CN104655266A - Sound field information acquisition method for sound field synthesis - Google Patents

Abstract

The invention discloses a sound field information acquisition method for sound field synthesis and belongs to the technical field of signal processing. Sound pressure data in specific discrete positions on an acquisition plane of a plane microphone array forms a training database; the training database is used for training GMM (Gaussian mixed model) parameters to establish a Gaussian mixed model of sound pressure and spatial information; and a regression prediction module carries out regression prediction on sound field sound pressure in any position on the acquisition plane based on the model, so that the complete sound field sound pressure information on the acquisition plane is obtained.

Description

A kind of sound field information acquisition method for sound field synthesis

Technical field

The present invention relates to a kind of sound field information acquisition method for sound field synthesis, belong to signal processing technology field.

Background technology

Sound field synthesis (wave field synthesis, WFS), based on Huygens' principle, synthesizes three-dimensional sound field in a big way.It has widespread use in the field such as consumer electronics, communication.The sound field of WFS technique reproducible can retain time, the spatial property of original sound field preferably, can give the experience of people " on the spot in person ".The perception of sound-filed simulation is the prerequisite of rebuilding sound field, usual WFS needs a large amount of microphone to carry out the distribution of perception three-dimensional sound field, and microphone distribution is more intensive, and the sound-filed simulation perceived is more accurate, but too much microphone can affect sound-filed simulation, the distribution of sound field even obviously can be changed.

From spatial sampling theorem, spatial sampling cutoff frequency is f _nyq≤ c/ (2 Δ x), wherein, Δ x is sampling unit spacing distance, precise acquisition frequency f≤f _nyqspace sound field in situation, microphone unit spaced furthest used during collection is Δ x=c/ (2f _nyq).Figure 13 in Figure of description is the region of L rice × L rice, at least needs to use (L/ Δ x) in sound field perception ²individual microphone.Such as, f is worked as _nyqwhen=2kHz, L=4, need arrange 2209 microphones in the region of 4 meters × 4 meters, even if do not consider the cost of these microphones, only with regard to the impact on sound field, the microphone of big figure like this, obviously can change original sound field.Therefore, local acquisition technique must be adopted to solve the too much problem of microphone number, WFS is applied in practice.

Patent " Method for reconstructing an acoustic the field. " (DK of Hald and Gomes, G01H 3/12 (2006.01), WO2010/003836.2010) devise a kind of reconstruction sound field scheme, the acoustic pressure for the whole viewing area of perception distributes.Microphone layout on 3 dimension regular grids, is tieed up in regular grid (z=0) 3, (L/ Δ x) by the program ²individual microphone uniform layout is in grid point of intersection, and namely under rectangular coordinate system, the interval between each microphone horizontal axis, vertical coordinate axle is Δ x.In addition, when the sound source of rebuilding sound field is steady, also can carries out collection with single microphone one by one at diverse location and measure.

This technology network style gathers sound field, and the highest frequency rebuilding sound field is like this larger, and the number of required microphone is more, and sound field can be caused to distort.If by single microphone station acquisition one by one, then rebuild sound field region larger, gather consuming time more.

Wang Peili, Li Ji, paper " the near field acoustic holography technique study based on the neural network " (applied acoustics of Zhou Lili, 2010,29 (1): 58-62) and Mao Rongfu, Zhu Haichao, paper " reducing the research of measuring and counting near field acoustic holography (NAH) " (acoustic technique of Jinsong ZHANG, 2009,28 (3): 287-294) scheme used is all to whole sound field modeling based on local Information Monitoring.In this scenario, D, D< (L/ Δ x) ²individual microphone uniform layout is in whole viewing area, first use spatial coordinate location as the importation of training dataset, the sound field acoustic pressure that each microphone perceives is as the output of training dataset, by training dataset by the training of SVR model, obtain the mapping model between locus and sound field acoustic pressure; On this basis, using position, whole observation area as this mode input, thus perception goes out the acoustic pressure distribution of whole viewing area.When training pattern, also SVR can be replaced with neural network (ANN).

There is problem concerning study in ANN model, needs training data sample larger; SVR find different classes of between optimization classifying face, reflection be difference between heterogeneous data, but its training time is long, and can not reflect the characteristic of training data itself, the precision of the sound field sound pressure information of reconstruct is not high.

Summary of the invention

The present invention is directed to the proposition of above problem, and develop a kind of sound field information acquisition method for sound field synthesis.

The technical scheme that the present invention takes is as follows:

A kind of sound field information acquisition method for sound field synthesis: local sound pressure information in first acquisition plane, and the data collected local are as tranining database, set up the gauss hybrid models of space acoustic pressure and positional information, then carry out regression forecasting by the sound field acoustic pressure of this model to optional position in acquisition plane, thus obtain overall sound pressure information in plane.

Beneficial effect of the present invention:

The three-dimensional audio sound field cognitive method returned based on Gaussian Mixture that the present invention proposes, when relative error controls within 10%, even if the microphone number used when sound field gathers is only 1/4 of classic method use number, acoustic pressure distribution still effectively can be reconstructed.Compared with neural net method, support vector regression method, when acoustic pressure relative error requires degree of precision (as 10%), the present invention's microphone number used is less, has clear superiority.

Accompanying drawing explanation

Fig. 1 is based on the sound field information acquisition method functional block diagram of GMR.

Microphone distribution plan in the A of Fig. 2 region.

The modeling procedure figure of Fig. 3 gauss hybrid models.

Fig. 4 f=1.6kHz, reconstructs the relation curve of acoustic pressure relative error and gaussian component number M during N=40.

Fig. 5 f=1.6kHz, the relation curve of relative error and training data sample length of side N during M=15.

Fig. 6 N=20, during M=15, reconstruct sound pressure amplitude distribution plan in 2 meters × 2 meters region A.

Fig. 7 N=20, during M=15, desirable sound pressure amplitude distribution plan in 2 meters × 2 meters region A.

Fig. 8 N=20, during M=15, reconstruct sound pressure phase distribution plan in 2 meters × 2 meters region A.

Fig. 9 N=20, during M=15, desirable sound pressure phase distribution plan in 2 meters × 2 meters region A.

Figure 10 N=20, during M=15, to reconstruct acoustic pressure and desirable sound pressure amplitude relative error histogram in 2 meters × 2 meters region A.

Figure 11 N=20, during M=15, to reconstruct acoustic pressure and desirable sound pressure phase relative error histogram in 2 meters × 2 meters region A.

Figure 12 Gaussian Mixture returns (GMR) method, neural network (ANN) method, the relative error of support vector regression (SVR) method and the relation curve of training data sample length of side N.

In the viewing area of Figure 13 L rice × L rice, (L/ Δ x) that sound field perception uses ²individual microphone distribution plan.

Embodiment

Below in conjunction with accompanying drawing, the present invention will be further described:

The present invention's gray-scale map can illustrate technique effect of the present invention, and spy provides gray-scale map and Fig. 6 to Fig. 9 to allow auditor better understand technique effect of the present invention.

The functional block diagram of the present invention program as shown in Figure 1, the present invention replaces overall sound pressure information by local sound pressure information in acquisition plane, and the data collected local are as tranining database, set up the gauss hybrid models (GMM) of acoustic pressure and spatial information, then carry out regression forecasting by the sound field acoustic pressure of this model to optional position in acquisition plane, thus obtain overall sound pressure information in plane.

A kind of sound field information acquisition method for sound field synthesis: comprise plane microphone array, gauss hybrid models (Gaussian mixed model, GMM) parameter training module, regression forecasting module.The acoustic pressure data composing training database of specific discrete position in described plane microphone array acquisition plane, described GMM parameter training module uses the data in this tranining database to set up the gauss hybrid models of acoustic pressure and spatial information, the described regression forecasting module sound field acoustic pressure of this model to optional position in acquisition plane carries out regression forecasting, thus obtains sound field sound pressure information complete in acquisition plane.

Microphone array is classified as:

In a region A to be collected, evenly choose D diverse location and carry out acoustic pressure collection, as shown in Figure 2, wherein microphone spacing is greater than Δ x, the span 1<N<L/ Δ x of D=N × N, N, any position r in region _qand the multiple sound pressure level p (r in corresponding frequency domain _q, Ω) and a four-dimensional observation signal x (x can be formed _q, y _q, p _amp, p _phase), wherein p _ampand p _phasebe respectively amplitude and the phase place of multiple acoustic pressure; By this D spatial positional information and the multiple sound pressure information of correspondence, be configured to training data vector sequence X={x _n, n=1,2 ..., D}, namely X is the training dataset Dataset that D × 4 are tieed up.

The modeling method of gauss hybrid models is as follows:

Definition M is gaussian component number in GMM, w _ibe mixed weighting value, meet u _iit is mean value vector Σ _icovariance matrix E [(x-u _i) ^t(x-u _i)], i=1,2 ..., M; Gauss hybrid models parameter lambda describes and is defined as

λ={w _i,u _i,Σ _i}，i=1,2,…,M， (1)

Determine that in GMM, gaussian component number is M according to experiences such as the complexities of sound-filed simulation, the present invention's suggestion arranges M=15; GMM modeling procedure as shown in Figure 3.

Cluster is carried out to training dataset Dataset, obtains the initial parameter λ of gauss hybrid models ⁰; Concrete steps are as follows: first M individual 4 front in Dataset is tieed up observation signal x as initial cluster center cx ₁, cx ₂..., cx _m; Institute's other observation signal remaining, then according to the Euclidean distance d of they and these cluster centres _j, respectively they are distributed to the cluster the most similar to it, i.e. Euclidean distance min cluster, see formula (2); Then the cluster centre mean of average as new cluster of all observation signals in a certain cluster is calculated according to formula (3) _i; Constantly iterative (2) and (3), until the absolute value of the difference of mean square deviation before and after iteration of M the cluster centre calculated according to formula (4) is 10 ^-10within; Finally calculate λ with M the cluster data obtained ⁰, such as formula (5) ~ (7);

${\mathrm{mean}}_i=\frac{1}{{\mathrm{num}}_i}{\mathrm{\&Sigma;}}_{i}x,---\left(3\right)$

$E_i=\underset{j=1}{\overset{{\mathrm{num}}_i}{\mathrm{\&Sigma;}}}{|x_j-{\mathrm{mean}}_i|}^2,---\left(4\right)$

$w_i^0=\frac{{\mathrm{num}}_i}{D},---\left(5\right)$

$u_i^0={\mathrm{mean}}_i,---\left(6\right)$

${\mathrm{\&Sigma;}}_i^0=E\left[{(x-u_i^0)}^T(x-u_i^0)\right],---\left(7\right)$

Wherein, num _ibe the number of the observation data vector contained in i-th cluster, meanwhile, the data that formula (3) and (7) middle x belong to i-th cluster just participate in calculating;

Expectation maximization (EM) algorithm is used to determine the parameter lambda of GMM, the objective function Q of EM algorithm:

$Q(\mathrm{\&lambda;},{\mathrm{\&lambda;}}^{\&prime;})=\frac{1}{D}\underset{n=1}{\overset{D}{\mathrm{\&Sigma;}}}\log \left[\underset{k=0}{\overset{M}{\mathrm{\&Sigma;}}}{w}_k{f}_k\left(x_n\right)\right],---\left(8\right)$

Wherein, the model parameter that λ ' goes out for last iterative estimate, and in the objective function computation process of this EM iteration; λ is the model estimated parameter obtained after an EM iteration; f _ix () is four-dimensional Gaussian probability-density function

$f_i\left(x\right)=\frac{1}{{\left(2\mathrm{\&pi;}\right)}^2{\left|{\mathrm{\&Sigma;}}_i\right|}^{1/2}}\exp [-{(x-u_i)}^T{\mathrm{\&Sigma;}}_i^{-1}(x-u_i)/2],---\left(9\right)$

Estimate posterior probability step---asking for the posterior probability of training data under i-th gaussian component is:

$p(i_n=i|x_n,\mathrm{\&lambda;})=\frac{w_i{f}_i\left(x_n\right)}{\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{w}_k{f}_k\left(x_n\right)},---\left(10\right)$

Wherein, λ is then λ when first time iteration ⁰, afterwards then for expecting the model parameter that maximization steps produces;

Expectation maximization step---calculate the various estimates of parameters of Q:

${\stackrel{\&OverBar;}{w}}_i=\frac{1}{D}\underset{n=1}{\overset{D}{\mathrm{\&Sigma;}}}p(i_n=i|x_n,\mathrm{\&lambda;}),---\left(11\right)$

${\stackrel{\&OverBar;}{u}}_i=\frac{\underset{n=1}{\overset{D}{\mathrm{\&Sigma;}}}p(i_n=i|x_n,\mathrm{\&lambda;})x_n}{\underset{n=1}{\overset{D}{\mathrm{\&Sigma;}}}p(i_n=i|x_n,\mathrm{\&lambda;})},---\left(12\right)$

${\stackrel{\&OverBar;}{\mathrm{\&Sigma;}}}_i^2=\frac{\underset{n=1}{\overset{D}{\mathrm{\&Sigma;}}}p(i_n=i|x_n,\mathrm{\&lambda;})x_n^2}{\underset{n=1}{\overset{D}{\mathrm{\&Sigma;}}}p(i_n=i|x_n,\mathrm{\&lambda;})}-{\stackrel{\&OverBar;}{u}}_i^2,---\left(13\right)$

Wherein, be respectively the estimation of the weighted value of i-th component, mean value and covariance matrix; After expectation maximization process terminates, will with upgrade; Consider the stability of numerical evaluation, in expectation maximization process, after estimate covariance matrix, the diagonal element of covariance matrix is added a very little constant 10 ^-5;

Then, according to formula (11) ~ (13) calculating formula (8), if the difference of absolute value is more than 10 between the Q calculated and last iteration result ^-10, then repeat to estimate posterior probability step and expectation maximization step, until formula (8) result of calculation is less than 10 ^-10, now, value form vector be the GMM parameter lambda trained.

The method of regression forecasting module is:

To viewing area need arbitrarily the spatial position data of rebuilding acoustic pressure as input data characteristics vector X _in={ x _{in n}, n=1,2 ..., L}, using the acoustic pressure of rebuilding as exporting data characteristics vector X _out={ x _{out n}, n=1,2 ..., L}, it is then be input as X that Gaussian Mixture returns _in=x _{in n}condition under, X _outexpectation as output x _{out n}:

$x_{\mathrm{outn}}=E\left[X_{\mathrm{out}}\right|X_{\mathrm{in}}=x_{\mathrm{inn}},\mathrm{\&lambda;}]=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{m}_i\left(x_{\mathrm{inn}}\right)p\left(i\right|x_{\mathrm{inn}},\mathrm{\&lambda;}),---\left(14\right)$

In formula, be input as x _{in n}condition under x _{out n}expectation, and p (i|x _{in n}, λ) and be input data x _{in n}posterior probability under i-th gaussian component.

The beneficial effect that technical solution of the present invention is brought

For checking this programme feasibility, if in free space, point sound source is positioned at (2.6 meters, 0 meter, 1.5 meters) place, produce spherical wave sound field, acoustical signal frequency is f=1.6kHz.According to spatial sampling theorem, will gather this sound field in space, spaced furthest placed by microphone is 0.1 meter undistortedly.If the only region A(x axial coordinate scope-2 meters ~ 2 meters of 4 meters × 4 meters on viewing plane, y-axis coordinate range-2 meters ~ 2 meter, z-axis coordinate range 0 meter), then at least need placement 40 × 40 microphones, the present invention only uses N × N(N<40) individual microphone, the acoustic pressure distribution of arbitrfary point in the plane domain of 4 meters × 4 meters can be obtained.

When microphone is chosen, need to determine gaussian component number M in gauss hybrid models.Work as N=40, when selecting different M, region A reconstructs the relative error maximal value of acoustic pressure and desirable acoustic pressure as shown in Figure 4, and wherein square point curve is the relative error curve of reconstruct sound pressure amplitude, and circular point curve is corresponding reconstruct sound pressure phase then.As seen from Figure 4, when gaussian component number M is increased to after 15 gradually, the change of relative error tends towards stability, if now increase M again, its calculated amount can obviously increase, but performance remains unchanged substantially.The present embodiment chooses M=15 as gaussian component number.

As M=15, select different training data sample size N × N, relative error maximal value region A being reconstructed to acoustic pressure and desirable acoustic pressure performs an analysis, relative error and training data sample length of side N relation curve are as shown in Figure 5, wherein square point curve is the relative error curve of reconstruct sound pressure amplitude, and circular point curve is corresponding reconstruct sound pressure phase then.As seen from Figure 5, when training sample Parameter N is increased to after 20, the change of its relative error slowly, tends towards stability gradually, if now increase N again, its calculated amount can obviously increase, but performance remains unchanged substantially.Consider, choose N=20.

Work as N=20, during M=15, use Gaussian Mixture homing method to be reconstructed to 100 × 100 acoustic pressures be uniformly distributed on position in the A of region, reconstruct acoustic pressure and desirable acoustic pressure distribute as shown in figs. 6-9, and their relative error histogram is as shown in Figure 10 ~ Figure 11.From Fig. 6 ~ Fig. 9, the distribution of reconstruct acoustic pressure distributes close with desirable acoustic pressure.From Figure 10 ~ Figure 11, when N=20, M=15, the acoustic pressure using method in this paper to reconstruct and the amplitude relative error of desirable acoustic pressure are 5.5% to the maximum, and phase place relative error is 4.5% to the maximum.

The performance that the method propose the present invention and ANN and SVR method carry out the reconstruct of sound field acoustic pressure compares analysis, and they are under same sample size, and its reconstruct sound pressure amplitude relative error as shown in figure 12.In Figure 12, square point curve is the relation curve using GMR method reconstruct acoustic pressure amplitude and desired amplitude relative error maximal value and training data sample size, and wherein the size of training sample represents with N, and N × N is the data volume of training sample.Circular point curve and trigpoint curve then distinguish the corresponding situation using neural network, SVR method.When using neural network, select three layers of BP neural network, target error is 0.001; When using SVR, SVR Selecting parameter is: use ε-SVR, kernel function is gaussian kernel γ=0.05, v=0.5, penalty coefficient is 2.2, tolerable deviation 10 ^-3.As seen from Figure 12, under Small Sample Size, SVR has advantage, but reconstructed error will control within 10%; If sample size is identical, GMR method is slightly better than other two kinds of methods.

More than experiment shows, the three-dimensional audio sound field cognitive method returned based on Gaussian Mixture that the present invention proposes, when relative error controls within 10%, even if the microphone number used when sound field gathers is only 1/4 of classic method use number, acoustic pressure distribution still effectively can be reconstructed.Compared with ANN, SVR method, when acoustic pressure relative error requires degree of precision (as 10%), the present invention's microphone number used is less, has clear superiority.

The above; be only the present invention's preferably embodiment; but protection scope of the present invention is not limited thereto; anyly be familiar with those skilled in the art in the technical scope that the present invention discloses; be equal to according to technical scheme of the present invention and inventive concept thereof and replace or change, all should be encompassed within protection scope of the present invention.

The abbreviation that the present invention relates to and Key Term definition

3D:Three-Dimension, three-dimensional.

WFS:Wave Field Synthesis, sound field is synthesized.

GMR:Gaussian Mixture Regression, Gaussian Mixture returns.

GMM:Gaussian Mixture Model, gauss hybrid models.

ANN:Artificial Neural Networks, artificial neural network.

SVR:Support Vector Regression, Support vector regression.

EM:Expectation Maximization, expectation-maximization algorithm.

Claims (4)

1., for a sound field information acquisition method for sound field synthesis, it is characterized in that: comprise plane microphone array, gauss hybrid models parameter training module, regression forecasting module; Overall sound pressure information is replaced by local sound pressure information in plane microphone array acquisition plane, and the data collected local set up the gauss hybrid models parameter training module of acoustic pressure and spatial information as tranining database, then carry out regression forecasting by the sound field acoustic pressure of this model to optional position in acquisition plane, thus obtain overall sound pressure information in plane; The acoustic pressure data composing training database of specific discrete position in described plane microphone array acquisition plane, described gauss hybrid models parameter training module uses the data in this tranining database to set up the gauss hybrid models of acoustic pressure and spatial information, the described regression forecasting module sound field acoustic pressure of this model to optional position in acquisition plane carries out regression forecasting, thus obtains sound field sound pressure information complete in acquisition plane.

2. a kind of sound field information acquisition method for sound field synthesis according to claim 1, it is characterized in that: microphone array is classified as: in a region to be collected, evenly choose D diverse location and carry out acoustic pressure collection, wherein microphone spacing is greater than Δ x, D=N × N, the span of N is 1<N<L/ Δ x, any position r in region _qand the multiple sound pressure level p (r in corresponding frequency domain _q, Ω) and a four-dimensional observation signal x (x can be formed _q, y _q, p _amp, p _phase), wherein p _ampand p _phasebe respectively amplitude and the phase place of multiple acoustic pressure; By this D spatial positional information and the multiple sound pressure information of correspondence, be configured to training data vector sequence X={x _n, n=1,2 ..., D}, namely X is the training dataset Dataset that D × 4 are tieed up.

3. a kind of sound field information acquisition method for sound field synthesis according to claim 1, is characterized in that: the modeling method of gauss hybrid models is as follows:

Definition M is gaussian component number in GMM, w _ibe mixed weighting value, meet u _iit is mean value vector Σ _icovariance matrix E [(x-u _i) ^t(x-u _i)], i=1,2 ..., M; Gauss hybrid models parameter lambda describes and is defined as

λ={w _i,u _i,Σ _i}，i=1,2,…,M， (1)

Determine that in GMM, gaussian component number is M according to experiences such as the complexities of sound-filed simulation, M=15 is set; Cluster is carried out to training dataset Dataset, obtains the initial parameter λ of gauss hybrid models ⁰; Concrete steps are as follows: first M individual 4 front in Dataset is tieed up observation signal x as initial cluster center cx ₁, cx ₂..., cx _m; Institute's other observation signal remaining, then according to the Euclidean distance d of they and these cluster centres _j, respectively they are distributed to the cluster the most similar to it, i.e. Euclidean distance min cluster, see formula (2); Then the cluster centre mean of average as new cluster of all observation signals in a certain cluster is calculated according to formula (3) _i; Constantly iterative (2) and (3), until the absolute value of the difference of mean square deviation before and after iteration of M the cluster centre calculated according to formula (4) is 10 ^-10within; Finally calculate λ with M the cluster data obtained ⁰, such as formula (5) ~ (7);

${\mathrm{mean}}_i=\frac{1}{{\mathrm{num}}_i}{\mathrm{\&Sigma;}}_{i}x,---\left(3\right)$

$E_i=\underset{j=1}{\overset{{\mathrm{num}}_i}{\mathrm{\&Sigma;}}}{|x_j-{\mathrm{mean}}_i|}^2,---\left(4\right)$

$w_i^0=\frac{{\mathrm{num}}_i}{D},---\left(5\right)$

$u_i^0={\mathrm{mean}}_i,---\left(6\right)$

${\mathrm{\&Sigma;}}_i^0=E\left[{(x-u_i^0)}^T(x-u_i^0)\right],---\left(7\right)$

Wherein, num _ibe the number of the observation data vector contained in i-th cluster, the data that same up-to-date style (3) and (7) middle x belong to i-th cluster just participate in calculating;

(3) expectation-maximization algorithm is used to determine the parameter lambda of GMM, the objective function Q of EM algorithm:

$Q(\mathrm{\&lambda;},{\mathrm{\&lambda;}}^{\&prime;})=\frac{1}{D}\underset{n=1}{\overset{D}{\mathrm{\&Sigma;}}}\log \left[\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{w}_k{f}_k\left(x_n\right)\right],---\left(8\right)$

Wherein, the model parameter that λ ' goes out for last iterative estimate, and in the objective function computation process of this expectation maximization iteration; λ is the model estimated parameter obtained after an EM iteration; f _ix () is four-dimensional Gaussian probability-density function

$f_i\left(x\right)=\frac{1}{{\left(2\mathrm{\&pi;}\right)}^2{\left|{\mathrm{\&Sigma;}}_i\right|}^{1/2}}\exp [-{(x-u_i)}^T{\mathrm{\&Sigma;}}_i^{-1}(x-u_i)/2],---\left(9\right)$

Estimate posterior probability step---asking for the posterior probability of training data under i-th gaussian component is:

$p(i_n=i|x_n,\mathrm{\&lambda;})=\frac{w_i{f}_i\left(x_n\right)}{\underset{k=1}{\overset{M}{\mathrm{\&Sigma;}}}{w}_k{f}_k\left(x_n\right)},---\left(10\right)$

Wherein, λ is then λ when first time iteration ⁰, afterwards then for expecting the model parameter that maximization steps produces;

Expectation maximization step---calculate the various estimates of parameters of Q:

${\stackrel{\&OverBar;}{w}}_i=\frac{1}{D}\underset{n=1}{\overset{D}{\mathrm{\&Sigma;}}}p(i_n=i|x_n,\mathrm{\&lambda;}),---\left(11\right)$

${\stackrel{\&OverBar;}{u}}_i=\frac{\underset{n=1}{\overset{D}{\mathrm{\&Sigma;}}}p(i_n=i|x_n,\mathrm{\&lambda;})x_n}{\underset{n=1}{\overset{D}{\mathrm{\&Sigma;}}}p(i_n=i|x_n,\mathrm{\&lambda;})},---\left(12\right)$

${\stackrel{\&OverBar;}{\mathrm{\&Sigma;}}}_i^2=\frac{\underset{n=1}{\overset{D}{\mathrm{\&Sigma;}}}p(i_n=i|x_n,\mathrm{\&lambda;})x_n^2}{\underset{n=1}{\overset{D}{\mathrm{\&Sigma;}}}p(i_n=i|x_n,\mathrm{\&lambda;})}-{\stackrel{\&OverBar;}{u}}_i^2,---\left(13\right)$

Wherein, be respectively the estimation of the weighted value of i-th component, mean value and covariance matrix; After expectation maximization process terminates, will with upgrade; Consider the stability of numerical evaluation, in expectation maximization process, after estimate covariance matrix, the diagonal element of covariance matrix is added a very little constant 10 ^-5; Then, according to formula (11) ~ (13) calculating formula (8), if the difference of absolute value is more than 10 between the Q calculated and last iteration result ^-10, then repeat to estimate posterior probability step and expectation maximization step, until formula (8) result of calculation is less than 10 ^-10, now, value form vector be the GMM parameter lambda trained.

4. a kind of sound field information acquisition method for sound field synthesis according to claim 1, is characterized in that: the method for regression forecasting module is:

To viewing area need arbitrarily the spatial position data of rebuilding acoustic pressure as input data characteristics vector X _in={ x _{in n}, n=1,2 ..., L}, using the acoustic pressure of rebuilding as exporting data characteristics vector X _out={ x _{out n}, n=1,2 ..., L}, it is then be input as X that Gaussian Mixture returns _in=x _{in n}condition under, X _outexpectation as output x _{out n}:

$x_{\mathrm{outn}}=E\left[X_{\mathrm{out}}\right|X_{\mathrm{in}}=x_{\mathrm{inn}},\mathrm{\&lambda;}]=\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{m}_i\left(x_{\mathrm{inn}}\right)p\left(i\right|x_{\mathrm{inn}},\mathrm{\&lambda;}),---\left(14\right)$

In formula, be input as x _{in n}condition under x _{out n}expectation, and p (i|x _{in n}, λ) and be input data x _{in n}posterior probability under i-th gaussian component.