A sound source localization method, system, electronic device and medium

Abstract

The present invention is applicable to the field of computer technology, and provides a sound source localization method, system, electronic device and medium. The method comprises: obtaining a plurality of sub-frequency domain signals, and using a generalized cross-correlation algorithm to extract a first generalized cross-correlation feature of the sub-frequency domain signal under the same preset frequency band; establishing a mapping relationship between the generalized cross-correlation feature and the sound source weight, bringing the first generalized cross-correlation feature and the mapping relationship into a conditional expectation function of an expectation maximization algorithm, and performing multiple iterations to obtain a target sound source weight; determining a second generalized cross-correlation feature according to the first generalized cross-correlation feature and the target sound source weight, and obtaining a target controllable response power according to a controllable response power function and the second generalized cross-correlation feature; obtaining position information corresponding to the target controllable response power, and obtaining target position information; and the method solves the problem of inaccurate sound source localization caused by environmental noise interference and reverberation.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a sound source localization method, system, electronic equipment and medium. Background Art

[0002] With the development of sensor networks, computers and other related technologies, sound source localization technology has been widely used in different fields such as industrial inspection, video conferencing, and intelligent robots. Its principle is to extract relevant feature parameters from the received sound signal to solve the location of the sound source target. However, in actual application scenarios, factors such as environmental noise interference and reverberation will cause the sound source localization performance to decline, resulting in reduced accuracy of sound source localization. Summary of the invention

[0003] The present invention provides a sound source localization method, system, electronic equipment and medium to solve the problem of inaccurate sound source localization caused by factors such as environmental noise interference and reverberation in the prior art.

[0004] The sound source localization method provided by the present invention comprises:

[0005] Using a plurality of sensors to collect target sound source signals generated by the same sound source, and performing spectrum transformation on the target sound source signals collected by the plurality of sensors respectively, to obtain a plurality of target frequency domain signals;

[0006] Dividing each of the target frequency domain signals into a plurality of sub-frequency domain signals according to a plurality of preset frequency bands, and extracting first generalized cross-correlation features of the sub-frequency domain signals in the same preset frequency band by using a generalized cross-correlation algorithm;

[0007] Establishing a mapping relationship between generalized cross-correlation features and sound source weights, bringing the first generalized cross-correlation features and the mapping relationship into a conditional expectation function of an expectation-maximization algorithm, and performing multiple iterations to obtain a target sound source weight;

[0008] Determining a second generalized mutual correlation feature according to the first generalized mutual correlation feature and the target sound source weight, and acquiring a target controllable response power according to a controllable response power function and the second generalized mutual correlation feature;

[0009] The preset candidate area is divided into a plurality of grid areas, and the position information corresponding to the target controllable response power is acquired according to the grid areas to obtain the target position information.

[0010] Optionally, the extracting the first generalized cross-correlation feature of the sub-frequency domain signal in the same preset frequency band by using a generalized cross-correlation algorithm includes:

[0011] Acquire a cross-power spectrum of the sub-frequency domain signal in the same preset frequency band, and process the cross-power spectrum using a generalized cross-correlation algorithm to obtain a first generalized cross-correlation feature;

[0012] The mathematical expression of the cross power spectrum is:

[0013]

[0014] Among them, Y m (ω) is the Fourier transform of the target sound source signal received by sensor m, is the conjugate of the Fourier transform of the target sound source signal received by sensor n, * is the conjugate, m and n are the numbers of the sensors, ω is the angular frequency, G mn (ω) is the cross power spectrum of sensor n and sensor m;

[0015] The mathematical expression of the first generalized cross-correlation feature is:

[0016]

[0017] Among them, R mn (τ) is the first generalized cross-correlation feature of sensor n and sensor m under the delay value τ, ψ mn is the phase transformation weight function of sensor n and sensor m, j is an imaginary number, ω is the angular frequency, and τ is the delay value of the target sound source signal;

[0018] in,

[0019] Optionally, the mapping relationship between the generalized cross-correlation feature and the sound source weight includes:

[0020]

[0021] In the formula, τ l is the delay value of the preset frequency band l, r l is the delay value τ l The generalized cross-correlation characteristics of the preset frequency band l under , L is the total number of preset frequency bands, l is the label of the preset frequency band, is the generalized cross-correlation characteristic of the sound source in the preset frequency band l, λ l ∈[0,1],λ l is the sound source weight of the preset frequency band l, n l is the noise component of the generalized cross-correlation characteristic of the preset frequency band l, n l are independent zero-mean Gaussian variables.

[0022] Optionally, before bringing the first generalized mutual correlation feature and the mapping relationship into a conditional expectation function of an expectation-maximization algorithm, the method further includes:

[0023] The full-band generalized cross-correlation feature is obtained according to the mapping relationship. The mathematical expression of the full-band generalized cross-correlation feature R is:

[0024]

[0025] According to the mapping relationship, complete data consisting of a mean and a covariance matrix is ​​obtained, and the complete data The mathematical expression is:

[0026]

[0027] Among them, r1 is the generalized cross-correlation feature of the preset frequency band 1, r2 is the generalized cross-correlation feature of the preset frequency band 1, and r L is the generalized cross-correlation feature of the preset frequency band L, T is the transpose of the matrix, and the variance of the noise in the preset frequency band l is The variance of the full-band noise is σ 2 ;

[0028] The mathematical expression of the mean μ of complete data is:

[0029]

[0030] Among them, λ1 is the sound source weight of the preset frequency band 1, λ2 is the sound source weight of the preset frequency band 2, and λ L is the sound source weight of the preset frequency band L, is the generalized cross-correlation feature of the sound source in preset frequency band 1, is the generalized cross-correlation feature of the sound source in preset frequency band 2, is the generalized cross-correlation characteristic of the sound source in the preset frequency band L;

[0031] The mathematical expression of the covariance matrix Q of the complete data is:

[0032]

[0033] Among them, diag is a diagonal matrix function, β l is the noise power of the preset frequency band l;

[0034]

[0035] Optionally, the step of bringing the first generalized mutual correlation feature and the mapping relationship into a conditional expectation function of an expectation-maximization algorithm and performing multiple iterations to obtain a target sound source weight includes:

[0036] Substituting the first generalized mutual correlation feature, the full-band generalized mutual correlation feature and the complete data into the conditional expectation function of the expectation-maximization algorithm, and performing multiple iterations, and obtaining the conditional expectation result after the iterations are completed;

[0037] Obtaining the target sound source weight according to the conditional expected result;

[0038] The parameters of the conditional expectation function θ=[θ1,θ2,…,θ l ],θ l =[τ l ,λ l ],

[0039] The mathematical expression of the conditional expectation function of the expectation maximization algorithm is:

[0040]

[0041] Where i is the number of iterations, θ i is the parameter of the conditional expectation function of the i-th iteration, is the conditional probability density, is the conditional expectation function, C is the preset constant, Delay value The i-th iteration generalized cross-correlation feature of the preset frequency band l under is the i-th iteration delay value of the preset frequency band l;

[0042] The mathematical expression of the conditional expectation result is

[0043]

[0044] The sound source weight of the last iteration of the preset frequency band The mathematical expression is:

[0045]

[0046] in, Delay value The last iteration generalized cross-correlation feature of the preset frequency band l under is the last iteration delay value of the preset frequency band l, s is the final iteration number, is the transpose of the generalized cross-correlation feature of the sound source in the preset frequency band l;

[0047] The mathematical expression of the target sound source weight k is:

[0048]

[0049] Optionally, the step of establishing a mapping relationship between generalized cross-correlation features and sound source weights, and bringing the mapping relationship into a conditional expectation function of an expectation-maximization algorithm comprises:

[0050] For the generalized cross-correlation feature r l Correction is performed to obtain the corrected generalized cross-correlation characteristics;

[0051] The mapping relationship is updated according to the modified generalized mutual correlation feature, and the updated mapping relationship is introduced into the conditional expectation function of the expectation-maximization algorithm.

[0052] Optionally, acquiring a target controllable response power according to the controllable response power function and the second generalized cross-correlation feature comprises:

[0053] Acquire a controllable response power according to a controllable response power function and the second generalized cross-correlation feature, and determine the maximum controllable response power as a target controllable response power, wherein the second generalized cross-correlation feature is a generalized cross-correlation feature of all sensors in a preset frequency band corresponding to the target sound source weight;

[0054] The mathematical expression of controllable response power is:

[0055]

[0056] Among them, m and n are the numbers of sensors. is the second generalized cross-correlation feature of sensor m and sensor n, τ mn (x) is the time delay value between sensor m and sensor n, and x is the grid point.

[0057] The present invention also provides a sound source localization system, comprising:

[0058] A frequency domain signal module is used to collect target sound source signals generated by the same sound source using a plurality of sensors, and perform spectrum transformation on the target sound source signals collected by the plurality of sensors respectively to obtain a plurality of target frequency domain signals;

[0059] A feature extraction module, used to divide each of the target frequency domain signals into a plurality of sub-frequency domain signals according to a plurality of preset frequency bands, and extract a first generalized cross-correlation feature of the sub-frequency domain signals in the same preset frequency band by using a generalized cross-correlation algorithm;

[0060] A target weight module is used to establish a mapping relationship between generalized mutual correlation features and sound source weights, bring the first generalized mutual correlation features and the mapping relationship into the conditional expectation function of the expectation-maximization algorithm, and perform multiple iterations to obtain a target sound source weight;

[0061] a target power module, configured to determine a second generalized mutual correlation feature according to the first generalized mutual correlation feature and the target sound source weight, and to obtain a target controllable response power according to a controllable response power function and the second generalized mutual correlation feature;

[0062] The target positioning module is used to divide the preset candidate area into a plurality of grid areas, obtain the position information corresponding to the controllable response power of the target according to the grid areas, and obtain the target position information. The frequency domain signal module, the feature extraction module, the target weight module, the target power module and the target positioning module are connected.

[0063] The present invention also provides an electronic device, comprising: a processor and a memory;

[0064] The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the electronic device executes the sound source localization method.

[0065] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the sound source localization method as described above is implemented.

[0066] As described above, the present invention provides a sound source localization method, system, electronic device and medium, which have the following beneficial effects: first, by dividing the target sound source signal collected by the sensor into a plurality of sub-frequency domain signals, and extracting the first generalized cross-correlation feature of the sub-frequency domain signal, the target sound source weight is obtained by the first generalized cross-correlation feature and the expectation maximization algorithm, and the target controllable response power is determined according to the second generalized cross-correlation feature corresponding to the target sound source and the controllable response power function, thereby realizing the positioning of the sound source and solving the problem of inaccurate sound source positioning caused by factors such as environmental noise interference and reverberation in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0068] Figure 1 is a schematic flow chart of a sound source localization method in an embodiment of the present invention;

[0069] Figure 2 is different from the embodiment of the present invention o Function graph;

[0070] Figure 3 is the full-band generalized cross-correlation characteristic before correction in the embodiment of the present invention;

[0071] Figure 4 is the corrected full-band generalized cross-correlation feature in the embodiment of the present invention;

[0072] Figure 5 It is the algorithm in each embodiment of the present invention in T 60 =0.2s, r = 0.1m performance diagram;

[0073] Figure 6 It is the algorithm in each embodiment of the present invention in T 60 =0.2s, r = 0.5m performance diagram;

[0074] Figure 7 It is the algorithm in each embodiment of the present invention in T 60 =0.8s, r=0.1m performance diagram;

[0075] Figure 8 It is the algorithm in each embodiment of the present invention in T 60 =0.8s, r = 0.5m performance diagram;

[0076] Fig. 9 It is the algorithm in each embodiment of the present invention in T 60 = Positioning success rate curve under 0.2s;

[0077] Fig.10 It is the algorithm in each embodiment of the present invention in T 60 = Positioning success rate curve under 0.8s;

[0078] Fig.11 is the average error comparison of the performance simulation of each algorithm in the embodiment of the present invention;

[0079] Fig.12 is a comparison of standard deviations of the performance simulations of the algorithms in the embodiments of the present invention;

[0080] Fig.13 is the relative position of the microphone and the sound source in the embodiment of the present invention;

[0081] Fig.14 are three sound source power spectra in the embodiments of the present invention;

[0082] Fig.15 is the signal-to-noise ratio of the all-pass signal in the embodiment of the present invention;

[0083] Fig.16 is the signal-to-noise ratio of the bandpass signal in the embodiment of the present invention;

[0084] Fig.17 is an average positioning error curve of the outdoor test results in an embodiment of the present invention;

[0085] Fig.18 Schematic diagram of a module of a sound source localization system in an embodiment of the present invention. DETAILED DESCRIPTION

[0086] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0087] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0088] In order to illustrate the technical solution of the present invention, a specific embodiment is provided below for illustration.

[0089] Figure 1 It is a flowchart of a sound source localization method provided in one embodiment of the present invention.

[0090] like Figure 1 As shown, the above-mentioned sound source localization method includes steps S110-S150:

[0091] S110, using a plurality of sensors to collect target sound source signals generated by the same sound source, and performing spectrum transformation on the target sound source signals collected by the plurality of sensors respectively, to obtain a plurality of target frequency domain signals;

[0092] S120, dividing each target frequency domain signal into a plurality of sub-frequency domain signals according to a plurality of preset frequency bands, and extracting first generalized cross-correlation features of the sub-frequency domain signals in the same preset frequency band by using a generalized cross-correlation algorithm;

[0093] S130, establishing a mapping relationship between the generalized cross-correlation feature and the sound source weight, bringing the first generalized cross-correlation feature and the mapping relationship into the conditional expectation function of the expectation-maximization algorithm, and performing multiple iterations to obtain the target sound source weight;

[0094] S140, determining a second generalized mutual correlation feature according to the first generalized mutual correlation feature and the target sound source weight, and acquiring a target controllable response power according to the controllable response power function and the second generalized mutual correlation feature;

[0095] S150, dividing the preset candidate area into a plurality of grid areas, acquiring position information corresponding to the target controllable response power according to the grid areas, and obtaining target position information.

[0096] In step S110 of the present embodiment, the same sound source includes but is not limited to the sound source generated in the actual scene, such as the sound source generated by recording, human movement, human speech, bird calls, rain, thunder, etc., and several sensors respectively collect the target sound source signals, and then perform spectrum conversion on the target sound source signals collected by the several sensors to obtain several target frequency domain signals. Specifically, the sound source signals collected by the several sensors can be Fourier transformed to obtain the target frequency domain signals corresponding to each sensor. The sensor includes a microphone, and the position of the sensor and the number of sensors can be set according to actual needs. It is also possible to first make a preliminary estimate of the position of the sound source that needs to be located, that is, the target sound source, and set the position of the sensor according to the preliminary estimation result, for example, with the preliminary estimated position as the center of the circle and a fixed value as the radius, and evenly set the sensors around it. The fixed value can be set according to actual needs.

[0097] In step S120 of the present embodiment, the frequency range set of all preset frequency bands includes the frequency range of the sound source. The number of preset frequency bands and the frequency band range can be set according to the frequency of the sound source. For example, if the frequency range of the sound source is 1000Hz-1500Hz, 800-1600Hz can be used as the frequency range of all preset frequency bands, and 800-1600Hz can be evenly divided into 4 preset frequency bands with a bandwidth of 200Hz; 900-1600Hz can also be used as the frequency range of all preset frequency bands, and 900-1600Hz can be evenly divided into 7 preset frequency bands with a bandwidth of 100Hz. Each target frequency domain signal is divided into several sub-frequency domain signals according to several preset frequency bands, that is, the target frequency domain signal collected by each sensor is divided into several sub-frequency domain signals corresponding to each sensor according to several preset frequency bands, and the same sub-frequency domain signal belongs to the same preset frequency band. The generalized cross-correlation algorithm is used to extract the first generalized cross-correlation features of the sub-frequency domain signals under the same preset frequency band, that is, the generalized cross-correlation algorithm is used to extract the first generalized cross-correlation features of the sub-frequency domain signals corresponding to different sensors under the same preset frequency band.

[0098] In one embodiment, extracting the first generalized cross-correlation feature of the sub-frequency domain signal in the same preset frequency band using the generalized cross-correlation algorithm includes: obtaining the cross-power spectrum of the sub-frequency domain signal in the same preset frequency band, and processing the cross-power spectrum using the generalized cross-correlation algorithm to obtain the first generalized cross-correlation feature;

[0099] The mathematical expression of the cross power spectrum is:

[0100]

[0101] Among them, Y m (ω) is the Fourier transform of the target sound source signal received by sensor m, is the conjugate of the Fourier transform of the target sound source signal received by sensor n, * is the conjugate, m and n are the numbers of the sensors, ω is the angular frequency, G mn (ω) is the cross power spectrum of sensor n and sensor m;

[0102] The mathematical expression of the first generalized cross-correlation feature is:

[0103]

[0104] Among them, R mn (τ) is the first generalized cross-correlation feature of sensor n and sensor m under the delay value τ, ψ mn is the phase transformation weight function of sensor n and sensor m, j is an imaginary number, ω is the angular frequency, and τ is the delay value of the target sound source signal;

[0105] in,

[0106] In step S130 of this embodiment, the mapping relationship between the generalized cross-correlation feature and the sound source weight includes:

[0107]

[0108] In the formula, τ l is the delay value of the preset frequency band l, r l is the delay value τ l The generalized cross-correlation characteristics of the preset frequency band l under , L is the total number of preset frequency bands, l is the label of the preset frequency band, is the generalized cross-correlation characteristic of the sound source in the preset frequency band l, λ l ∈[0,1],λ l is the sound source weight of the preset frequency band l, n l is the noise component of the generalized cross-correlation characteristic of the preset frequency band l, n l are independent zero-mean Gaussian variables.

[0109] Before bringing the first generalized mutual correlation feature and the mapping relationship into the conditional expectation function of the expectation maximization algorithm, the method also includes: obtaining the full-band generalized mutual correlation feature according to the mapping relationship, and obtaining complete data consisting of the mean and covariance matrix according to the mapping relationship.

[0110] The mathematical expression of the full-band generalized cross-correlation feature R is:

[0111]

[0112] Complete data The mathematical expression is:

[0113]

[0114] Among them, r1 is the generalized cross-correlation feature of the preset frequency band 1, r2 is the generalized cross-correlation feature of the preset frequency band 1, and r L is the generalized cross-correlation feature of the preset frequency band L, T is the transpose of the matrix, and the variance of the noise in the preset frequency band l is The variance of the full-band noise is σ 2 ;

[0115] The mathematical expression of the mean μ of complete data is:

[0116]

[0117] Among them, λ1 is the sound source weight of the preset frequency band 1, λ2 is the sound source weight of the preset frequency band 2, and λ L is the sound source weight of the preset frequency band L, is the generalized cross-correlation feature of the sound source in preset frequency band 1, is the generalized cross-correlation feature of the sound source in preset frequency band 2, is the generalized cross-correlation characteristic of the sound source in the preset frequency band L;

[0118] The mathematical expression of the covariance matrix Q of complete data is:

[0119]

[0120] Among them, diag is a diagonal matrix function, β l is the noise power of the preset frequency band l;

[0121]

[0122] In one embodiment, the first generalized mutual correlation feature and the mapping relationship are brought into the conditional expectation function of the expectation maximization algorithm, and multiple iterations are performed to obtain the target sound source weight, including: obtaining the full-band generalized mutual correlation feature and complete data consisting of a mean and a covariance matrix according to the mapping relationship, bringing the first generalized mutual correlation feature, the full-band generalized mutual correlation feature and the complete data into the conditional expectation function of the expectation maximization algorithm, and performing multiple iterations, and after the iterations are completed, obtaining the conditional expectation result; obtaining the target sound source weight according to the conditional expectation result;

[0123] The parameters of the conditional expectation function θ=[θ1,θ2,…,θ l ],θ l =[τ l ,λ l ],

[0124] The mathematical expression of the conditional expectation function of the expectation maximization algorithm is:

[0125]

[0126] Where i is the number of iterations, θi is the parameter of the conditional expectation function of the i-th iteration, is the conditional probability density, is the conditional expectation function, C is the preset constant, Delay value The i-th iteration generalized cross-correlation feature of the preset frequency band l under is the i-th iteration delay value of the preset frequency band l;

[0127] In the first iteration, the first generalized cross-correlation feature is substituted into the conditional expectation function of the expectation maximization algorithm mentioned above; by using the conditional expectation function U(θ,θ i ) is maximized to estimate the parameter θ, which is equivalent to performing simultaneous parameter θ estimation on multiple frequency bands. l Estimation can effectively reduce the computational complexity.

[0128] Using the conditional expectation result of the joint Gaussian random variable, the mathematical expression of the conditional expectation result is

[0129]

[0130] The last iteration sound source weight of preset frequency band l The mathematical expression is:

[0131]

[0132] The last iteration delay value of preset frequency band l The mathematical expression is:

[0133]

[0134] in, Delay value The last iteration generalized cross-correlation feature of the preset frequency band l under is the last iteration delay value of the preset frequency band l, s is the final iteration number, is the transpose of the generalized cross-correlation feature of the sound source in the preset frequency band l; for The conjugate transpose of ;

[0135] At the beginning of the iteration, the θ of each frequency band is l Assign any initial value, when δ is the preset discriminant value, or the iteration is terminated when the number of iterations is reached.

[0136] Finally The largest one is selected from the parameter set, which is the generalized cross-correlation feature of the optimal frequency band. The sound source weight corresponding to the generalized cross-correlation feature of the optimal frequency band is the target sound source weight. The mathematical expression of the target sound source weight k is:

[0137]

[0138] In one embodiment, a mapping relationship between a generalized mutual correlation feature and a sound source weight is established, and the mapping relationship is introduced into a conditional expectation function of an expectation maximization algorithm, comprising: l The modified generalized mutual correlation feature is obtained; the mapping relationship is updated according to the modified generalized mutual correlation feature, and the updated mapping relationship is brought into the conditional expectation function of the expectation maximization algorithm. Replace r in each frequency band l , participate in each iteration, and finally select the optimal frequency band Substitute into the conditional expectation function of the expectation-maximization algorithm.

[0139] The mathematical expression of the modified generalized cross-correlation feature is:

[0140]

[0141] in, is the generalized cross-correlation feature after correction of the preset frequency band l, E{r l r o} is r l With r o expectations, for With r o The expectation, E{n l r o} is n l With r o expectations.

[0142] See also Figure 2 , r o is a function. The size of the main lobe width will affect the positioning result. A larger width will make the spatial spectrum of the controllable response power (SRP) smoother and improve the robustness. On the contrary, a smaller width will make the peak sharper and improve the resolution. o The waveform will help reduce the error of sound source localization. In the method of generating generalized cross-correlation features of each frequency band, Compared to r l The waveform is more flexible and controllable. o This affects the waveform of the sub-band generalized cross-correlation characteristics in the time domain; on the other hand, the r of different frequency bands in the frequency domain o It plays a filtering role, so r o The power spectrum distribution is relatively concentrated. Figure 2 For different oThe function curves are the time domain waveform comparisons of the generalized cross correlation (GCC) function, sinc function, rectangular window function, and Gaussian function. Figure 2 It can be seen that under the condition of achieving roughly similar main peak widths, the sinc function still has some side lobes in the time domain, which is not conducive to positioning compared to SRP, while the frequency domain function corresponding to the rectangular window function is sinc, which also has more side lobe interference, which is not conducive to effective filtering.

[0143] In order to minimize the impact of side lobes, the appropriate r o It should also satisfy the following properties:

[0144] r o The function has no side lobes, that is, for any |t1|<|t2|,|r o (t1)|>|r o (t2)|established.

[0145] Therefore, r o The Gaussian function is selected, and its waveform function has no side lobes in both the frequency domain and the time domain, which is suitable as an ideal function r o , which is expressed as follows:

[0146]

[0147] In the formula, f s is the signal frequency, and d is the grid scale. The initial value of the parameter τ comes from the peak of the full-band cross-correlation function. Each time the parameter value is updated, r is used. o Reconstruct the generalized cross-correlation features of each frequency band, where n is the number of discrete sampling points.

[0148] Through the above analysis, it can be observed that under this condition, the generalized cross-correlation feature has the following properties: 1) It extracts the time delay information of the signal component of the frequency band GCC in the interval; 2) It is only related to this frequency domain interval and has nothing to do with the components outside the interval; 3) It plays an important role in improving the positioning effect of the SRP function.

[0149] The full-band GCC under low signal-to-noise ratio has more side lobes and pseudo peaks, which will cause a ripple effect in SRP, making the SRP spatial spectrum more susceptible to interference from other factors, resulting in positioning failure.

[0150] See also Figure 3 and Figure 4By comparing the full-band GCC and the modified sub-band GCC in the actual received signal, it can be observed that the generalized cross-correlation feature based on the expected maximum algorithm (EM-GCC) improves the delay accuracy and reduces the pseudo-peak amplitude, which plays a role in waveform smoothing. Since the peak range of the generalized cross-correlation feature based on phase transformation (GCC-PHAT) is narrow, random values ​​will be used when calculating the controllable response power (SRP) at the grid point, and EM-GCC ensures that it samples at the grid to obtain an available maximum value, so that the SRP can be generated normally at the grid point.

[0151] In step S140 of the present embodiment, a second generalized mutual correlation feature is determined according to the first generalized mutual correlation feature and the target sound source weight, the second generalized mutual correlation feature is a generalized mutual correlation feature of a preset frequency band corresponding to the target sound source weight, and according to the controllable response power function and the second generalized mutual correlation feature, obtaining the target controllable response power includes: obtaining the controllable response power according to the controllable response power function and the second generalized mutual correlation feature, and determining the maximum controllable response power as the target controllable response power, the second generalized mutual correlation feature is a generalized mutual correlation feature of all sensors in the preset frequency band corresponding to the target sound source weight;

[0152] The mathematical expression of controllable response power is:

[0153]

[0154] Among them, m and n are the numbers of sensors. is the second generalized cross-correlation feature of sensor m and sensor n, τ mn (x) is the time delay value between sensor m and sensor n, and x is the grid point.

[0155] In step S150 of this embodiment, the preset candidate area is divided into a number of grid areas, and the position information corresponding to the target controllable response power is obtained according to the grid areas to obtain the target position information; thereby realizing the positioning of the sound source and solving the problem of inaccurate sound source positioning caused by factors such as environmental noise interference and reverberation in the prior art.

[0156] The sound source localization method provided by the embodiment of the present invention. The SRP positioning algorithm with less information loss directly uses the original data for positioning calculation, and accordingly proposes to first perform the expectation maximization algorithm processing and then obtain the controllable response power on this basis, that is, the sound source localization method (EM-SRP) in the present invention. First, the waveform is modulated using the Gaussian function on the basis of the reconstructed sub-band GCC, and then the optimal frequency band is selected by comparing the weighted coefficients of each sub-band GCC, and finally applied to the SRP positioning algorithm to solve the sound source position. While retaining the delay information, EM-SRP can effectively reduce the pseudo-peaks of the function, and the peak width corresponds to the grid scale. EM-SRP was compared with other commonly used methods through data simulation and outdoor experiments, and still maintained good accuracy and stability under large-scale grids.

[0157] In one embodiment, the performance of the sound source localization method (EM-SPR) of this scheme is analyzed based on data simulation using real recordings as the sound source. The experiments involved in the entire process of analyzing sound source localization based on data simulation are all carried out under two-dimensional positioning. The simulation scene is a 4×7m rectangular room with 6 sensors. The 6 microphones are located at [0,0], [0,4], [0,7], [4,0], [4,4], [4,7], and the sound speed c is set to 340m / s. In the same scene, the sound source uses a 3s pure male voice with a sampling rate of 44.1kHz. The performance of different algorithms is compared by the root mean square error (RMSE) between the real sound source position and the estimated sound source position. The signal processing uses a window function with a length of 4096 and a 50% overlap. A total of 93 frames of source audio data are obtained, and Gaussian white noise is added according to the signal-to-noise ratio (SNR∈{-10,-5,0,5,10}dB). The frequency range of the preset frequency band of the sound source localization method (EM-SPR) in this scheme is 0-1500 Hz. The target frequency domain signal collected by each sensor is divided into 8 sub-frequency domain signals corresponding to each sensor, and 48 sub-frequency domain signals are obtained. The bandwidth of each preset frequency band is 200 Hz, the total number of preset frequency bands is L=8, and the number of iterations η=5.

[0158] Under different signal-to-noise ratios, two different reverberation times (T 60 = 0.2s and T 60 =0.6s) and compared the RMS errors of various algorithms at different spatial grid resolutions (r1 = 0.1m, r2 = 0.5m).

[0159] See also Figure 5-8 , Figure 5-8 The horizontal axis is the signal-to-noise ratio (SNR), and the vertical axis is the root mean square error (RMSE) of performance accuracy. The performance comparison of each algorithm is given through simulation data. Figure 5 and Figure 6 It can be seen that under low reverberation and high signal-to-noise ratio conditions, all algorithms perform well. For smaller grid resolutions, it can be observed that the positioning error of the phase-transformed controllable power response (PHAT-SRP) can also be consistent with other algorithms; however, for larger grid resolutions, the positioning deviations of each algorithm increase, and the performance of the traditional PHAT-SRP decreases most significantly, but EM-SRP is always below the curve, especially under low signal-to-noise ratios, the positioning deviation of EM-SRP is significantly lower than that of other algorithms. SVD-SPR is the singular value decomposition controllable response power, WSVD-SRP is the weighted singular value decomposition controllable response power, SCOT-SRP is the smoothed coherent controllable response power, and ROTH-SRP is the Wiener filter impulse response controllable response power. Figure 7 and Figure 8 The root mean square error under high reverberation is given. The SRP-PHAT method will have irrelevant and large deviation estimates. This is mainly because many spatial maxima generated by the wrong position make the spatial peak of the true sound source position less obvious, thus making the search for the global peak wrong. With the improvement of the signal-to-noise ratio, the positioning deviation of each algorithm gradually decreases, but it is always above EM-SRP. Therefore, under the same grid resolution, the EM-SRP method has achieved better improvements in both noise resistance and reverberation resistance.

[0160] See also Figure 9-10 , Figure 9-10 The horizontal axis is the signal-to-noise ratio, and the vertical axis is the positioning success rate (Px). In order to further verify the effectiveness of the sound source localization method in this paper, the positioning success rate is used to evaluate the robustness of each algorithm. s Indicates the actual sound source position, Represents the sound source position estimation obtained by the k-th frame signal. If the estimated error is not greater than ε, that is, ε is the preset discrimination value, then the positioning estimation is successful, otherwise it is unsuccessful. The mathematical expression of the positioning success rate is:

[0161]

[0162] Where N suc Indicates the number of successful positioning, N total Indicates the total number of positioning times. is the positioning success rate. The closer the positioning error distribution curve is to the upper right corner, that is, the closer it is to 100%, the fewer the divergent results are estimated to be, and the better the robustness of the corresponding algorithm. The positioning success rate curves of different algorithms are shown in Figure 2. Figure 9-10 As shown, in the reverberation and Gaussian white noise environment, the grid size is 0.3m and the predicted value The positioning accuracy of the proposed method is better than that of the SRP-PHAT algorithm and other algorithms; especially in low signal-to-noise ratio environments, the positioning accuracy of the EM-SRP algorithm is significantly improved. Simulation results show that the proposed method is more robust to reverberation and low signal-to-noise ratio.

[0163] See also Figure 11-12 , Figure 11-12 The statistical positioning error under different signal-to-noise ratios can be observed. Fig.11 The horizontal axis is the signal-to-noise ratio, and the vertical axis is the average error of performance accuracy. Fig.12 The horizontal axis is the signal-to-noise ratio, and the vertical axis is the standard deviation of performance accuracy. As the signal-to-noise ratio decreases, the signal component is gradually obliterated by the noise, and the positioning error gradually increases. The average error and standard deviation of EM-SRP are always kept to a minimum, and stable positioning performance is maintained above -10dB, which is suitable for environments with lower signal-to-noise ratios.

[0164] In one embodiment, an outdoor test experiment is used to analyze the performance of the sound source localization method (EM-SPR) of this solution, see Figure 13-17 . Fig.13 The horizontal and vertical coordinates in are the position coordinates of the sensor and the sound source, node represents the node, and source represents the sound source. The outdoor test experiment was conducted in an open-air grassland, with a total of 7 wireless microphone nodes deployed. Each node contains a microphone for signal acquisition, a Wi-Fi module for data transmission, and a high-precision differential self-positioning and global time synchronization module. The node layout spacing is greater than 20 meters, and the detection search range is set to a 200×200×10 area. There are undulating terrain in the layout area. A portable speaker box was used to play three sounds (sound sources) at random in 12 locations, including:

[0165] 1) Gaussian signal (SG for short), representing a general broadband sound source;

[0166] 2) Motor vehicle horn (SV for short), which represents a typical application of urban noise source monitoring;

[0167] 3) Birdsong (SB for short), representing a typical application of field environment monitoring;

[0168] Seven sensors are used to collect the sound source signals generated by these three sound sources, and the sound source localization method (EM-SPR) of this scheme is used to process them. For the plane positions of sensors and sound sources, please refer to Fig.13 The sampling frequency of the system is 10000Hz, the system sampling synchronization error is less than 1us, and the node self-positioning error is less than 0.1m. For the three types of sound sources, all available signal sources are divided into 1242 frames, and the frame length is 2s for 7-channel recording signals.

[0169] In outdoor environments, the actual received signal-to-noise ratio of each node varies greatly with the distance from the sound source and the shielding conditions. The signal components are extracted by spectral subtraction, and the signal-to-noise ratio is estimated at different distances. The signal at 30m from the sound source is used to estimate the power spectrum of the background noise in the received signal and the three sound source signals. Please refer to Fig.14 , Fig.14 The horizontal axis is frequency, and the vertical axis is power spectrum density. The environmental noise is concentrated below 1500Hz. For all types of sound sources, the frequency range of 1500-3500Hz is uniformly used as the signal passband. The signal-to-noise ratio curves of the full-pass signal (0-5000Hz) and the band-pass signal (1500-3500Hz) are as follows: Fig.15 and Fig.16 As shown, Figure 15-16 The horizontal axis is the grid spacing, and the vertical axis is the signal-to-noise ratio. For the three types of sound sources, the passband range significantly improves the signal-to-noise ratio (20-30dB).

[0170] The parameters of the singular value decomposition (SVD-SPR) and weighted singular value decomposition (WSVD-SPR) methods are 1024 (spectral window length) and 128 (frequency hopping), and the total number of frequency bands is preset to be L = 76. The algorithm in this paper uses a bandwidth of 200 Hz to divide the frequency band from 1500 to 3500 Hz, the total number of frequency bands is preset to be L = 10, and the number of iterations η = 10. The results are compared with those of the PHAT-SPR, ROTH-SPR, and SCOT-SPR methods.

[0171] The mean absolute error (MAE) of the positioning results and the cumulative distribution function (CDF) of the relative distance estimation error are calculated to evaluate the accuracy and stability of these algorithms, namely

[0172]

[0173] In the formula, x represents the actual sound source position, Indicates the SRP estimated sound source position, L e Indicates positioning distance error.

[0174] See also Fig.17 , Fig.17 The horizontal axis is the grid spacing, and the vertical axis is the mean positioning error (MAE) of the performance accuracy. It can be seen from the average positioning error curve. When the grid spacing is very small, EM-SRP, SVD-SRP and WSVD-SRP can achieve good positioning. As the grid spacing increases, the positioning errors of the SVD-SRP and WSVD-SRP algorithms increase rapidly. Since the TDOA algorithm is independent of the grid, its CDF curve does not change. The EM-SRP algorithm obtains better results at all sizes. It can be seen that when EM-GCC is applied to SRP positioning, more accurate positioning results can be obtained.

[0175] Based on the same inventive concept as the sound source localization method, this embodiment also provides a semantic segmentation system accordingly. In this embodiment, the sound source localization system executes the sound source localization method described in any of the above embodiments, and the specific functions and technical effects can be referred to the above embodiments, which will not be repeated here.

[0176] Fig.18 This is a schematic structural diagram of the sound source localization system provided by the present invention.

[0177] like Fig.18 As shown, the sound source localization system includes: 11 a mapping module, 12 a target model building module, and 13 a target result acquisition module.

[0178] The frequency domain signal module is used to collect target sound source signals generated by the same sound source using a plurality of sensors, and perform spectrum transformation on the target sound source signals collected by the plurality of sensors respectively to obtain a plurality of target frequency domain signals;

[0179] A feature extraction module, used to divide each target frequency domain signal into a plurality of sub-frequency domain signals according to a plurality of preset frequency bands, and extract the first generalized cross-correlation feature of the sub-frequency domain signals in the same preset frequency band by using a generalized cross-correlation algorithm;

[0180] A target weight module is used to establish a mapping relationship between the generalized cross-correlation feature and the sound source weight, bring the first generalized cross-correlation feature and the mapping relationship into the conditional expectation function of the expectation-maximization algorithm, and perform multiple iterations to obtain the target sound source weight;

[0181] A target power module, used to determine a second generalized cross-correlation feature according to the first generalized cross-correlation feature and the target sound source weight, and to obtain a target controllable response power according to the controllable response power function and the second generalized cross-correlation feature;

[0182] The target positioning module is used to divide the preset candidate area into several grid areas, obtain the position information corresponding to the target controllable response power according to the grid area, and obtain the target position information. The frequency domain signal module, feature extraction module, target weight module, target power module and target positioning module are connected.

[0183] This embodiment further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, any one of the methods in this embodiment is implemented.

[0184] In one embodiment, this embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method described in any of the above embodiments when executing the computer program.

[0185] The computer-readable storage medium in this embodiment can be understood by ordinary technicians in this field: all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to the computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other media that can store program codes.

[0186] The electronic device provided in this embodiment includes a processor, a memory, a transceiver and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication with each other. The memory is used to store computer programs, the communication interface is used to communicate, and the processor and the transceiver are used to run the computer program so that the electronic device executes each step of the above method.

[0187] In this embodiment, the memory may include a random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0188] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0189] In the above embodiments, the description refers to "this embodiment", "an embodiment", "another embodiment", "in some exemplary embodiments" or "other embodiments" to indicate that the specific features, structures or characteristics described in conjunction with the embodiments are included in at least some embodiments, but not necessarily all embodiments. Multiple occurrences of "this embodiment", "an embodiment", "another embodiment" do not necessarily all refer to the same embodiment.

[0190] In the above-described embodiments, although the invention has been described in conjunction with specific embodiments of the invention, many substitutions, modifications, and variations of these embodiments will be apparent to those of ordinary skill in the art based on the foregoing description. For example, other storage structures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed. Embodiments of the invention are intended to encompass all such substitutions, modifications, and variations that fall within the broad scope of the appended claims.

[0191] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0192] The present invention can be used in many general or special computing system environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like.

[0193] The present invention may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0194] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A sound source localization method, characterized in that: include: Using a plurality of sensors to collect target sound source signals generated by the same sound source, and performing spectrum transformation on the target sound source signals collected by the plurality of sensors respectively, to obtain a plurality of target frequency domain signals; Dividing each of the target frequency domain signals into a plurality of sub-frequency domain signals according to a plurality of preset frequency bands, and extracting first generalized cross-correlation features of the sub-frequency domain signals in the same preset frequency band by using a generalized cross-correlation algorithm; The extracting the first generalized cross-correlation feature of the sub-frequency domain signal in the same preset frequency band by using the generalized cross-correlation algorithm comprises: obtaining the cross-power spectrum of the sub-frequency domain signal in the same preset frequency band, and processing the cross-power spectrum by using the generalized cross-correlation algorithm to obtain the first generalized cross-correlation feature; Establish a mapping relationship between generalized mutual correlation features and sound source weights, bring the first generalized mutual correlation features and the mapping relationship into the conditional expectation function of the expectation-maximization algorithm, and perform multiple iterations to obtain the target sound source weight; bringing the first generalized mutual correlation features and the mapping relationship into the conditional expectation function of the expectation-maximization algorithm, and performing multiple iterations to obtain the target sound source weight includes: obtaining full-band generalized mutual correlation features according to the mapping relationship; obtaining complete data consisting of a mean and a covariance matrix according to the mapping relationship; bringing the first generalized mutual correlation features, the full-band generalized mutual correlation features, and the complete data into the conditional expectation function of the expectation-maximization algorithm, and performing multiple iterations, and after the iterations are completed, obtaining a conditional expectation result; obtaining the target sound source weight according to the conditional expectation result; Determine a second generalized mutual correlation feature according to the first generalized mutual correlation feature and the target sound source weight, and obtain a target controllable response power according to a controllable response power function and the second generalized mutual correlation feature; obtaining the target controllable response power according to the controllable response power function and the second generalized mutual correlation feature includes: obtaining a controllable response power according to the controllable response power function and the second generalized mutual correlation feature, and determining that the maximum controllable response power is the target controllable response power, and the second generalized mutual correlation feature is a generalized mutual correlation feature of all sensors in a preset frequency band corresponding to the target sound source weight; The preset candidate area is divided into a plurality of grid areas, and the position information corresponding to the target controllable response power is acquired according to the grid areas to obtain the target position information.

2. The sound source localization method according to claim 1, characterized in that: The mathematical expression of the cross power spectrum is: Among them, Y m (ω) is the Fourier transform of the target sound source signal received by sensor m, is the conjugate of the Fourier transform of the target sound source signal received by sensor n, * is the conjugate, m and n are the numbers of the sensors, ω is the angular frequency, G mn (ω) is the cross power spectrum of sensor n and sensor m; The mathematical expression of the first generalized cross-correlation feature is: Among them, R mn (τ) is the first generalized cross-correlation feature of sensor n and sensor m under the delay value τ, ψ mn is the phase transformation weight function of sensor n and sensor m, j is an imaginary number, ω is the angular frequency, and τ is the delay value of the target sound source signal; in, 3. The sound source localization method according to claim 1, characterized in that: The mapping relationship between the generalized cross-correlation feature and the sound source weight includes: In the formula, τ l is the delay value of the preset frequency band l, r l is the delay value τ l The generalized cross-correlation characteristics of the preset frequency band l under , L is the total number of preset frequency bands, l is the label of the preset frequency band, is the generalized cross-correlation characteristic of the sound source in the preset frequency band l, λ l ∈[0,1],λ l is the sound source weight of the preset frequency band l, n l is the noise component of the generalized cross-correlation characteristic of the preset frequency band l, n l are independent zero-mean Gaussian variables.

4. The sound source localization method according to claim 3, characterized in that: The mathematical expression of the full-band generalized cross-correlation feature R is: The complete data The mathematical expression is: Among them, r1 is the generalized cross-correlation feature of the preset frequency band 1, r2 is the generalized cross-correlation feature of the preset frequency band 1, and r L is the generalized cross-correlation feature of the preset frequency band L, T is the transpose of the matrix, and the variance of the noise in the preset frequency band l is The variance of the full-band noise is σ 2 ; The mathematical expression of the mean μ of complete data is: Among them, λ1 is the sound source weight of the preset frequency band 1, λ2 is the sound source weight of the preset frequency band 2, and λ L is the sound source weight of the preset frequency band L, is the generalized cross-correlation feature of the sound source in preset frequency band 1, is the generalized cross-correlation feature of the sound source in preset frequency band 2, is the generalized cross-correlation characteristic of the sound source in the preset frequency band L; The mathematical expression of the covariance matrix Q of the complete data is: Among them, diag is a diagonal matrix function, β l is the noise power of the preset frequency band l; 5. The sound source localization method according to claim 4, characterized in that: The parameters of the conditional expectation function θ=[θ1,θ2,…,θ l ],θ l =[τ l ,λ l ], The mathematical expression of the conditional expectation function of the expectation maximization algorithm is: Where i is the number of iterations, θ i is the parameter of the conditional expectation function of the i-th iteration, is the conditional probability density, is the conditional expectation function, C is the preset constant, Delay value The i-th iteration generalized cross-correlation feature of the preset frequency band l under is the i-th iteration delay value of the preset frequency band l; The mathematical expression of the conditional expectation result is The sound source weight of the last iteration of the preset frequency band The mathematical expression is: in, Delay value The last iteration generalized cross-correlation feature of the preset frequency band l under is the last iteration delay value of the preset frequency band l, s is the final iteration number, is the transpose of the generalized cross-correlation feature of the sound source in the preset frequency band l; The mathematical expression of the target sound source weight k is:

6. The sound source localization method according to claim 3, characterized in that: The step of establishing a mapping relationship between generalized cross-correlation features and sound source weights and introducing the mapping relationship into a conditional expectation function of an expectation maximization algorithm includes: For the generalized cross-correlation feature r l Make corrections to obtain the corrected generalized cross-correlation characteristics; The mapping relationship is updated according to the modified generalized mutual correlation feature, and the updated mapping relationship is introduced into the conditional expectation function of the expectation-maximization algorithm.

7. The sound source localization method according to claim 1, characterized in that: The mathematical expression of controllable response power is: Among them, m and n are the numbers of sensors. is the second generalized cross-correlation feature of sensor m and sensor n, τ mn (x) is the time delay value between sensor m and sensor n, and x is the grid point.

8. A sound source localization system, characterized in that: include: A frequency domain signal module is used to collect target sound source signals generated by the same sound source using a plurality of sensors, and perform spectrum transformation on the target sound source signals collected by the plurality of sensors respectively to obtain a plurality of target frequency domain signals; A feature extraction module, used to divide each of the target frequency domain signals into a plurality of sub-frequency domain signals according to a plurality of preset frequency bands, and extract a first generalized cross-correlation feature of the sub-frequency domain signals in the same preset frequency band by using a generalized cross-correlation algorithm; The extracting the first generalized cross-correlation feature of the sub-frequency domain signal in the same preset frequency band by using the generalized cross-correlation algorithm comprises: obtaining the cross-power spectrum of the sub-frequency domain signal in the same preset frequency band, and processing the cross-power spectrum by using the generalized cross-correlation algorithm to obtain the first generalized cross-correlation feature; A target weight module is used to establish a mapping relationship between generalized mutual correlation features and sound source weights, bring the first generalized mutual correlation features and the mapping relationship into the conditional expectation function of the expectation-maximization algorithm, and perform multiple iterations to obtain the target sound source weight; bringing the first generalized mutual correlation features and the mapping relationship into the conditional expectation function of the expectation-maximization algorithm, and performing multiple iterations to obtain the target sound source weight includes: obtaining full-band generalized mutual correlation features according to the mapping relationship; obtaining complete data composed of a mean and a covariance matrix according to the mapping relationship; bringing the first generalized mutual correlation features, the full-band generalized mutual correlation features, and the complete data into the conditional expectation function of the expectation-maximization algorithm, and performing multiple iterations, and after the iterations are completed, obtaining the conditional expectation result; obtaining the target sound source weight according to the conditional expectation result; A target power module, used to determine a second generalized mutual correlation feature according to the first generalized mutual correlation feature and the target sound source weight, and obtain a target controllable response power according to a controllable response power function and the second generalized mutual correlation feature; the obtaining of the target controllable response power according to the controllable response power function and the second generalized mutual correlation feature includes: obtaining a controllable response power according to the controllable response power function and the second generalized mutual correlation feature, and determining the maximum controllable response power as the target controllable response power, the second generalized mutual correlation feature being a generalized mutual correlation feature of all sensors in a preset frequency band corresponding to the target sound source weight; The target positioning module is used to divide the preset candidate area into a plurality of grid areas, obtain the position information corresponding to the controllable response power of the target according to the grid areas, and obtain the target position information. The frequency domain signal module, the feature extraction module, the target weight module, the target power module and the target positioning module are connected.

9. An electronic device, characterized in that: including a processor, a memory and a communication bus; The communication bus is used to connect the processor and the memory; The processor is configured to execute the computer program stored in the memory to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is used to make the computer execute the method according to any one of claims 1 to 7.

Images