JP4691662B2 - Out-of-head sound localization device - Google Patents

Description

  The present invention relates to an out-of-head sound image localization technique that makes it possible to perceive the direction of a sound source outside the head when listening to stereo headphones.

Out-of-head sound image localization reproduces a sound image with a sense of openness outside the head by correcting the acoustic propagation characteristics of the ear canal when wearing headphones with a digital filter and presenting it to the listener.
As shown in FIG. 13, the spatial acoustic characteristics are given by complex multiplication of an out-of-head sound image localization transfer function SLTF (Sound Localization Transfer Function) and the sound source signal S (ω).
The subscripts L and R in the figure indicate the left ear and the right ear, respectively.
The SLTF obtains a spatial sound transfer function SSTF (Spacial Sound Transfer Function) from the speaker 3 in the free space to the microphone 2 at the entrance of the ear canal, and divides this by the transfer function LSTF (Loud Speaker Transfer Function) of the speaker 3. The characteristic compensated is further divided by the ear canal transfer function (ECTF) from the headphone 4 to the microphone 2 of the listener 8. That is, SLTF = SSTF / (LSTF · ECTF).
The SLTF is complex-multiplied with the sound source signal S (ω) and presented to the listener 8 via the headphones 4 to realize out-of-head sound image localization.

However, the binaural characteristics such as arrival time difference, level difference, frequency characteristics, etc. necessary for the listener to perceive the sound source position, and the monoaural characteristics are subtle differences such as the head, trunk, and pinna. It is known to change.
For this reason, there are individual differences in the human head, torso, and auricle, and so-called front / rear misjudgment where a sound image that should be perceived forward is perceived backward in a general-purpose transfer function measured using a dummy head. The sound image localization is inaccurate or, in the worst case, the sound image is localized in the head.

  The problems to be solved are as described above, and the present invention provides an out-of-head sound image localization apparatus for stereo headphones that can obtain a good localization feeling for an unspecified number of listeners using a general-purpose transfer function. It was made for the purpose of doing.

Therefore, the present invention sets the moving sound source positions A and B having the angle difference θ (swing angle) to the left and right of the front two-channel stereo real sound source positions L and R having the azimuth angle ± Φ to the listening point of both ears. Transfer functions Ha (ω) and Hb (ω) for each path are obtained for each moving sound source position, and a waveform cut-out function w ₁ (front and back) overlaps the sound source signal s (t) of each channel of stereo headphones. t), the frame is sequentially cut out, the cut-out means for dividing the sound source signal s (t) into a plurality of frame signals sn (t), and the transfer function Ha (ω), alternately for the frame signal sn (t), Convolution means for generating frame signals sa (t) and sb (t) including position information of the moving sound source positions A and B by convolving Hb (ω), and frame signals sa (t) and sb (t) contrast, the waveform synthesis function w The multiplied by frame signal sa' obtained (t), Sb' and overlap-add (t) are alternately while smooth discontinuities in the waveform, stereo actual sound source position L, the sound source installed in R movement Adding means for generating a synthesized signal s ′ (t) including movement information that reciprocates between the sound source positions A and B at a constant period T (switching time), and thereby presenting a sound image to both ears of the listener The main feature is that the sound image is localized out of the head by presenting a swing sound image having an angle φ, a swing angle θ, and a switching time T.

The present invention takes advantage of the fact that animals including humans are generally sensitive to the perception of moving sound sources, and transfers the sound source positions A and B to the sound source signal s (t) of each channel of stereo headphones. Since the sound image is moved by alternately convolving the functions Ha (ω) and Hb (ω), high-accuracy out-of-head sound image localization using a general-purpose transfer function can be realized.
In addition, since the waveform of the overlap interval is synthesized to smooth the discontinuity of the waveform at the time of overlap addition, there is no sense of incongruity due to amplitude fluctuation that occurs when the transfer function is switched, and a more natural signal waveform sound image can be presented .

  Embodiments of the present invention will be described below.

FIG. 1 shows a configuration diagram of an out-of-head sound image localization apparatus embodying the present invention.
In the figure, only one channel series of the stereo headphone system is shown in order not to make the figure complicated.
The out-of-head sound image localization apparatus has a measuring system microphone 2 connected to the input side of the personal computer 1 and a measuring system speaker 3 and a reproducing system headphone 4 connected to the output side by a switch 5 so as to be switchable.
The microphone 2 is connected to the personal computer 1 through an A / D converter 21, a low pass filter 22 as an anti-aliasing filter, and an amplifier 23.
The speaker 3 is connected to the switch 5 via the selector 31 and the amplifier 32.
The headphone 4 is connected to the switch 5 via the amplifier 41.
The switch 5 is connected to the personal computer 1 through a D / A converter 51 and a low pass filter 52 as a smoothing filter.

Measurement is performed by installing a dummy head 7 simulating a human head, torso, and auricle shape in the measurement chamber 6, setting the microphone 2 at the entrance of the ear canal of both ears of the dummy head 7, and mounting the dummy head 7. The speakers 3 are arranged at equal angular intervals on the forward arc centered.
The selector 31 is switched to move the position of the speaker 3 that outputs the measurement sound, and the measurement sound of the speaker 3 is picked up by the microphone 2 inserted into the ear of the dummy head 7 and is impulse response in order at predetermined measurement angle intervals. Measure.
When the measurement angle interval is narrower than the actual arrangement interval of the speakers 3, for example, an interval of 1 degree, the impulse response is obtained by calculation using a linear interpolation method considering the arrival time difference.
The speakers 3 may be attached at equal angular intervals to the circular frame.
In that case, the frame can be rotated in the horizontal direction so that the measurement angle interval is narrower than the actual arrangement interval of the speakers 3, for example, an interval of 1 degree.

  The measurement is performed for both ears of the dummy head 7, and as shown in FIG. 2, the impulse response h1L (t) between the left speaker 3L and the left ear, and the impulse response h2L between the left speaker 3L and the right ear. (T) The impulse response h1R (t) between the right speaker 3R and the right ear and the impulse response h2R (t) between the right speaker 3R and the left ear are measured.

FIG. 3 shows a block diagram of a measurement system processed in the personal computer 1.
The measurement system includes a signal generation unit 11, an impulse response calculation unit 12, and a memory storage unit 13. The measurement system measures the impulse response between the speaker 3 and the microphone 2 in the measurement chamber 6 and determines the path from the sound source to the listening point. Find the transfer function.

The signal generator 11 generates an impulse response measurement input signal x (t) such as an M-sequence signal (Maximum Length Sequence) or a time stretched pulse (Time Stretched Pulse) and outputs it to the speaker 3.
The input signal x (t) is output as sound by the speaker 3 and collected by the microphone 2 inserted into the ear of the dummy head 7.
The sound collected by the microphone 2 is converted into a digital signal, and the impulse response together with the input signal x (t) is output as the output signal y (t) when x (t) is input to the linear system with the impulse response h (t). Input to the calculator 12.
The impulse response calculation unit 12 performs a transfer function H () that is a Fourier transform of the impulse response h (t) from the Fourier transform X (ω) of the input signal x (t) and the Fourier transform Y (ω) of the output signal y (t). ω) = Y (ω) / X (ω) is calculated.
Measurement is performed by moving the position of the speaker 3, impulse responses h (t) of different sound source positions are sequentially acquired at predetermined measurement angle intervals, and transfer functions H (ω) of different sound source positions are sequentially calculated therefrom. To do.
The memory storage unit 13 sequentially stores the transfer functions H (ω) of different sound source positions calculated by the impulse response calculation unit 12 in the memory.

As shown in FIG. 2 and FIG. 4, the sound source signal SL (t) and a signal obtained by convolving the impulse responses h1L (t) and h2L (t) of the transfer function from the position of the real sound source 3L to both ears of the listener, By adding, for each ear, a signal obtained by convolving the sound source signal SR (t) and the impulse response h1R (t), h2R (t) of the transfer function from the position of the real sound source 3R to the listener's ears, respectively. By presenting the obtained two-channel virtual sound sources SiL (t) and SiR (t) using the headphones 4, the listener perceives the synthesized stereo sound image Si (t).
At this time, the presentation angle φ of the synthesized stereo sound image can be controlled by adding a level difference and a time difference to the sound source signals SL (t) and SR (t).
As shown in FIG. 4, the sound image swing method is synthesized by reciprocating the virtual sound sources SiL (t) and SiR (t) between positions A and B displaced by the swing angle θ with a constant switching time T. Further, the stereo sound image Si (t) is displaced in the left-right direction to be localized outside the head of the listener 8.
In FIG. 4, the central angle θ of the arc AB is set as a swing angle and set in a range of 3 to 10 degrees.
Also, the distance from the center of the head of the listener 8 to the stereo real sound source positions L and R is set to about 1.5 m, and the switching time T is set to 200 msec or more.
The sound image presentation angle φ is set by adding a time difference and a level difference to the sound source signal s (t) of each channel.

As shown in FIG. 5, the virtual sound sources SiL (t) and SiR (t) set at the stereo real sound source positions L and R are reciprocated between the moving sound source positions A and B in the reverse direction. As shown in FIG. 6, there is a compound method and a twist method in which the virtual sound sources SiL (t) and SiR (t) are reciprocated between the moving sound source positions A and B in the same direction.
In the companding method, the position of the sound image presented by the virtual sound sources SiL (t) and SiR (t) set at the left and right stereo real sound source positions L and R is expanded and contracted to the left and right, and the sound image is placed out of the head in front of the listener 8. Let it be localized.
In the twist method, the position of the sound image presented by the virtual sound sources SiL (t) and SiR (t) set at the left and right stereo real sound source positions L and R is swung left and right, and the sound image is out of the head in front of the listener 8. Is localized.

FIG. 7 shows a block diagram of a reproduction system for processing in the personal computer 1.
The reproduction system includes a first sound image generating unit 14, a second sound image generating unit 15, a third sound image generating unit 16, a fourth sound image generating unit 17, a first sound image synthesizing unit 18, and a second sound image synthesizing unit 19. The stereo sound signal is input to generate a swing sound image of both ears and output to the left and right channels of the headphones 4.
The swing sound image extracts the transfer functions Ha (ω) and Hb (ω) of the moving sound source positions A and B from the transfer functions H (ω) of different sound source positions stored in the memory storage unit 13 for the left and right channels. It is obtained by complex multiplication with the sound source signal alternately.

The first sound image generation unit 14 generates a swing sound image s1L (t) for the left ear by convolving and multiplying the left stereo signal and the impulse response h1L (t).
The second sound image generation unit 15 convolves and multiplies the left stereo signal and the impulse response h2L (t) to generate a right ear swing sound image s2L (t).
The third sound image generation unit 16 generates a swing sound image s2R (t) for the left ear by convolving and multiplying the right stereo signal and the impulse response h2R (t).
The fourth sound image generating unit 17 generates a swing sound image s1R (t) for the right ear by convolving and multiplying the right stereo signal and the impulse response h1R (t).
The first sound image synthesis unit 18 adds the swing sound images s1L (t) and s2R (t) to generate the left channel output signal of the headphones 4.
The second sound image synthesizer 19 adds the swing sound images s1R (t) and s2L (t) to generate the right channel output signal of the headphones 4.

FIG. 8 shows a processing flow of the sound image generation unit.
First, the sound source signal s (t) is multiplied by a waveform cut-out function w ₁ (t) that overlaps between frames before and after, thereby sequentially cutting out frames, and the sound source signal s (t) is converted into a plurality of frame signals sn (t) = s. Divide into (t) and w ₁ (t) (step 101).
As a result, the length of the sound source signal is divided into the same length as the impulse response, and the processing efficiency of the convolution calculation is increased.
Next, the Fourier transform Sn (ω) = F {sn (t)} of the frame signal sn (t) is obtained by fast Fourier transform (FFT) (step 102).
Next, the A position frame signal Sa (ω) = Sn (ω) is obtained by alternately complex-multiplying the frequency domain frame signal Sn (ω) and the transfer functions Ha (ω) and Hb (ω) of the moving sound source positions A and B. ) · Ha (ω) and B position frame signal Sb (ω) = Sn (ω) · Hb (ω) are generated (step 103).
As a result, transfer functions measured and calculated at different sound source positions A and B are convoluted, and the frame signal becomes a sound source signal including spatial position information.

In the case of the companding method, the order of the transfer functions Ha (ω) and Hb (ω) for performing complex multiplication on the frame signal Sn (ω) of each stereo channel is reversed.
In the case of the twist method, the order of the transfer functions Ha (ω) and Hb (ω) for performing complex multiplication on the frame signal Sn (ω) of each stereo channel is the same on the left and right.

Next, the inverse Fourier transform sa (t) = F ⁻¹ {Sa (ω)}, sb (t) of the A position frame signal Sa (ω) and the B position frame signal Sb (ω) by inverse fast Fourier transform (IFFT). ) = F ⁻¹ {Sb (ω)}, and the sound source signal is returned to the time domain (step 104).
Next, the A position frame signal sa (t) and the B position frame signal sb (t) are multiplied by the waveform synthesis function w ₂ (t) to synthesize the waveforms before and after the frame, and the waveform synthesis A position frame signal sa ′ ( t) = sa (t) · w ₂ (t) and the waveform synthesis B position frame signal sb ′ (t) = sb (t) · w ₂ (t) are generated. As a result, the amplitude of the overlap interval is adjusted, and fluctuations in amplitude at the time of overlap addition are suppressed to smooth the frame connection. Next, the waveform synthesis A position frame signal sa ′ (t) and the waveform synthesis B position frame signal sb ′ (t) are alternately overlap-added and combined, and the synthesis signal s ′ (t) = sa ′ (t ₁ ) + Sb ′ (t ₂ ) +... (Step 105).
As a result, frame signals including positional information of different sound source positions A and B are alternately connected, and the synthesized signal becomes a sound source signal including spatial movement information.

The waveform cut-out function w ₁ (t) and the waveform synthesis function w ₂ (t) may use a fade-in / fade-out function or a modified Hamming window.
If a relationship of multiples of L = (4 · M) is established between the cut-out section L and the frame shift amount M, the waveform can be cut out and synthesized using a modified Hamming window to smoothly form the waveform. Can be synthesized. The modified Hamming window smoothes the signal waveform by simultaneously reducing the amplitude of one and the other so that the power sum in the overlap section is constant, so that the sound image moves smoothly.
Also, the waveform can be synthesized smoothly by using a fade-in / fade-out function. As shown in FIG. 9, when the frame signal is switched from a to b, or when switching from b to a, the fade-in / fade-out function uses the overlap interval of the signals a and b as a cross-fade region. The signal a that fades out is multiplied by a fade-out function wa (t) that linearly inclines and falls, and the signal b that fades in is multiplied by a fade-in function wb (t) that inclines and rises linearly.
Thus, one amplitude is monotonously decreased and the other amplitude is monotonously increased so that the power sum of the overlap section is constant, thereby smoothing the signal waveform so that the sound image moves smoothly.

Hereinafter, examples (evaluation results) of the present invention will be described.
FIG. 10 shows the relationship between the sound image presentation angle φ and the front / rear erroneous determination rate.
FIG. 10 shows an evaluation result when the out-of-head sound image localization apparatus of the present invention is applied to a subject whose localization accuracy deteriorates when a general-purpose transfer function is used, and a sound image presentation angle φ of 0 to 20 degrees on the left and right is shown. On the horizontal axis, the misjudgment rate before and after sound image localization perception is arranged on the vertical axis.
From this, it can be seen that when the sound image presentation angle is 0 degree, the previous technique has a front / rear misjudgment rate of 60%, but the compound method and twist method of the present invention are reduced to 10-30%.
Further, when the sound image presentation angle is 10 degrees, the front-rear error determination rate of 25% in the conventional technique is reduced to 10-20% in the compound method and the twist method of the present invention.
The value at this time is represented by an average value of all the subjects who do not match the switching time T, the swing angle θ, and the transfer function, which will be described later.
From the above, it can be seen that the front / rear misjudgment rate is the worst when the sound image presentation angle is 0 degrees, and that the front / rear misjudgment rate decreases as the sound image presentation angle leaves the front. This proves that the localization accuracy in the front direction is poor both theoretically and experimentally.
In the following, the optimum swing angle θ and switching time T are determined by focusing on the front localization.

FIG. 11 shows the relationship between the swing angle θ and the forward / backward misjudgment rate.
FIG. 11 shows the evaluation result of the twist method when the out-of-head sound localization apparatus of the present invention is applied to a subject (Group B) that substantially matches a subject (Group A) that does not match a general-purpose transfer function. The presentation angle φ is 0 degree, the switching time T is from 200 milliseconds to 1 second, the swing angle θ is set on the horizontal axis, and the pre- and post-judgment error determination rates for sound image localization are arranged on the vertical axis.
From this, it can be seen that the range of the optimum value of the swing angle θ is 3 to 10 degrees.
In addition, although description is abbreviate | omitted about the companding method, the evaluation result similar to the twist method is obtained.

FIG. 12 shows the relationship between the switching time T and the front / rear misjudgment rate.
FIG. 12 shows the evaluation results of the twist method when the out-of-head sound localization apparatus of the present invention is applied to subjects (Group B) that substantially match subjects (Group A) that do not match the general-purpose transfer function. The presentation angle φ is 0 degree, the swing angle θ is 4 degrees and 8 degrees, the switching time T is set on the horizontal axis, and the error determination rate before and after sound image localization perception is arranged on the vertical axis.
From this, it is understood that the range of the optimum value of the switching time T is 200 milliseconds or more.
Similarly, although the description of the companding method is omitted, the same evaluation results as the twist method are obtained.

It is a block diagram of the out-of-head sound image localization apparatus which implemented this invention. It is a conceptual diagram of the measuring method of an impulse response. 2 is a block diagram of a measurement system that is processed in the personal computer 1. FIG. It is a conceptual diagram of the sound image presentation method of the out-of-head sound image localization apparatus which implemented this invention. It is a conceptual diagram of the sound image presentation method by a companding method. It is a conceptual diagram of the sound image presentation method by a twist method. 2 is a block diagram of a playback system that is processed in the personal computer 1; FIG. It is a processing flow of a sound image generation part. It is a conceptual diagram of a fade-in / fade-out process. It is a graph showing the relationship between a sound image presentation angle and a back-and-front error determination rate. It is a graph showing the relationship between swing angle (theta) and a back-and-front misjudgment rate. It is a graph showing the relationship between the switching time T and the back-and-front error determination rate. It is a conceptual diagram of the measuring method of an out-of-head sound image localization transfer function.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Personal computer 11 Signal generation part 12 Impulse response calculation part 13 Memory preservation | save part 14 1st sound image generation part 15 2nd sound image generation part 16 3rd sound image generation part 17 4th sound image generation part 18 1st sound image synthesis part 19 2nd sound image synthesis Part 2 Microphone 21 A / D converter 22 Low-pass filter 23 Amplifier 3 Speaker 31 Selector 32 Amplifier 4 Headphone 41 Amplifier 5 Switch 51 D / A converter 52 Low-pass filter 6 Measurement room 7 Dummy head 8 Audience

Claims (6)

Transfer function of path to listening point of both ears by setting moving sound source positions A and B having an angle difference θ (swing angle) to the left and right of the front two-channel stereo real sound source positions L and R having azimuth angles ± Φ Ha (ω) and Hb (ω) are obtained for each moving sound source position, and for the sound source signal s (t) of each channel of the stereo headphones,
A cutout unit that sequentially cuts out frames by multiplying the waveform cutout function w ₁ (t) in which the front and back of the frame overlap, and divides the sound source signal s (t) into a plurality of frame signals sn (t);
The frame signals sa (t) and sb (t) including the position information of the moving sound source positions A and B by alternately convolving the transfer functions Ha (ω) and Hb (ω) with respect to the frame signal sn (t). A convolution means to generate,
The frame signals sa (t) and sb (t) are alternately overlapped with the frame signals sa ′ (t) and sb ′ (t) obtained by multiplying the waveform synthesis function w ₂ to generate waveform discontinuities. And a synthesized signal s ′ (t) including movement information in which the sound source installed at the stereo real sound source positions L and R reciprocates between the moving sound source positions A and B at a constant period T (switching time). Adding means for generating
With
An out-of-head sound image localization apparatus that presents a sound image presentation angle φ, a swing angle θ, and a switching time T to the listener's ears to localize the sound image out of the head.   2. The head according to claim 1, wherein the swing sound image is expanded and contracted to the left and right by reversing the order of the transfer functions Ha (ω) and Hb (ω) for the frame signal sn (t) in the left and right channels. Outside sound image localization device.   The swing sound image is swung left and right by making the order of the transfer functions Ha (ω) and Hb (ω) to be convoluted in the left and right channels with respect to the frame signal sn (t). Out-of-head sound image localization device. The out-of-head sound localization apparatus according to claim 1, wherein the waveform cut-out function w ₁ (t) and the waveform synthesis function w ₂ (t) are either a fade-in / fade-out function or a modified Hamming window, respectively. .   The out-of-head sound image localization apparatus according to claim 1, wherein the swing angle θ is 3 to 10 degrees.   2. The out-of-head sound image localization apparatus according to claim 1, wherein the switching time T is 200 milliseconds or more.

Images