function, and the unknown It is characterized by obtaining a transfer function.

Further, according to the present invention, a band dividing means for dividing an acoustic signal into a plurality of frequency bands, a band synthesizing means for synthesizing the divided acoustic signals of respective frequency bands, and a parameter of a filter relating to each sound image localization position are set. And a sound image localization means for performing a filter calculation with an acoustic signal, an acoustic signal input section for one or more channels, and an acoustic signal output section for two channels, the acoustic signal input section being connected to the input section of the band dividing means. The output unit of the band dividing unit is connected to the input unit of the sound image localization unit corresponding to each frequency band, the output unit of the sound image localization unit is connected to the input unit of the corresponding band synthesis unit, the band An output section of the synthesizing means is connected to the acoustic signal output section.

Further, in the present invention, the sound image localization means is composed of a direction localization means and a crosstalk canceling means, the input part of the sound image localization means serves as an input part of the direction localization means, and the output part of the sound image localization means. Is the output section of the crosstalk canceling means, the direction locating means has two output sections, and all the first output sections of the direction locating means pass through the adders and the first of the crosstalk canceling means.
Is connected to the second input part of the crosstalk canceling means via an adder.

Further, in the present invention, the crosstalk canceling means is removed, and all the first output parts of the direction locating means become the first output parts of the sound image locating means via the adder,
All the second output parts of the direction localization means become the second output parts of the sound image localization means via an adder.

Further, the present invention is characterized in that a reflected sound generating means is provided between the output portion of the direction localization means and the input portion of the band synthesizing means.

Further, the present invention is characterized in that a reflected sound generating means is provided in parallel with the direction localization means.

Further, the present invention is characterized in that a reflected sound generating means is provided between the output section of the band synthesizing means and the acoustic signal output section.

Further, the present invention is characterized in that a reflected sound generating means is provided between the sound input section and the sound output section.

[0029]

Considering the system as shown in FIG. 5, two transfer functions having common inputs are A and B, respectively. Let the common input sequence at time t be in (t), and transfer function A
Let outa (t) be the output sequence via and outb (t) be the output sequence via the transfer function B: outa (t) = A × in (t) outb (t) = B × in (t) Since it can be expressed and can be transformed into in (t) = A ^-1 × outa (t), it can be expressed as outb (t) = (B / A) × outa (t). That is, outa (t) and outb can be obtained without individually obtaining the transfer functions A and B.
B / A can be directly obtained from the data of (t) only by the method of least squares or the like.

Therefore, since it is not necessary to perform the inverse calculation of the transfer function, the transfer function can be obtained stably and easily.

If the input in (t) is white noise in the above equation, the transfer function A can be identified by an autoregressive model (AR model). What is AR model?

[0032]

[Equation 1]

It is a model represented by Since the transfer function obtained in this manner has only poles and no zeros, its inverse function is only zeros, and thus a stable inverse function A ^-1 is obtained. White noise is input to the inverse function A ^-1 thus obtained, and its output is input to the transfer function B. The output of the transfer function B is outb (t) = B × {A ⁻¹ × in ( t)} = (B /
A) × in (t). Therefore, the transfer function (B / A) can be obtained by modeling with an autoregressive moving average model (ARMA model) or the like and identifying by the least square method or the like.

The transfer function B can also be identified in the form of an AR model, and the transfer function (B / A) can be obtained by combining it with the inverse function A ^-1 obtained as described above.

Further, in the filter used in the sound image localization means, if the order is the same, the frequency resolution is proportional to the sampling frequency. Although it depends on the order, if the order is about 30 with the IIR filter, empirically the sampling frequency is 1
The accuracy of frequency components of / 50 to 1/20 or less is not very good. Also, of course, due to the Nyquist principle, it is not possible to represent frequency components that are 1/2 or more of the sampling frequency. Therefore, a low frequency acoustic signal may use a low sampling frequency, and a high frequency acoustic signal may use a high sampling frequency.

Therefore, band division processing is performed in which sampling is first performed at a high frequency, a high frequency component is used to obtain a transfer function at the same sampling frequency, and low frequencies are subjected to decimation processing to reduce the sampling frequency to obtain a transfer function.

Similarly, for the sound image localization apparatus, a sound image localization means composed of a filter represented by the transfer function obtained above is prepared for each frequency band. The original acoustic signal is divided into bands, the sound image localization processing operation is performed for each frequency band, and then the sound image localization processing acoustic signals of each frequency band are combined by the band synthesizing means, resulting in an acoustic signal having a normal band. Output.

As described above, by dividing the frequency band, it is possible to perform a highly accurate sound image localization calculation with a low order. Because of the low order, the poles are separated from the vicinity of the unit circle and the stability is improved as compared with the case of the high order. Also, since the filter order, operation word length, and scaling for operation can be set for each band, the band sensitive to the human ear can be highly accurate, and the band that is relatively dull can be simplified to improve the sound quality and sound image. It is possible to achieve both localization accuracy and reduction of arithmetic processing. Further, in the band where the individual difference is small, the number of individual parameters of the filter can be reduced by using the standardized parameter of the filter of the sound image localization means.

[0039]

【Example】

 (Embodiment 1) Refer to FIG. 6 and FIG. 7 for the first embodiment of the present invention.
While explaining. First, the part that satisfies the desired reflected sound condition
Shop (If you don't need reflected sound
Virtual sound source position in a room with that condition if
Place a speaker in the room and generate white noise from the speaker
Let White noise is generated at all frequencies up to the Nyquist frequency.
It contains the wave number component, and the continuous excitation condition is satisfied.
It The sound propagates through the space and is left or right of the listener or the simulated head
Signal and source of microphone attached to the ear canal of
The signal is recorded on a data recorder (eg DAT). Left ear
The signal sequence at y _L1, The signal sequence at the right ear is y_R1, Original white
Let the color noise signal sequence be e. Listen to the virtual sound source position here
It is assumed that it is on the left side when viewed from the person. Left from virtual sound source
The spatial transfer function to the ear microphone is T_M, Right ear microphone
The spatial transfer function at_CAnd then y_L1= T_M・ E y_R1= T_C・ Because it becomes e, the e is eliminated from the above two equations, and the space ratio transmission is
Reaching function F_S= T_C/ T_MThen y_R1= (T_C/ T_M) × y_L1= F_S・ Y_L1 Can be expressed as_L1, Output series
y_R1, The least squares method, the maximum likelihood method, etc.
F using the system identification method_SCan be asked.
Although FIG. 6 shows the case of the method of obtaining sequentially,
Even if this is an offline batch processing method, the effect is the same.
is there.

Next, the speaker is placed at the actual speaker position, and the same operation as before is repeated. It is desirable for this person to measure in an anechoic room. The same white noise e output from the speaker is used. In this case, if the actual speakers are placed symmetrically, either one of the left and right speakers may be used, but here the left speaker is used. At this time, the signal sequence in the left ear is y _L2 , the signal sequence in the right ear is y _R2 , the spatial transfer function from the sound source to the left ear microphone is S _M , and the spatial transfer function from the right ear microphone is S _C Then, if the system is identified as y _R2 = (S _C / S _M ) × y _L2 = F _C · y _{L2, the} spatial ratio transfer function F _C =
S _C / S _M is required. That is, considering an unknown system, the unknown system may be identified by regarding the input sequence as y _L2 and the output sequence as y _R2 .

Also, using the first recorded y _L1 and the next recorded y _L2 , the spatial ratio transfer can be obtained by identifying with a system of y _L1 = (T _M / S _M ) × y _L2 = F _M · y _L2 Function F _M = T _M / S
_M is required. At this time, if the original signal sequence e is used, the two data y _L1 and y _L2 can be easily synchronized. By such a procedure, FIG. 4 (a)-(c)
All the filters of the three-dimensional sound field processing circuit shown in can be obtained.

According to this procedure, the inverse calculation of the transfer function is not necessary. Therefore, it is possible to directly obtain a stable spatial ratio transfer function. If a single spatial transfer function is identified as in the conventional case, the characteristics of the speaker, microphone, and ear that do not depend on the direction of sound arrival are added in addition to the characteristics of the direction of sound arrival, and the order is increased accordingly. Without identification, it is not possible to accurately extract the characteristics of the sound arrival direction. According to the method of the present invention, the characteristics of the ear that do not depend on the direction of arrival of the speaker, the microphone, and the sound are canceled from the beginning, and thus the identification can be performed at a low order.

Although the original signal e is recorded here, the same effect can be obtained even if the original signal e is not recorded if the original signal is known. This also applies to the following examples.

If the microphone signal is filtered and the data signal is used for identification, the frequency characteristic of the spatial ratio transfer function can be modified.

As a model of the spatial ratio transfer function,

[0046]

[Equation 2]

IIR filter type transfer function model represented by

[0048]

(Equation 3)

A transfer function model in the form of an FIR filter represented by the following may be used, but a model in which the present output can be represented by a linear combination of past output and past to future input can also be used. this is

[0050]

[Equation 4]

By shifting the time series,

[0052]

(Equation 5)

It can be expressed as Using this model,
This is a non-minimum phase system, and even if the impulse response has a relatively long rise time, it has the effect of enabling accurate identification and stability.

(Embodiment 2) A second embodiment of the present invention is shown in FIG.
A description will be given with reference to FIGS. 8 and 9. Similar to the first embodiment, the speaker is placed at the virtual sound source position. Output impulse signal or step signal from the speaker,
From the output of the microphones installed in the left and right ears,
The difference in sound arrival time between the ears is measured. FIG. 8 shows an example of an impulse signal.

Alternatively, the white noise is output from the speaker, the cross-correlation coefficient of the microphone output signals of both ears is obtained, and the maximum value point is found to find the arrival time difference between both ears. That is, when the signal sequence in the left ear at time t is y _L1 (t) and the signal sequence in the right ear is y _R1 (t), R (d) = E [y _L1 (t + d) · y _R1 (T)] Here, E [•] represents an average. Cross correlation coefficient R
(D) is obtained, and d having the maximum R (d) is regarded as the arrival time difference between both ears.

The time difference thus obtained is expressed as z- ^d using a delay operator in a discrete time system. A signal that satisfies the continuous excitation condition is output from the speaker, and the signal sequence for the left ear is y _L1 and the signal sequence for the right ear is y _R1 . Here, assuming that the route to the left ear is the main path, the arrival time of the sound to the right ear is later than that of the left ear, so the data in the left ear is shifted accordingly. That is, y _R1 = (T _C / T _M ) × z ^−d y _L1 , which means that the arrival time difference component and the timbre change component of the spatial ratio transfer function are obtained separately. Hereinafter, the same operation is repeated even in the case of the actual speaker position. The above procedure is shown in FIG.

The method of the present embodiment is capable of further reducing the number of parameters of the spatial ratio transfer function by obtaining the arrival time difference separately as compared with the first embodiment.

The spatial ratio transfer function can be obtained by regarding a value slightly smaller than the actually obtained d as the arrival time difference component.

(Embodiment 3) A third embodiment of the present invention is shown in FIG.
A description will be given with reference to 0. First, at the actual speaker position
Place the speaker only on the left side, and the white noise from that speaker
Generate. The sound propagates through space and is
Microphones attached to the left and right external ear canals
The signal and the original signal are recorded in the data recorder. Belief in the left ear
Y series _L3, The signal sequence at the right ear is y_R3And First,
In the same manner as in Example 2, y_L3And y_R3Spatial ratio transmission between
Find the time delay element of the function. Next, the signal sequence y_L3Than,
Space transfer function S_MThe Levinson-Durbin method, Berg
Method, least squares method, etc. to identify in the form of an AR model.
The AR model has the advantage that it can be identified quickly and stably.
Therefore, the inverse function is easy and cheap because it has only zero poles and no zeros.
You can always ask. Inverse of the obtained spatial transfer function
Function S_M ^-1Convolve white noise into the output signal sequence
p. Is the signal series p again the above-mentioned speaker?
Signal sequence y in the right ear at that time_R4Record
It This y_R4Levinson-Durbin method, least squares
Method in the form of a model such as ARMA model,
By adding the previously obtained time delay factor to the result, the spatial ratio
Transfer function F_C= S_C/ S_MIs required.

Next, a speaker is placed at the virtual sound source position and
Similarly, the arrival time difference between both ears is obtained. Next, the signal system described above
Output the column p and the signal sequence y at the left ear_L5, Signal system in the right ear
Row y _R5To record. This is also y_L5And y_R5The
The ARMA model with the Binson-Durbin method, the least squares method, etc.
Space ratio transfer function F_L= T_L/
S _M, F_R= T_R/ S_MThe tone change component of
In addition, a time delay element is added to the spatial ratio transfer function of the crosstalk path.
Just add it to the number. By doing this, in FIG.
It is possible to obtain the filter of the three-dimensional sound field processing circuit shown.
It

(Embodiment 4) A fourth embodiment of the present invention is shown in FIG.
This will be described with reference to 1. First, at the actual speaker position
Place the speaker only on the left side, and the white noise from that speaker
Generate. The sound propagates through space and is
Microphones attached to the left and right external ear canals
The signal and the original signal are recorded in the data recorder. Belief in the left ear
Y series _L3, The signal sequence at the right ear is y_R3And First,
In the same manner as in Example 2, y_L3And y_R3Spatial ratio transmission between
Find the time delay element of the function. Next, the signal sequence y_L3, Y_R3
Therefore, the spatial transfer function S_M, S_CLevinson Durbin
Method, Berg method, least square method, etc.
I do.

Next, a speaker is placed at the position of the virtual sound source, and similarly, the arrival time difference of the sound to the left and right ears from the signal sequence y _{L4 in} the left ear and the signal sequence y _{R4 in the} right ear, the AR model type spatial transmission. Find the functions T _L and T _R.

[0063] Then, the obtained AR modeling of spatial transfer function T _L, T _R, the S _C, is multiplied by the inverse function S _M ^-1 of the S _M, obtaining the space ratio transfer function of the IIR filter type ( (See FIG. 12). To this, add the time delay factor obtained previously.

This method has the advantage of being fast and simple because the identification is only for the AR model. However, there is a demerit that the order of the completed IIR filter becomes large.

In order to overcome the above-mentioned drawbacks, it is possible to reduce the order by obtaining the poles and zeros of the IIR filter thus found and removing those which are almost overlapped and canceled.

Further, a signal (for example, white noise) satisfying the condition of continuous excitation is input to this IIR filter, and the output sequence and the input sequence are used to perform identification again by the least-square method or the like, whereby the low-order IIR filter is identified. It is also possible to convert it to a filter.

(Embodiment 5) A fifth embodiment of the present invention is shown in FIG.
The description will be made with reference to FIGS. In the first embodiment, the band dividing means is provided, the acoustic signal from the data recorder is band-divided, and the spatial ratio transfer function is obtained in each frequency band.

First, in FIG. 1A, in the process of obtaining the spatial ratio transfer function F _S = T _C / T _M , the outputs y _L1 and y _R1 from the data recorder are input to the band dividing means, and a plurality of outputs are obtained. It is divided into frequency bands. Here, consider an example of division into four bands. Here, as a band dividing means, FIG.
Using 1), the coefficient of each filter is set so that the frequency characteristic shown in FIG. 13 is obtained. High pass filter 19a
_Has a characteristic of F _HH , the bandpass filter 20a has a characteristic of F _HL , 20b has a characteristic of F _LH , and the low pass filter has a characteristic of F _LL . These filters also have a role of preventing aliasing when lowering the sampling frequency by the subsequent decimation process. Then, the sampling frequency is reduced by the thinning means. For example, the decimation order is 2 in the thinning means 22a, the decimation order is 4 in the thinning means 22b, and the decimation order is 8 in the thinning means 22c.

FIG. 15 shows the outline of the data. After the original signal (a) is filtered to remove the out-of-band components and the waveform (b) is obtained, the decimation process ( In the figure, the decimation order 4) is performed, and data is taken once every four times. Therefore, the sampling frequency is reduced to 1/4.

The signal thus band-divided and the sampling frequency dropped is input to the system identification means prepared for each band, and the spatial ratio transfer function is obtained.

Other unknown spatial ratio transfer functions F _C and F _M can be obtained by the same procedure.

The high-pass filter may be omitted and the band-pass filter may be replaced with a low-pass filter. In addition, the decimation order by the thinning means is
Other values may be used as long as (degree of 22a) <(degree of 22b) <(degree of 22c).

As a band dividing means, FIG.
The same applies when 2) to (a5) are used. Where (a
When the thinning means is omitted as in 4) and (a5), it is necessary to use data for each d (original decimation order). In this case, since the number of data is not reduced by the thinning means, it is possible to identify even if the original number of data is small.

(Embodiment 6) A sixth embodiment of the present invention will be described with reference to FIGS. 1 and 13 to 17. FIG.
An example of a three-dimensional sound field processing device using the spatial ratio transfer function obtained in the fifth embodiment is shown in (b). This device has 1 input channel, 2 output channels, and 4 band divisions.
Is. The direction localization means is shown in FIG. 4 (a) or (b), and the crosstalk canceling means is shown in FIG. 4 (c). The number of band divisions is 4, and the band dividing means is shown in FIG.
1) and the band synthesizing means is shown in FIG. 14 (b1).

First, a digital signal such as a CD or A
The signal converted to digital by the / D converter is input to the input section and sent to the band dividing means. Due to the band division,
Low pass filter (LPF) in the lowest frequency band
However, in the highest frequency band, a high-pass filter (HP
F), but band pass filter (BPF) for other bands
Are connected in parallel and are divided into preset frequency bands. The operation of the band dividing means is the same as that of the fifth embodiment. The band-divided signal is input to the sound image localization means.

In the sound image localization means, first, the virtual sound image localization calculation is performed by the direction localization means. The output of the two channels is input to the crosstalk canceling means, the crosstalk component expected in the actual speaker arrangement is removed, and the output of the two channels is obtained. The signal is input to the band synthesizing means.

In the band synthesizing means, first, in order to increase the sampling frequency, the waveform diagram after sound image localization processing is shown in FIG.
Insert interpolation data into. Although zero is inserted in FIG. 15 (e), it may be a hold of the data at the previous time or an internally divided value of the data before and after. After that, a band pass filter (or a low pass filter) is applied and smoothed, and converted into the original sampling frequency data (f). The operation in the highest band is the same except that a high-pass filter is used for band division / synthesis and no thinning means and interpolation means are used.

The signals after the sound image localization processing in each of these bands are added and combined, and then sent to the output section. The signal from the output section is converted into an analog signal by the D / A converter,
It is amplified by the amplifier and output by the speaker.

The band dividing means is shown in FIGS.
3), the same applies when the band synthesizing means is shown in FIGS. 14B2 to 14B3. In this case, the band is p octave (p
Is a positive real number), the same parameters can be used for the low-pass filter and the high-pass filter, and the memory for storing the parameters can be reduced.

In FIG. 14, (a4) and (a5) show the thinning means, and (b4) and (b5) show the interpolating means omitted. In FIG. 16A, the transfer function T (z ⁻¹ ) is

[0081]

(Equation 6)

In the system that can be represented as shown in FIG.
In (b), the transfer function T (z ^-D ) is

[0083]

(Equation 7)

The system represented by, that is, FIG.
It can be converted into a system represented by (b). Therefore, as shown in FIG. 17, the sound image localization processing may be performed using the data for each D. In this case, the number of calculations can be reduced by omitting the filter of the band synthesizing means.

As shown in FIG. 18, the number of input channels is k, the number of band dividing means is k, and the number of direction localization means provided in one sound image localization means is also k. The outputs of the respective direction localization means are added. Then, the input channel can be expanded to a plurality by inputting it to the crosstalk canceling means.

Although the number of band divisions is 4 in this embodiment, the same effect can be obtained when the number of band divisions is 3 or less or 5 or more.

(Embodiment 7) A seventh embodiment of the present invention is shown in FIG.
This will be described with reference to FIG. In the apparatus of FIG. 1B, the sound image localization means 10a in the high frequency band does not include the crosstalk canceling means. In the high frequency range, the phase shift due to the shift in the listening position becomes large, and it becomes difficult to cancel the crosstalk. For example, frequency 1
At 0 kHz, the wavelength is about 3.4 cm, and if the propagation paths of the sound to the left and right ears deviate by half a wavelength of 1.7 cm, the phases are opposite to each other, and crosstalk is strengthened instead of canceling crosstalk. Normally, the ratio of high frequency energy to all frequency components is not so large, so
There is no particular problem even if the high-frequency crosstalk canceling means is removed, and it is possible to reduce the parameters and the number of calculations and expand the listening range.

(Embodiment 8) FIG. 2 shows the eighth embodiment of the present invention.
Description will be given with reference to 0. This shows an out-of-head localization processing device for headphones with 2-channel input. The number of band divisions is three. In this case, since the propagation of sound from one speaker of the headphones to the other ear is quite small, the crosstalk canceling means can be omitted. Also,
Reflected sound generating means may be added to the present sound image localization device. This is because the crosstalk canceling means omitted as shown in FIG. 20 (a), the direction localizing means as shown in FIG. 20 (b) and the band synthesizing means as shown in FIG. It may be connected in parallel with the real sound image localization apparatus as shown in FIG. By adding the reflected sound generating means in this way, it is possible to obtain an effect that there is more realism, there are few individual differences, and that many people localize the sound image outside the head.

In particular, when the reflected sound generating means is included in the sound image localization means divided into bands as shown in (a) and (b), it is possible to separately process the high range where the reverberation time is short and the low range where the reverberation time is long. It is also possible to compress the order of the filter of the reflected sound generating means.

[0090]

As described above, according to the present invention, in the process of designing a filter in the sound image localization method, an unstable pole is not generated, and a troublesome operation for stabilization is unnecessary.
The spatial ratio transfer function can be easily obtained, and its effect is great. In addition, the sound image localization apparatus of the present invention divides the frequency band and performs sound image localization processing, thereby
Both sound image localization and stability are compatible, and the effect is great.

[Brief description of drawings]

FIG. 1 is a conceptual diagram showing a measurement system of a spatial ratio transfer function and a block diagram of a three-dimensional sound field processing device.

FIG. 2 is a principle diagram for performing stereoscopic sound field processing and spatial localization processing using a spatial transfer function.

FIG. 3 is a block diagram of a circuit that performs three-dimensional sound field processing.

FIG. 4 is a block diagram of a circuit of a direction localization unit and a crosstalk cancellation unit.

FIG. 5: Block diagram for identification of spatial ratio transfer function.

FIG. 6 is a conceptual diagram showing a measurement system of a spatial ratio transfer function.

FIG. 7 is a block diagram showing the procedure of the first embodiment.

FIG. 8 is a conceptual diagram for measuring an inter-aural arrival time difference using an impulse signal.

FIG. 9 is a block diagram showing the procedure of the second embodiment.

FIG. 10 is a block diagram showing the procedure of the third embodiment.

FIG. 11 is a block diagram showing the procedure of the fourth embodiment.

FIG. 12 is a conceptual diagram for obtaining a spatial ratio transfer function from two spatial transfer functions.

FIG. 13 is a conceptual diagram of band division.

FIG. 14 is a block diagram showing a band dividing unit and a band synthesizing unit.

FIG. 15 is a diagram showing operations of a band dividing unit and a band synthesizing unit.

FIG. 16 is a block diagram showing band division processing.

FIG. 17 is a diagram showing data used by filters of a band dividing unit and a band synthesizing unit that do not use an interpolating unit and a thinning unit.

FIG. 18 is a block diagram showing a circuit of a stereophonic sound field processing apparatus having a plurality of inputs.

FIG. 19 is a block diagram showing a circuit of Example 7.

FIG. 20 is a block diagram showing a circuit of Example 8.

[Explanation of symbols]

 1a, 1b Acoustic signal generator 2a, 2b Speaker 2c, 2d Headphone speaker 3a-3d Microphone 4A-4D System identification means 4E AR model identification means 4F Linear model identification means 4a Unknown system input signal input section 4b Unknown system output signal input section 4c Moving average section parameter 4d Auto-regression section parameter 4e Parameter successive correction means 5a, 5b Data recorder 6a, 6b Band division means 7a, 7b Input section 7c, 7d Output section 8a-8d Sound image localization means 9a-9h Directional localization means 10a , 10b, 10c Crosstalk canceling means 11a, 11b Band synthesizing means 12a to 12g Adder 13a Tone correction filter 13b Interaural difference filter 13c, 13d Directional localization filter 13e, 13f Crosstalk generation filter 14A, 14B, 14C Arrival time difference Estimator 15 Inverse filter calculator 16A, 16B Spatial transfer function 17 Inverse calculator 18 Spatial ratio transfer function 19a-19g High pass filter 20a, 20b Band pass filter Filters 21a to 21g Low pass filter 22a to 22i Decimation means 23a to 23i Interpolation means 24a to 24d Low pass filter or band pass filter 25a, 25b Digital filter 26a to 26d Digital filter 27a to 27e Reflected sound generation means

Claims (15)

[Claims] 1. A first spatial transfer function from a sound source position in an actual sound field to a specific location of a listener or a left ear of a pseudo head, a second spatial transfer function to a specific location of the right ear, and a virtual sound source. A sound image including a calculation of a third spatial transfer function from the position to the specific location of the left ear and a fourth spatial transfer function of the specific location of the right ear, and localizing the sound image to an arbitrary position based on the calculation result. A localization method, wherein an acoustic signal satisfying a continuous excitation condition is generated from a sound source position or a virtual sound source position, the acoustic signal is recorded at the specific places of the left and right ears, and any one of the first to fourth An unknown transfer function represented by a product of an inverse function of the spatial transfer function and any one of the remaining spatial transfer functions is obtained by regarding the recorded acoustic signal as an input / output of the unknown transfer function. Sound image localization method. 2. A model in which the output data is relatively delayed by the input / output of an unknown transfer function, and the present output is relatively expressed by a linear operation of the past output and the present and past and future inputs. The sound image localization method according to claim 1, wherein the unknown transfer function is obtained by using. 3. The arrival time difference of the sound between both ears is measured, and the data of the earliest arrival time is shifted by the obtained arrival time difference to shift the arrival time difference and the above-mentioned arrival time difference among unknown transfer functions. 3. The sound image localization method according to claim 1, wherein the sound image localization method is obtained by separating from elements other than the arrival time difference. 4. A first spatial transfer function from a sound source position in a real sound field to a specific location of a listener or the left ear of a pseudo head, a second spatial transfer function to a specific location of the right ear, and a virtual sound source. A sound image including a calculation of a third spatial transfer function from the position to the specific location of the left ear and a fourth spatial transfer function of the specific location of the right ear, and localizing the sound image to an arbitrary position based on the calculation result. A localization method, wherein a first acoustic signal satisfying a continuous excitation condition is generated from a sound source position or a virtual sound source position, the first acoustic signal is recorded at the specific places of the left and right ears, and the sound source position or Estimating the difference between the propagation time of the sound from the virtual sound source position to the specific location of the left ear and the propagation time of the sound to the specific location of the right ear, and autoregressing the spatial transfer function of any one of the first to fourth Estimated in the form of a model, the inverse of the spatial transfer function A function is subjected to a convolution operation on a signal satisfying the continuous excitation condition to obtain a second acoustic signal, and the second acoustic signal is output again from the sound source position or the virtual sound source position, and the specific portions of the left and right ears are output. And recording the second acoustic signal with an unknown transfer function represented by the product of the inverse function of any of the first to fourth spatial transfer functions and any of the remaining spatial transfer functions, A sound image localization method, wherein the sound image localization method is performed by using the second acoustic signal or including the first acoustic signal as the input / output of the unknown transfer function. 5. A first spatial transfer function from a sound source position in a real sound field to a specific location of a listener or a left head of the pseudo head, a second spatial transfer function to a specific location of the right ear, and a virtual sound source. A sound image including a calculation of a third spatial transfer function from the position to the specific location of the left ear and a fourth spatial transfer function of the specific location of the right ear, and localizing the sound image to an arbitrary position based on the calculation result. A localization method, in which an acoustic signal satisfying a continuous excitation condition is generated from a sound source position or a virtual sound source position, the acoustic signal is recorded at a specific position of the left and right ears, and the sound source position or the virtual sound source position of the left ear is recorded. The difference between the propagation time of the sound to the specific location and the propagation time of the right ear to the specific location is estimated, and the first to fourth spatial transfer functions are estimated by the system identification means in the form of an autoregressive model. , Any of the first to fourth The unknown transfer function represented by the product of the inverse function of the spatial transfer function and any one of the remaining spatial transfer functions is obtained in the form of an IIR filter that can be expressed by division of two autoregressive models. Sound image localization method. 6. A transfer function having a lower order than that of the transfer function is re-identified by inputting a signal series satisfying a continuous excitation condition to the obtained transfer function and re-identifying from the output series and the input signal series. The sound image localization method according to claim 1, wherein 7. The recorded acoustic signal is replaced with the recorded acoustic signal for use as an input / output of an unknown transfer function, according to any one of claims 1 to 5. Sound image localization method described. 8. The recorded acoustic signal is divided into a plurality of frequency bands by band dividing means, and the unknown transfer function is obtained by regarding the band-divided acoustic signal as the input / output of an unknown transfer function. The sound image localization method according to any one of claims 1 to 5, wherein 9. A first spatial transfer function from a sound source position in an actual sound field to a specific location of a listener or a left ear of a pseudo head, a second spatial transfer function to a specific location of the right ear, and a virtual sound source. A sound image including a calculation of a third spatial transfer function from the position to the specific location of the left ear and a fourth spatial transfer function of the specific location of the right ear, and localizing the sound image to an arbitrary position based on the calculation result. In the localization device, a band dividing unit that divides an acoustic signal into a plurality of frequency bands, a band synthesizing unit that synthesizes the divided acoustic signals of each frequency band, and a filter parameter for each sound image localization position are provided. The sound image localization means for performing the filter calculation of 1, the acoustic signal input section of one or more channels, and the acoustic signal output section of two channels, the acoustic signal input section is connected to the input section of the band dividing means, The output unit of the dividing unit is connected to the input unit of the sound image localization unit corresponding to each frequency band, the output unit of the sound image localization unit is connected to the input unit of the corresponding band synthesizing unit, and the output of the band synthesizing unit. A sound image localization apparatus in which a section is connected to the acoustic signal output section. 10. The sound image localization means comprises a direction localization means and a crosstalk canceling means, an input section of the sound image localization section serves as an input section of the direction localization section, and an output section of the sound image localization section is the crosstalk. The output part of the canceling means, the direction locating means has two output parts, and all the first output parts of the direction locating means are connected to the first input part of the crosstalk canceling means via the adder. 10. The sound image localization apparatus according to claim 9, wherein all the second output sections of the direction localization means are connected to the second input section of the crosstalk cancellation means via an adder. 11. The crosstalk canceling means is removed, and all the first output parts of the direction localization means become the first output parts of the sound image localization means via the adder, and all the second output parts of the direction localization means. 11. The sound image localization apparatus according to claim 10, wherein the output unit of the above-mentioned becomes the second output unit of the sound image localization means through the adder. 12. The sound image localization apparatus according to claim 10, wherein reflected sound generating means is provided between the output portion of the direction localization means and the input portion of the band synthesizing means. 13. The sound image localization apparatus according to claim 10, further comprising reflection sound generation means provided in parallel with the direction localization means. 14. The sound image localization apparatus according to claim 9, wherein reflected sound generating means is provided between the output portion of the band synthesizing means and the acoustic signal output portion. 15. The sound image localization apparatus according to claim 9, further comprising reflection sound generating means provided between the sound input section and the sound output section.