Sound source localization method based on double-microphone rotating array

Abstract

The invention relates to a sound source positioning method based on a double-microphone array, and belongs to the technical field of acoustic positioning. The method comprises the steps that sound source signals received when two microphones rotate at a constant speed along with a rotating disc are collected every a period of fixed time, and the collected multiple sound source signals with time delay are equivalently simulated into a circular array composed of the multiple microphones; converting the time domain signal into a frequency domain signal, calculating a cross-correlation function between different frequency domain signals, and weighting and filtering the cross-correlation function; inverse Fourier transform is carried out on the processed cross-correlation function, so that frequency domain information is converted into a time domain, and cross-correlation functions among virtual microphone positions at different time points are obtained; in a unit space for grid division, the position of a sound source is found by adopting a strategy of first rough search and then fine search. According to the method, the use number of microphones can be reduced, the hardware cost is reduced, meanwhile, the calculated amount can be greatly reduced, and the real-time performance of sound source positioning is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of acoustic positioning, and relates to a sound source positioning method based on a dual-microphone rotating array. Background Art

[0002] Microphone array sound source localization technology has gradually matured and plays an important role in speech enhancement, automatic speech recognition, and intelligent robots. Microphone arrays use the spatial characteristics of signals to perform weighted processing by arranging and combining multiple microphones to achieve signal enhancement and interference suppression, thereby improving signal quality and performance. The accuracy of sound source localization is affected by many factors, especially the sound source localization algorithm, which directly processes the sound data received by the microphone to accurately estimate the location of the sound source. The current mainstream sound source localization algorithms are divided into three categories: controllable beamforming algorithms, subspace sound source localization algorithms, and arrival delay difference algorithms.

[0003] Among them, the controllable beamforming algorithm mainly uses beamforming technology to process the sound signals received by the array. After removing useless frequencies through filtering technology, the signals are combined and weighted, and the beam direction of the array is controlled to enhance the signal energy in a specific direction, thereby realizing the location of the sound source. Commonly used maximum output power controllable beamforming algorithms include the delayed accumulation beamforming algorithm and the adaptive beamforming algorithm. The delayed accumulation beamforming algorithm offsets the propagation delay through time delay processing, with small computational complexity and low signal distortion, but poor anti-noise performance, and usually requires more microphones to improve the effect. The adaptive beamforming algorithm adapts to complex environments, but the computational complexity is large, the output signal may be distorted, and reverberation needs to be avoided when the effect is good.

[0004] The subspace sound source localization algorithm is a sound source localization method based on array signal processing. The commonly used high-resolution spectrum estimation method extends the time-domain Fourier method to the spatial domain, constructs the cross-spectral matrix of the sound source signal and the microphone array, and estimates the direction of arrival (DOA). The conventional beamforming (CBF) algorithm is the earliest DOA algorithm. Although it simplifies the time domain data processing, its resolution within the beam width is insufficient. In order to improve the resolution, it is usually necessary to increase the array aperture, but this is often not feasible in practical applications, so the improvement of the algorithm has become a research focus.

[0005] The sound source localization algorithm based on arrival delay difference calculates the delay difference between multiple array elements receiving the same sound signal, and uses these delay differences and the geometric relationship of the microphones to estimate the sound source position. The key to this method is to accurately calculate the arrival time difference of the sound source between different microphones. The delay estimation method of the cross-correlation function is usually used, such as the generalized cross-correlation method (GCC), the maximum likelihood weighted method (ML) and the cross-power spectrum phase method (CSP). Although the GCC algorithm is commonly used, there may be differences between theoretical and actual results when there are interfering sound sources, and a large amount of data calculation is required to improve accuracy. By obtaining the delay difference and the geometric position information of the microphone, the sound source position can be calculated using the triangulation positioning method or the least squares method. The algorithm calculates the distance between the sound source and the microphone based on time and sound speed, and determines the exact position of the sound source through the curve equation constructed by multiple microphones. However, errors may cause the curves determined by two microphones to not necessarily intersect at one point, so the sound source position is often determined by the least squares fitting of the intersection area of ​​multiple microphones. In general, delay difference estimation and sound source localization are two steps, and the results reflect the past position of the sound source. Therefore, the accuracy is affected by the delay difference estimation and the microphone geometry information. In the case of multiple sound sources, the delay difference estimation becomes more complicated, and the geometry of the microphone array needs to be optimized to meet such requirements. In addition, the accuracy of the algorithm is closely related to the signal sampling rate, and continuous sampling and calculation fitting are required to improve the positioning accuracy.

[0006] In summary, the existing sound source localization algorithm based on arrival time delay difference has the problems of using multiple microphone arrays, high hardware cost, large placement space, and inconvenient movement. At the same time, the calculation of the cross-correlation values ​​of all grid points is computationally intensive, the ability to resist noise interference is weak, and the real-time performance is poor. Summary of the invention

[0007] In view of this, the purpose of the present invention is to provide a sound source localization method based on a dual-microphone rotation array to reduce the number of microphones used, reduce the amount of calculation of cross-correlation values, improve the anti-interference ability of the cross-correlation function, and at the same time speed up the sound source position search speed and improve real-time performance.

[0008] In order to achieve the above object, the present invention provides the following technical solutions:

[0009] A sound source localization method based on a dual microphone array, the method comprising:

[0010] The two microphones are made to rotate at a constant speed along with the rotating disk, and the sound source signals received by the two microphones are collected at fixed intervals. The collected multiple sound source signals with time delays are regarded as a group, which is equivalent to simulating a circular array composed of multiple microphones.

[0011] Convert the collected multiple time domain signals into frequency domain signals, calculate the cross-correlation function between different frequency domain signals, and then perform weighted and filtering processing on the cross-correlation function;

[0012] Performing inverse Fourier transform on the processed cross-correlation function, thereby converting the frequency domain information into the time domain, and obtaining the cross-correlation function between the virtual microphone positions at different time points;

[0013] A unit space is gridded by a triangle recursive method, and a simulated microphone array is set at the origin of the unit space. The time delay from any grid point in the unit space to each pair of microphones is obtained, and the cross-correlation value of each microphone pair is calculated according to the time delay and the signal received by the microphone pair. The cross-correlation values ​​of each microphone pair are accumulated as the energy value of the grid point. After assigning energy values ​​to all grid points in the unit space, the grid point with the largest energy value among all grid points is searched, and the grid point with the largest energy value is the location of the sound source.

[0014] The fixed time is the time required for the microphone to rotate 30°. The microphone collects a sound source signal every 30° rotation. During the 360° rotation of the rotating disk, 12 sound source signals with time delay are collected, which is equivalent to a circular array composed of 12 microphones.

[0015] Further, the collected multiple time domain signals are converted into frequency domain signals, and then the cross-correlation function between different frequency domain signals is calculated, including: for the collected multiple time domain signals, they are first divided into L frames, and then the L frame signals corresponding to each time domain signal are converted into the frequency domain; and then the cross-correlation function between different frequency domain signals is calculated by the following formula:

[0016]

[0017] Where, X i [k] and X j [k] are the k-th frame time domain signal x of virtual microphone i i (k) and the k-th frame time domain signal x of virtual microphone j j (k) is the discrete Fourier transform, τ represents x i (k) and x j (k) The time delay between (X i [k]X j [k]) * For x i (k) and x j The cross spectrum of (k), (·) * represents complex conjugate.

[0018] Furthermore, the cross-correlation function is weighted by the weighting function shown in the following formula:

[0019]

[0020] Where S(k) is the cross-correlation function between different frequency domain signals in the kth frame, k = 0, 1, ..., L-1, and L represents the number of frames for framing the acquired time domain signal;

[0021] Kalman filtering is added every time the rotating disk completes a 360° rotation, and the corresponding weight is calculated for each cross-correlation function, as shown in the following formula:

[0022]

[0023] In the formula, Represents the weighted prediction value calculated at the last moment, S n (k) represents the cross-correlation function between different frequency domain signals of the kth frame currently predicted; N n (·) indicates the weight assignment operation.

[0024] Furthermore, after assigning energy values ​​to all grid points in the unit space, the four grid points with the largest energy values ​​among all grid points are first searched to preliminarily define the area of ​​the sound source location; then, the grid points are further divided in the defined area to improve the resolution of the area;

[0025] For the area with improved resolution, an energy value is reassigned to each grid point in the area, and the grid point with the largest energy value is searched again in the limited area, which is the location of the sound source.

[0026] The beneficial effect of the present invention is that the present invention can effectively reduce hardware costs, improve practicality, and reduce the influence of sound source propagation barriers on the sound source localization effect by simulating the signals collected at different time points equivalently as a circular array composed of multiple microphones, and then calculating the cross-correlation values ​​of the virtual microphone components. In addition, the present invention divides the space grid with different resolutions, pre-calculates the time delay value from the grid point position to each pair of microphones, and adopts the strategy of performing a coarse search first and then a fine search when searching for the sound source position, which greatly reduces the amount of calculation, effectively speeds up the speed of sound source localization, and improves the real-time performance of sound source localization.

[0027] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0029] Figure 1 is a flow chart of a sound source localization method shown in an exemplary embodiment of the present application;

[0030] Figure 2 is a schematic diagram of a handheld rotating microphone shown in an exemplary embodiment of the present application;

[0031] Figure 3 is a schematic diagram of a simulated circular microphone array shown in an exemplary embodiment of the present application;

[0032] Figure 4 is a spatial grid division diagram shown in an exemplary embodiment of the present application;

[0033] Figure 5 is a schematic diagram of microphone directivity shown in an exemplary embodiment of the present application;

[0034] Figure 6 is a simulation gain function diagram shown in an exemplary embodiment of the present application;

[0035] Figure 7 It is a schematic diagram of a unit sphere search strategy shown by an exemplary embodiment of the present application.

[0036] Figure numerals: 1-motor module; 2-acrylic plate; 3-MEMS microphone; 4-communication module; 5-handheld handle. DETAILED DESCRIPTION

[0037] 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 illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0038] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0039] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0040] like Figure 2 As shown, a handheld rotating microphone device provided by an embodiment of the present invention includes a motor module 1, a circular acrylic plate 2, a MEMS microphone 3, a communication module 4 and a handheld handle 5. The motor module 1 includes a motor and a battery, which is responsible for the rotation of the motor. The acrylic plate 2 is used to fix the microphone. The MEMS microphone 3 is used to receive sound signals and convert analog signals into digital signals. The communication module 4 is used to remotely transmit the sound signals output by the microphone to a computer.

[0041] The diameter of the circular acrylic plate 2 is 10 cm. Two MEMS microphones 3 are symmetrically fixed on a diameter of the circular acrylic plate 2 and are located at the edge of the circular acrylic plate 2, so that the sound collection point of the MEMS microphone 3 is located at the outermost side of the circular acrylic plate 2, thereby enhancing the spatial resolution.

[0042] Based on the above-mentioned rotating microphone device, the sound source localization method based on the dual-microphone rotating array proposed in the present invention is as follows: Figure 1 As shown, it includes the following steps:

[0043] N1: The two microphones receive sound source signals at different angles as the rotating disk rotates. The collected analog signals are converted into digital signals through the sound card and transmitted to the computing device. The preprocessing module performs frame processing on the time domain signals collected by each microphone at different time points. The signals collected at each rotation are divided into L = 1024 independent frames.

[0044] In the computing device, the signal is first processed by a power complementary window to reduce the impact of spectrum leakage, and then the time domain signal is converted into a frequency domain signal by fast Fourier transform (FFT). The signals obtained at different time points can be regarded as receiving sound source signals from multiple virtual microphone positions at the same time. Specifically, according to the rotation speed of the circular acrylic plate, the time interval required for the microphone to rotate 30° is calculated, and a total of 12 time intervals are obtained. Therefore, the 12 delayed signals are regarded as a group, which is equivalent to simulating an array consisting of 12 microphones. Figure 3The figure shows a simulated circular array formed by two microphones after rotation. The simulated circular array is formed by twelve microphones, and the angle between each two adjacent microphones and the line connecting the center of the circle is 30°.

[0045] The output of the M-microphone delay-sum beamformer is defined as:

[0046]

[0047] Where M = 2, x m (n) is the time domain signal collected by the mth microphone, τ m is the time delay when the sound source reaches the mth microphone relative to the first microphone. According to the output of the delay-sum beamformer, the output energy of the beamformer in the frame length of L = 1024 can be obtained:

[0048]

[0049] From formula (2), it can be seen that if the maximum value of E is required, then it is only necessary that the signals of the microphones are in the same direction, that is, the peak value of each microphone signal corresponds to the peak value. Expanding the energy formula through the square sum formula, we get:

[0050]

[0051] When calculating the maximum value of E in formula (3), the square term on the right side of the equal sign is approximately a constant and can be ignored.

[0052] In order to reduce the amount of calculation and facilitate the later whitening of the signal, the time domain signal is converted into a frequency domain signal, and the cross-correlation value between the frequency domain signals of the two microphones is calculated to obtain:

[0053]

[0054] Among them, X i [k] is x i The discrete Fourier transform of (k), (X i [k]X j [k]) * For x i (k) and x j The cross spectrum of (k), (·) * represents complex conjugate, R ij(τ) represents the cross-correlation function of the frequency domain signal of the kth frame of virtual microphone i and the frequency domain signal of the kth frame of virtual microphone j. In the signal collected at a sampling rate of 16kHz, an overlapping window with a length of L = 1024 sampling points (50% overlap) is used to calculate the power spectrum and cross power spectrum for each window. By calculating the cross-correlation value between the frequency domain signals of two microphones, the similarity between the two signals can be evaluated. By finding the maximum value in the cross-correlation function, the time delay τ of the sound source signal can be determined. The time delay reflects the time delay difference of the sound source reaching each microphone, which provides key data for the subsequent sound source localization.

[0055] N2: After calculating the frequency domain signals of the microphones at different time points, the frequency domain summation function of the signals is rewritten using the timing relationship of these signals to obtain the cross-correlation function based on the signals collected at different time points (as shown in formula (4)). By whitening these cross-correlation functions, the influence of environmental noise is reduced, and a weighting function is introduced to further suppress the noise, thereby obtaining an optimized cross-correlation function. Then, the optimized cross-correlation function is inversely Fourier transformed to convert the frequency domain information back to the time domain, and the cross-correlation function between the virtual microphone positions at different time points is obtained, providing accurate time delay estimation information for sound source localization.

[0056] The cross-correlation value R in formula (4) ij (τ) is the average cross power spectrum (X i [k]X j [k]) * Calculated. Pre-calculated R ij (τ) can be calculated using only M(M-1) / 2 lookup and accumulation operations, while calculating E in the time domain requires 2L(M+2) operations. For M=2 and 2562 directions, the complexity of the search itself is reduced from 2.4Gflops to only 3.5Mflops. After calculating all time-frequency transforms, the complexity is only 96.8Mflops, which is 25 times lower than the time domain search at the same resolution.

[0057] In order to facilitate the observation and comparison of the calculated results, equation (4) is normalized:

[0058]

[0059] The normalized cross-correlation will produce a sharper cross-correlation peak, but it also has an obvious disadvantage, that is, the influence of each frequency domain of the spectrum on the final cross-correlation result calculation is the same, that is, the influence of noise is obvious, the system has poor anti-interference ability to noise, and it is more difficult to detect the sound source. To solve this problem, a weighting function is introduced, which is based on the signal-to-noise ratio, as shown in the following formula:

[0060]

[0061] Wherein, S(k) is the cross-correlation function of the two microphones on the frequency domain signal of the kth frame.

[0062] Kalman filtering is added at each update (that is, when the circular acrylic plate rotates 360° and the microphone starts collecting signals for the next cycle), so as to calculate the corresponding weight for each signal, as shown in the following formula:

[0063]

[0064] in, is the weighted prediction value calculated at the previous moment, S n (k) is the cross-correlation function of the different frequency domain signals of the currently predicted kth frame. n (·) indicates the weight assignment operation.

[0065] N3: After obtaining the optimized cross-correlation function, the sound source position estimation strategy is first based on the time delay information between these virtual microphone positions. It uses the time delay differences collected by the microphones at various angles on the rotating disk and searches for the most likely position of the sound source on the simulated circular microphone array through a spatial search algorithm.

[0066] This strategy makes full use of the spatial characteristics of virtual array coverage and can quickly converge to the optimal position of the sound source with less computing resources, which is suitable for the needs of real-time sound source localization.

[0067] In order to facilitate the search for the sound source position and reduce the amount of calculation, a unit space is defined, and the unit space is divided by a uniform triangular grid to obtain a spatial grid for sound source localization. To create a spatial grid, start with an initial 20-sided sphere grid. Each triangle in the initial 20-element grid is recursively subdivided into 4 smaller triangles. Each smaller triangle after recursive subdivision is recursively subdivided again, such as Figure 4 As shown in the figure, the final grid consists of 5120 triangles and 2562 points. The accuracy of the grid can be determined by the number of recursions. In order to quickly search for the sound source location, the grid is divided twice, corresponding to the coarse search and the fine search respectively.

[0068] The microphone array is set at the origin of the unit space. According to the coordinates of each grid point and the position parameters of the microphone, the time delay difference from any grid point to all microphone pairs can be calculated. According to the time delay difference and the microphone signal, the cross-correlation value of each pair of microphones can be calculated. In microphone arrays, it is usually assumed that the microphones are omnidirectional, that is, they acquire signals from all directions with the same gain. However, in practice, the microphones are mounted on a rigid body, which may block the direct propagation path between the sound source and the microphone, causing attenuation of the sound source signal. The attenuation is mainly caused by diffraction and varies with frequency.

[0069] Since an accurate diffraction model is not available, the proposed model relies on simpler assumptions: (1) there is unity gain for a sound source with a direct propagation path, and (2) the gain is zero when the path is blocked by an object. Since the signal-to-noise ratio of an obstructed microphone is usually unknown, it is safer to assume a lower signal-to-noise ratio, and setting the gain to zero prevents noise from being injected into the observation. In addition, the gain is constant for all frequencies, and a smooth transition band connects the unity gain region and the zero gain region. This transition band prevents abrupt changes in gain when the sound source position changes. Figure 5 θ(u, d) is shown, which represents the angle between the sound source at position u and the direction of the microphone modeled by the unit vector d. θ(u, d) is expressed as follows:

[0070]

[0071] Let the analog gain G(u,D) be a function of θ(u,d), as shown below:

[0072]

[0073] Where D is a set of parameters {d, α, β}, α represents the angle where the gain is 1, β represents the angle where the gain is zero, and the area between α and β can be regarded as a transition zone, such as Figure 6 shown.

[0074] To make the sound source localization more reverberant, the scan space is limited to a specific direction. For example, the scan space is limited to a hemisphere pointing toward the ceiling to ignore reflections from the floor, which also reduces the number of grid points that need to be searched and speeds up the search.

[0075] Based on the spatial grid constructed above and the assumptions, after calculating the cross-correlation relationship of each pair of microphones, we start to search each grid point of the spatial grid in a loop to search for the best direction of the sound source. First, set the energy value of each grid point to 0, loop the microphone pair i, j, obtain the pre-calculated time delay τ from each grid point to each pair of microphones, and calculate the cross-correlation value of each team of microphones. The cross-correlation value of each microphone team The energy value E accumulated to the grid point p p Therefore, the energy value of each grid point is the sum of the cross-correlation values ​​of all microphone pairs, and the direction of the sound source is the grid point with the largest energy value.

[0076] The search strategy is to find the grid point with the largest energy value as quickly as possible, such as Figure 7 As shown in the figure, the approximate sound source position is first searched in the A sphere with a lower resolution. After the approximate sound source position is found, the refined grid points near the approximate sound source point are searched, as shown in the B sphere. Specifically, the sphere with a lower resolution is first divided for a coarse search to find the point with the largest energy value (recorded as the first grid point) in the grid points of the sphere, which is the approximate direction of the sound source; then a high-resolution grid is divided in a certain space around the first grid point, and the energy value of each grid point in this certain space is recalculated to perform a fine search in the certain space, and the grid point with the largest energy value is found again, which is the precise position of the sound source. This search strategy can effectively reduce calculations while improving the accuracy of positioning.

[0077] At the same time, in order to draw a visual heat map, four cycles are performed when searching for the sound source position in the coarse search and fine search stages. In the first cycle, all grid points are traversed to find the maximum value point, and the position coordinate information and energy value of the point are output, and then the point is removed from the space; then the second cycle is entered, and the second cycle also traverses all grid points to find the maximum value. The maximum value found in the second cycle is equivalent to the second largest value point of the whole. The position coordinates and energy value of this point are also output, and the point is removed from the space; the third cycle performs the same operation, and the fourth cycle finds the grid point with the largest energy value in the space and outputs the position coordinate information of the point. After four cycles of the coarse search and fine search stages, 8 points are obtained, and a visual heat map is drawn according to the energy values ​​of these 8 points.

[0078] As a practical application scenario of the present invention, the present invention can be used to detect the broken parts of the motor. During the detection, the operator holds the Figure 2 The audio-visual device shown in the figure is kept at a certain detection distance from the motor to be tested. The audio-visual device has a built-in camera to capture the image information of the motor surface in real time, and transmit the acquired images and related data to the computer for processing and display over long distances by wireless or wired means. At the same time, the audio-visual device uses the built-in sound source positioning module to synchronously detect abnormal sound source signals emitted during the operation of the motor, and combines the camera image information to accurately identify and locate the broken part of the motor, providing the operator with real-time and intuitive fault diagnosis auxiliary functions. The device is designed to be portable and suitable for rapid detection needs in complex industrial scenarios.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A sound source localization method based on a dual microphone array, characterized in that: The method includes: The two microphones are made to rotate at a constant speed along with the rotating disk, and the sound source signals received by the two microphones are collected at fixed intervals. The collected multiple sound source signals with time delays are regarded as a group, which is equivalent to simulating a circular array composed of multiple microphones. Convert the collected multiple time domain signals into frequency domain signals, calculate the cross-correlation function between different frequency domain signals, and then perform weighted and filtering processing on the cross-correlation function; Performing inverse Fourier transform on the processed cross-correlation function, thereby converting the frequency domain information into the time domain, and obtaining the cross-correlation function between the virtual microphone positions at different time points; A unit space is gridded by a triangle recursive method, and a simulated microphone array is set at the origin of the unit space. The time delay from any grid point in the unit space to each pair of microphones is obtained, and the cross-correlation value of each microphone pair is calculated according to the time delay and the signal received by the microphone pair. The cross-correlation values ​​of each microphone pair are accumulated as the energy value of the grid point. After assigning energy values ​​to all grid points in the unit space, the grid point with the largest energy value among all grid points is searched, and the grid point with the largest energy value is the location of the sound source.

2. The sound source localization method according to claim 1, characterized in that: The fixed time is the time required for the microphone to rotate 30°.

3. The sound source localization method according to claim 2, characterized in that: The microphone collects a sound source signal every time it rotates 30°. During the 360° rotation of the rotating disk, 12 sound source signals with time delays are collected, which is equivalent to a circular array composed of 12 microphones.

4. The sound source localization method according to claim 1, characterized in that: The multiple time domain signals collected are converted into frequency domain signals, and then the cross-correlation function between different frequency domain signals is calculated, including: for the multiple time domain signals collected, they are first divided into L frames, and then the L frame signals corresponding to each time domain signal are converted into the frequency domain; and then the cross-correlation function between different frequency domain signals is calculated by the following formula: Where, X i [k] and X j [k] are the k-th frame time domain signal x of virtual microphone i i (k) and the k-th frame time domain signal x of virtual microphone j j (k) is the discrete Fourier transform, τ represents x i (k) and x j (k) The time delay between (X i [k]X j [k]) * For x i (k) and x j The cross spectrum of (k), (·) * represents complex conjugate.

5. The sound source localization method according to claim 1 or 4, characterized in that: The cross-correlation function of the frequency domain signal is weighted and filtered, including: weighting the cross-correlation function by a weighting function shown in the following formula: Where S(k) is the cross-correlation function between different frequency domain signals in the kth frame, k = 0, 1, ..., L-1, and L represents the number of frames for framing the acquired time domain signal; Kalman filtering is added every time the rotating disk completes a 360° rotation, and the corresponding weight is calculated for each cross-correlation function, as shown in the following formula: In the formula, Represents the weighted prediction value calculated at the last moment, S n (k) represents the cross-correlation function between different frequency domain signals of the kth frame currently predicted; N n (·) indicates an empowerment operation.

6. The sound source localization method according to claim 1, characterized in that: After assigning energy values ​​to all grid points in the unit space, first search for the grid point with the largest energy value among all grid points to preliminarily define the area of ​​the sound source position; then continue to divide the grid points in the defined area to improve the resolution of the area; For the area with improved resolution, an energy value is reassigned to each grid point in the area, and the grid point with the largest energy value is searched again in the limited area, which is the location of the sound source.

Images