JPH09135499A - Sound image localization control method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve sound quality by correcting the discrepancy in transmission characteristics, which are made different for each person, from a target localized position and a speaker to the head. SOLUTION: The head transmission functions (HRTF) from the speaker set and arranged for practical measurement and from respective target localized positions to the human head or the dummy human head are respectively found (steps 101 and 102) and among these HRTF, move averaging processing is performed to the HRTF provided from the speaker on frequency amplitude characteristics (step 103). Based on this move averaged HRTF and the HRTF provided from the respective sound image localized positions, the transmission characteristics of a signal converting circuit are found and based on these transmission characteristics, coefficients for sound image localization are respectively calculated (step 105). By supplying these coefficients to the signal converting circuit at any suitable prescribed timing, the localization of sound image under inputting is controlled (step 107).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention reproduces signals processed by a plurality of signal conversion circuits supplied with the same sound source from a plurality of transducers arranged apart from each other, and an actual transducer (speaker). The present invention relates to a sound image localization control method that makes a user feel that a sound image is localized at a desired arbitrary position different from the above, and particularly relates to improvement of calculation of data for sound image localization control (transfer characteristics of a signal conversion circuit). .

[0002]

2. Description of the Related Art The applicant of the present application has previously filed Japanese Patent Application No. 4-3434.
No. 59 and Japanese Patent Application No. 4-356358 have been filed. These contents are related to the head related transfer function (Head Related Transfer Fu
nction; hereinafter referred to as HRTF), the characteristics are determined based on actual measurement of HRTF, and the sound image is localized. As a realization means, a sound image is localized by digitally processing the frequency amplitude characteristic and the frequency phase characteristic by signal processing. FIG. 7 is a diagram showing the HRTF measurement system. In this system, a pair of microphones ML and MR are provided on both ears of a dummy head (or human head) DM.
Is installed, the measurement sound from the speaker SP is received, and the recorder D
Source sound (reference data) refL, re to AT
The fR and the to-be-measured sound (measurement data) L and R are recorded in synchronization.

As the source sound XH, impulse sound, white noise, other noises, etc. are used.
As shown in, the position of the speaker SP is 0 degrees (°) on the front side, and the speaker SP is installed at a spatial position of 12 points every 30 degrees.
Each of them is continuously recorded for a predetermined time to obtain the data.

Then, by using these data, for example, the sound image localization control is performed by using the basic principle system of FIG. In the figure, sp1,
sp2 is a speaker arranged on the left and right of the front of the listener. The head-related transfer characteristic (impulse response) from sp1 to the listener's left ear is h1L, and the head-related transfer characteristic to the right ear is h1.
The head-related transfer characteristics from R, sp2 to the left and right ears are h2L, h
2R and pL is the head-related transfer characteristic to the left and right ears of the listener when an actual speaker is placed at the desired localization position x.
x and pRx, the data obtained from the above-mentioned measurement system is subjected to a predetermined waveform processing to obtain the respective coefficients cfLx and cfRx, and the sound image localization control is performed by using the data obtained by performing the convolution calculation processing on these respective coefficients. It was made to do.

[0005]

By the way, in the above-mentioned contents, the components to be folded as HRTFs from the speaker to both ears at the time of calculation of each data (coefficient) used for the convolution operation, and the components at the time of reproduction. HR from the speaker to both ears of the listener
Due to TF mismatch, the quality of reproduced sound may be impaired. Further, due to the above disagreement, there is also a problem that a sound of a high frequency component remains in the front when the sound is localized in the rear, and the sound image is inverted and heard.

The present invention considers the adverse effect of the HRTF variation from the speaker to both ears during reproduction, and the HRTF assumed during calculation of the sound image localization coefficient and the HR during reproduction.
It is intended to provide a sound image localization control method in which the above problem does not occur even if TF mismatch occurs.

[0007]

In order to solve the above problems, the present invention reproduces a signal processed by a plurality of signal conversion circuits to which the same sound source is supplied from a plurality of transducers arranged apart from each other. Then, in the sound image localization control method that makes the listener feel that the sound image is localized at an arbitrary position different from the transducer, in the human head or pseudo from the localization target position and the transducer set and set for actual measurement. The transfer functions up to the human head are obtained respectively, and of these transfer functions, the transfer function obtained from the transducer is subjected to moving averaging processing on the frequency-amplitude characteristics, and the transfer function subjected to this moving averaging processing and the respective sound images. A sound image localization control method characterized in that the transfer characteristic of the signal conversion circuit is obtained based on a transfer function obtained from a localization position.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of the sound image localization control method according to the present invention will be described with reference to the drawings. FIG. 1 is a schematic configuration diagram according to an embodiment of a sound image localization control device to which the control method of the present invention is applied. This device includes a synthesizer 1 which is a sound source, convolvers 2 and 3 for performing a folding operation of measured coefficients, and a coefficient R for storing the coefficients to these convolvers 2 and 3.
OM5, a control sub CPU 4 for controlling the coefficient supply timing, amplifiers 6 and 7, a pair of speakers SP1 and SP2, and the like.

Here, the signals obtained by passing the sound source X to be localized through signal converters cfLx and cfRx (transfer characteristics by a convolver or the like) are sp1 and s, respectively.
Consider playing on p2. At this time, if the signals obtained in the left and right ears of the listener are eL and eR,

EL = h1L · cfLx · X + h2L · cfRx · X (Equation 1) eR = h1R · cfLx · X + h2R · cfRx · X (〃)

On the other hand, when the signals obtained in the left and right ears of the listener when the source X is reproduced from the target localization position are dL and dR, dL = pLxX (Equation 2) dR = pRxX (〃)

In this case, if the signals obtained in the left and right ears of the listener by reproducing sp1 and sp2 match the signals when the source is reproduced from the target position, the listener is as if the speaker exists at the target position. The sound image will be recognized. This condition eL = dL, eR = dR (Equation 1),
From (Equation 2), h1L · cfLx + h2L · cfRx = pLx (Equation 3) h1R · cfRx + h2R · cfRx = pRx (〃) From (Equation 3), cfLx = cfRx−cfLx = (h2R · pLx− h2L · pRx) / H (equation 4a) cfRx = (− h1R · pLx + h1L · pRx) / H (〃) where H = h1L · h2R−h2L · h1R (equation 4b)

Therefore, if a signal to be localized is processed by a convolver (convolution operation processing circuit) or the like using the transfer characteristics cfLx and cfRx calculated by (Equation 4a) and (Equation 4b), a sound image is obtained at a target position x. Can be localized. There are various concrete methods for realizing the signal conversion apparatus, but it is sufficient to use an asymmetric FIR digital filter (convolver). The final transfer characteristic when used in the FIR digital filter is a time response function.

That is, as the transfer characteristics cfLx and cfRx at the required localization position x, (Equation 4a) and (Equation 4b)
The coefficients cfLx and cfRx are created in advance as the coefficients for realizing the one obtained in step 1 by the FIR filter processing, and are prepared as ROM data.
If the coefficient of the required sound image localization position is transferred from the ROM to the FIR digital filter and the signal from the sound source is subjected to the convolutional arithmetic processing and reproduced from the pair of speakers, the sound image is localized at any desired position.

The sound image localization control method based on the above principle will be described in detail with reference to FIG. The figure shows the steps of the method for controlling the sound image localization.

Measurement of head related transfer function (step 101) The HRTF of each measurement point is measured by the method described with reference to FIGS.

HRTF impulse response (Impulse Re
sponse; hereinafter referred to as IR) (step 10)
2) In step 101, the source sounds (reference data) refL and refR and the to-be-measured sounds (measurement data) L and R, which are synchronously recorded, are stored in a workstation (not shown).
Process above. Letting X (S) be the frequency response of the source sound (reference data), Y (S) be the frequency response of the sound to be measured (measurement data), and IR (S) be the frequency response of the HRTF at the measurement position (Equation 5). There is an input / output relationship as shown in. Y (S) = IR (S) · X (S) (Equation 5) Therefore, the frequency response of the HRTF is IR (S) = Y (S) / X (S) (Equation 6) is there.

Therefore, the frequency response X of the reference
(S), the frequency response Y (S) of the measurement data is obtained by cutting out the data obtained in step 101 with a window that is time-synchronized, and performing a finite Fourier series expansion by FFT conversion to calculate a discrete frequency, From 6), HRT
The frequency response IR (S) of F is obtained by a known calculation method. In this case, in order to improve the accuracy of IR (S) (improvement of SN ratio), it is preferable to calculate IR (S) for several hundred windows that are temporally different and average them. Then, the frequency response IR (S) of the calculated HRTF is inverse F
FT conversion is performed to obtain a time axis response (impulse response) IR (first IR) of HRTF.

Each head-related transfer characteristic h1L, h1R, h2
L, h2R moving averaging process (smoothing process) (step 103) Each head-related transfer characteristic h1L, obtained by actual measurement as described above.
The frequency response is obtained by FFT-transforming h1R, h2L, and h2R, and moving average of the frequency response is performed with a width according to the critical bandwidth. Here, it is the gist of the present invention to perform this moving averaging process, which will be described below with reference to FIGS.
As a technique for removing unnecessary peaks and dips, the above h1L, h1R, h2L, and h2R are FFT-converted to H1L (S), H1R (S), H2L (S), H2R.
(S) is obtained and H1 obtained as this discrete frequency response
L (S), H1R (S), H2L (S), H2R (S)
Is moving averaged and the moving frequency averaged discrete frequency response is inverse F
It is conceivable to perform FT conversion to obtain the time response of the head-related transfer characteristic.

Generally, when performing moving averaging, it is general that a certain bandwidth is set and each frequency band is moving averaged with the same bandwidth. However, human hearing has a characteristic called "critical band", which is that "the sound is discriminated and the frequency is analyzed based on the bandpass characteristics that are arranged in the entire audible frequency band, and the passband width is narrower in the lower range and wider in the higher range". There is. The present application focuses on this critical band, and adopts a method of optimizing the bandwidth for moving and averaging depending on the frequency band according to the critical band. The critical bandwidth CBc (Hz) is expressed by the following equation when expressed by the center frequency f (Hz).

[0021]

(Equation 1)

A concrete example of the above processing contents is shown in FIG. FIG. 3A shows the time response of the head related transfer function obtained by actual measurement. The figure (B) displays the discrete frequency response and the critical bandwidth CBc obtained by the FFT transformation of the figure (A). The figure (C) shows the discrete frequency response obtained by moving and averaging the figure (B) with the critical bandwidth. FIG. 6D is a time response of the head related transfer function obtained by performing an inverse FFT transform on FIG.

By doing so, FIGS. 3C and 3D are used.
As is clear from the above, it is possible to remove unnecessary peaks and dips in the high frequency range while keeping the characteristic of the frequency response in the low frequency range necessary for sound image localization. By the above processing, the characteristic of the HRTF that a general person has is left, and the sharp dip is eliminated, and as a result, generation of unnecessary sharp peaks in the frequency amplitude characteristic reproduced at the time of reproduction is prevented.

IR (impulse response) shaping processing (step 104) Here, the IR obtained in step 102 is shaped. First, for example, by the FFT transform, the first obtained in step 102
The IR of the is expanded at discrete frequencies over the audio spectrum, and unnecessary band (a large dip occurs in the high range, but this is unnecessary that does not affect sound image localization) is removed by a BPF (bandpass filter). To do. By limiting the band in this way, unnecessary peaks and dips on the frequency axis are removed and unnecessary coefficients do not occur in the cancel filter, so that the convergence is improved and the coefficients can be shortened.

Then, the band-limited IR (S) is inverted F
FT conversion is performed, and IR (impulse response) is cut out on the time axis by a window (for example, a cosine function window) and subjected to window processing (second IR). By performing the window processing, the effective length of the IR is not long, the convergence of the cancel filter is improved, and the sound quality is not deteriorated. FIG. 4 shows a specific example of IR (impulse response) of HRTF. The horizontal axis is time (sample clock is 4
The clock unit is 8 kHz), and the vertical axis is the amplitude level. The chain double-dashed line shows the window.

Cancellation filters cfLx, cfRx
(Step 105) The cancel filters cfLx and cfRx, which are convolvers (convolutional integration circuits), are cfLx = (h2R.pLx-h2L.pRx) / H (Formula 4a) cfRx = (-h1R.pLx + h1L.pRx) / H (〃) However, it is H = h1L.h2R-h2L.h1R (Formula 4b).

Here, the speakers sp1 and sp to be arranged
Head related transfer characteristics h1L, h1R, h2L, h2R
And, as the head-related transfer characteristics pLx and pRx when the actual speaker is arranged at the target localization position x, the second IR (reshaped) for each angle θ obtained in steps 101 to 104 described above. Impulse response).

The head-related transfer characteristics h1L and h1R are L in FIG.
Corresponding to the position of the channel speaker, if it is installed at 30 degrees (θ = 330 degrees) from the front to the left, an IR of θ = 330 degrees is used. Head-related transfer characteristic h2
R and h2L correspond to the position of the R channel speaker in the figure, and are, for example, 30 degrees (θ = 30) from the front to the right.
, The IR of θ = 30 degrees is used.

The head-related transfer characteristics pLx and pRx are 90 degrees from the front, which is the target sound source localization position.
The cfLx of the entire space corresponding to that by substituting IR for every 30 degrees in a wide space (overall space) beyond that of 180 °
cfRx, that is, 12 sets of cancellation filters cfLx and cfRx are obtained every 30 degrees (in FIG. 5,
The 240 ° position is taken as an example). The cancel filters cfLx and cfRx group are finally obtained as IR (impulse response) which is a response on the time axis.

The calculation of the cancel filters cfLx and cfRx according to (Equation 4a) is as follows. First, H ^-1 , which is a kind of inverse filter for H in (Equation 4b), is obtained by the least squares method, and this is subjected to inverse FFT conversion to obtain a time function h (t). Also, the terms h1L and h1 of (Equation 4a)
By expressing R, h2L, pRx, pLx, and h2R by a time function, respectively, the following equation holds.

CfLx (t) = (h2R · pLx−h2L · pRx) · h (t) (Formula 7) cfRx (t) = (− h1R · pLx + h1L · pRx) · h (t) (〃)

Therefore, the coefficients of the cancel filters cfLx and cfRx can be obtained from these (formula 7). As is clear from (Equation 7), in order to shorten the coefficients of the cancel filters cfLx and cfRx, the head-related transfer characteristics h1L, h1R, h2L, pRx, pLx,
It is extremely important to shorten each h2R. Therefore, as described above, various processes such as window processing and shaping processing are performed in steps 101 to 104 to perform the head-related transfer characteristics h1L, h1R, h2L, pRx, pLx, h.
2R is shortened.

FIG. 6 shows the cancel filter coefficient cfLx,
The concrete coefficient sequence of cfRx is shown. The horizontal axis is time (time in clock units where the sample clock is 48kHz), and the vertical axis is the amplitude level. The chain double-dashed line shows the window.

Scaling of the cancellation filter at each localization point x (step 106) Further, the spectrum distribution of the sound source (source sound) actually processed by the convolver (cancellation filter) is like pink noise statistically. There are things that are distributed or that drop gently in the high range.In any case, since the sound source is different from a single sound, there is a risk that overflow will occur when convolution calculation (integration) is performed and distortion will occur. is there.

Therefore, in order to prevent overflow,
Of the coefficients of the cancel filters cfLx, cfRx, the maximum gain (for example, the cancel filters cfLx, c
Find the square sum of each sampled value of fRx) and scale all the coefficients so that overflow does not occur when the coefficient and the white noise of 0 db are convoluted. In practice, it is advisable to attenuate the maximum absolute value to be about 0.1 to 0.4 (for example, 0.2) with respect to the allowable level (amplitude) 1.

Then, the window (cosine window) shown in FIG. 6 is used to perform window processing so that both ends become 0 according to the number of coefficients of the actual convolver, and the effective length of the coefficient is shortened. In this way, the data group (in this example, 12 sets of convolver coefficient groups capable of sound localization for every 30 degrees) cfLx and cfRx that are subjected to scaling processing and finally supplied to the convolver as coefficients are obtained.

A signal from a sound source is convoluted and reproduced (step 107) For example, as a sound reproducing device of a game machine, as shown in FIG.
A pair of loudspeakers sp1 and sp2 are arranged at intervals so that the acoustic signals processed by the pair of convolvers (convolution operation processing circuits) 2 and 3 are reproduced on the pair of loudspeakers sp1 and sp2. Configure.

A signal from the same sound source X (for example, a flight sound from a game synthesizer) 1 is supplied to the pair of convolvers, and the IR coefficients cfLx and cfRx (for example, the flight sounds generated at step 105) (for example, , When the sound of the flight is to be localized at a position of 120 degrees leftward (θ = 240 degrees), a coefficient of θ = 240 degrees) is selected and set in the convolvers 2 and 3.

For example, a coefficient for a desired localization position is transferred from the coefficient ROM 5 to the pair of convolvers 2 and 3 by the control sub CPU 4 based on a sound image localization command from a main CPU (central processing unit) of a game machine or the like. .

In this way, the pair of convolvers 2, 3
Thus, the signal from the sound source X1 is subjected to convolution calculation processing on the time axis, and a pair of speakers sp arranged apart from each other.
It is reproduced from 1, sp2. Pair of speakers sp1, s
The sound reproduced from p2 has its sound image localized such that the crosstalk to both ears is canceled and the sound source is located at the desired position, and is heard by the game operator (listener) M, which is extremely realistic. Played as a sound.

The coefficient of the convolver is switched by sequentially selecting the optimum sound image position along with the transition of the movement of the airplane according to the operation of the operator. Further, when the flight sound is changed to, for example, a missile sound, the source sound from the sound source X is changed from the flight sound to the missile sound. In this way, the sound image can be freely localized at any position.

In the above description, the critical bandwidth is specifically described as being defined by an equation, but the critical bandwidth is not limited to this. What is similar to this definitional expression, or approximated by a logarithmic expression, may be any as long as its width is narrower in the lower range and wider in the higher range.

As a transducer for reproduction, a headphone can be used instead of the pair of speakers sp1 and sp2. In this case, since the HRTF measurement conditions are different, it is advisable to prepare the coefficient separately and switch it according to the reproduction situation. I shown in step 103
R (impulse response) shaping processing is not always necessary, and sound image localization control is possible even if omitted. Further, the configuration described in the embodiment, in which the signal processed by the pair of convolvers to which the same sound source is supplied from the pair of transducers arranged apart from each other is reproduced, is the minimum configuration for obtaining the effect of the present application. Is shown. Therefore, as a matter of course, a pair, that is, two or more transducers and convolvers may be additionally configured, and when the convolver coefficient is long, the coefficient is divided into a plurality of parts. It may be configured with a convolver.

[0044]

As described above in detail, in the sound image localization control method according to the present invention, a signal processed by a plurality of signal conversion circuits to which the same sound source is supplied from a plurality of transducers arranged at a distance. In a sound image localization control method for reproducing a sound image so that the listener feels that the sound image is localized at an arbitrary position different from that of the transducer, and a transfer characteristic from a preset transducer to both ears of the listener. The obtained transfer characteristic is subjected to a moving averaging process on the frequency amplitude characteristic, and the signal conversion circuit is based on the transfer characteristic subjected to the moving average process and the transfer function measured at each sound image localization position. Since it is the transfer characteristic of the high frequency range, it is moving averaged in the optimized bandwidth with the critical bandwidth, leaving the characteristics of the frequency response in the middle and low range necessary for sound localization. It is possible to eliminate the unnecessary peaks and dips in etc., H normally general public has
It is possible to obtain the natural sound quality by eliminating the sharp dip while keeping the characteristics of the RTF, and as a result, preventing the generation of an unnecessary sharp peak in the frequency amplitude characteristic reproduced at the time of reproduction.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a sound image localization apparatus based on a sound image localization control method according to the present invention.

FIG. 2 is a chart showing steps of a sound image localization control method according to the present invention.

FIG. 3 is a diagram illustrating a sound image localization control method according to the present invention, in which FIG. 3A is a time response of a signal processing circuit (convolver) obtained based on a head related transfer function obtained by measurement; The figure (B) shows the discrete frequency response and the critical bandwidth obtained by the FFT transformation of the figure (A), and the figure (C) shows the moving average of the figure (B) with the critical bandwidth. The converted discrete frequency response is displayed, and FIG. 6D is the time response of the signal processing circuit (convolver) obtained by performing the inverse FFT conversion on FIG.

FIG. 4 is a diagram showing a specific example of IR (impulse response) of HRTF.

FIG. 5 is a diagram illustrating a calculation example of a cancel filter.

FIG. 6 is a diagram showing a specific example of coefficients of a cancel filter.

FIG. 7 is a configuration diagram showing an HRTF (head related transfer function) measurement system.

FIG. 8 is a diagram illustrating the points of HRTF measurement.

FIG. 9 is a configuration diagram showing a basic principle of a sound image localization control method according to the present invention.

[Explanation of symbols]

1 sound source (synthesizer) 2, 3 convolver 5 coefficient ROM 101 step for measuring head related transfer function (HRTF) 102 step for calculating impulse response of HRTF 103 head related transfer function h1L, h1R, h2L, h2
Moving averaging step of R 105 Step of calculating cancellation filters cfLx, cfRx 107 Step of convoluting and reproducing a signal from a sound source sp1, sp2 Speakers h1L, h1R Head-related transfer characteristics from the speaker sp1 to the listener's left and right ears h2L , H2R Head-to-head transfer characteristics from speaker sp2 to the left and right ears of the listener pLx, pRx Head-to-head transfer characteristics to the left and right ears of the listener when an actual speaker is placed at the intended localization position x

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI Technical display location H04R 1/40 310 H04S 1/00 K H04S 1/00 7/00 F 7/00 G06F 15/36 A

Claims (1)

[Claims] 1. A signal processed by a plurality of signal conversion circuits to which the same sound source is supplied is reproduced from a plurality of transducers arranged apart from each other, and a signal is reproduced by a listener at an arbitrary position different from that of the transducer. In the sound image localization control method that makes the user feel that the sound image is localized, the transfer function from the transducer set for measurement and each localization target position to the human head or the pseudo human head is obtained, and the transfer function is calculated from these transfer functions. , The transfer function obtained from the transducer is subjected to moving averaging processing on the frequency-amplitude characteristic, and the signal is obtained based on the transfer function obtained from this moving averaging processing and the transfer function obtained from each of the sound image localization positions. A sound image localization control method characterized in that a transfer characteristic of a conversion circuit is obtained.