Sound source localization method based on same-order relative circular harmonic coefficients

Abstract

The invention discloses a sound source localization method based on same-order relative circular harmonic coefficients. The method is used for solving the singular problem of an open type circular array in a Bessel zero frequency interval and a high-order low-frequency range. When a circular harmonic domain characteristic wave beam is processed, mode intensity compensation is usually adopted to eliminate dependence on array characteristics. However, the problem of singularity exists in the frequency with very small modal intensity amplitude. In combination with the same-order mode intensity consistency characteristic, a new positioning feature same-order relative circular harmonic coefficient is designed, dependence on microphone array characteristics is removed, and the influence of a singular problem on a positioning result is also solved. The positioning algorithm provided by the invention has relatively low calculation complexity, and can be efficiently deployed on a hardware platform with limited resources. On the basis of keeping low calculation complexity, the method still shows excellent positioning precision in a high-noise and high-reverberation environment.

Description

Technical Field

[0001] This invention relates to a sound source localization method based on the same-order relative circular harmonic coefficients, belonging to the field of speech signal arrival direction estimation technology. Background Technology

[0002] Direction of arrival (DOA) estimation for speech signals is a key technology in microphone array processing, widely used in fields such as robotic hearing, hearing aids, video conferencing, and speech enhancement. However, real-world application environments (such as homes and offices) often suffer from reverberation and noise interference, leading to performance degradation of traditional DOA algorithms. Reverberation introduces multipath effects, complicating sound source localization. Therefore, achieving high-accuracy and robust DOA estimation in high-noise, high-reverberation environments has become a research focus.

[0003] Open circular arrays are a common array type, but they suffer from the Bessel zero problem. The abnormal amplification of circular harmonic coefficients during mode intensity compensation due to periodic Bessel zeros and the low amplitude of modal intensities at high-order, low-frequency levels is called the singularity problem. To avoid the singularity problem caused by the Bezier zero, some articles suggest mounting the circular array around a rigid sphere (J. Meyer, "Beamforming for a circular microphone array mounted on spherically shaped objects" J. Acoust. Soc. Am, vol. 109, no. 1, pp. 185-193, 2001) or on an infinitely long rigid cylindrical baffle (GADaigle, MR Stinson, and JG Ryan, "Beamforming with air-coupled surface waves around a sphere and circular cylinder (L)," J. Acoust. Soc. Am, vol. 117, no. 6, pp. 3373-3376, 2005), but this solution is not practical.

[0004] Torres AM (AM Torres, M. Cobos, B. Pueo, and JJ Lopez, “Robustacoustic source localization based on modal beamforming and time-frequency processing using circular microphone arrays” J. Acoust. Soc. Am, vol. 132, no. 3, pp. 1511-1520, 2012) proposed using Tikhonov-regularized TF-CHB at each time-frequency point to address the singularity problem. Tikhonov regularization alleviates the Bessel zero problem, and this method achieves higher localization accuracy than DSB. However, applying Tikhonov-regularized TF-CHB to each time-frequency point leads to significant computational complexity.Hu et al. (Y. Hu, PN Samarasinghe, S. Gannot, and TD Abhayapala, "Evaluation and comparison of three source direction-of-arrival estimators using relative harmonic coefficients" in ICASSP. IEEE, 2021, pp. 815-819; 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 726-730; Y. Hu, W. Wang, Z. Gu, T.Mao, X. Zhu, and J. Jin, "Closed form multiple source direction-of-arrivalestimator under reverberant environments" The Journal of the AcousticalSociety of America, vol. 154, no. 4, pp. 2349-2364, (2023) proposed a relative harmonic coefficient (RHC) suitable for spherical arrays, applicable to high-order microphone arrays, and has been shown to be robust against room reverberation. Although RHC has proven to be very useful for spherical arrays, the RHC method for spherical arrays (using different order harmonic coefficients) is no longer applicable to circular arrays due to singularity issues. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a sound source localization method based on the same-order relative circular harmonic coefficients, which solves the singularity problem caused by the periodic Bessel zeros and the excessively low amplitude of the intensity of high-order low-frequency modes in the traditional relative harmonic coefficient localization method in the circular array.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A sound source localization method based on the same-order relative circular harmonic coefficients includes the following steps: Step 1: Perform a short-time Fourier transform on the sound pressure signal acquired by the uniform open circular microphone array to convert the sound pressure signal to the short-time Fourier domain. Then, use the orthogonal circular harmonic function to transform the sound pressure signal converted to the short-time Fourier domain to the circular harmonic domain to obtain the characteristic beam. Step 2: Perform mode intensity compensation calculation on the characteristic beam to obtain the circular harmonic coefficients. Construct relative circular harmonic coefficients of the same order based on the circular harmonic coefficients for source location estimation. The relative circular harmonic coefficients of the same order are defined as the ratio of the power spectrum and cross power spectrum of the circular harmonic coefficients of the same order. Step 3: In the presence of environmental noise, perform frequency smoothing on the same-order relative circular harmonic coefficients constructed in Step 2 to obtain frequency-smoothed same-order relative circular harmonic coefficients. Step 4: Based on the frequency smoothing of the same order relative circular harmonic coefficients, a number selection method is used to select some characteristic beams from the characteristic beams for sound source azimuth estimation, and a delay summation beamforming algorithm is used to determine the final sound source arrival direction angle.

[0007] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects: 1. This invention utilizes the principle that the mode intensity of an open circular array is consistent within the same order. The proposed positioning feature, Same Order-Relative Circular Harmonic Coefficients (SO-RCHC), removes the influence of array characteristics while avoiding the effects of singularity problems. Even under conditions of high noise and high reverberation, it can still obtain high-precision positioning results with very low computational complexity.

[0008] 2. To address the issues of strong dependence on sensor array characteristics and singularity problems when using open circular arrays in existing circular harmonic domain positioning methods, the positioning feature SO-RCHC used in this invention is constructed by removing harmonic coefficients that depend on array characteristics, thereby avoiding the influence of sensor array configuration on the positioning feature. Simultaneously, SO-RCHC effectively solves the singularity problem by utilizing the same-order consistency of mode intensity. The SO-RCHC positioning method solves the angle ambiguity problem with minimal increase in computation by separately calculating the beam power at the estimated angle and the reverse angle of the estimated angle. Attached Figure Description

[0009] Figure 1 This is a flowchart of the sound source localization method based on the same-order relative circular harmonic coefficients of the present invention; Figures 2(a), 2(b), and 2(c) are schematic diagrams illustrating the effect of reverberation time on the performance of each method at SNR=0dB, 5dB, and 10dB, respectively. Figure 3 These are schematic diagrams illustrating the intensity amplitudes of different modes; Figure 4 This is a schematic diagram comparing the amplitudes of the first three orders of DO-RCHC and SO-RCHC. Detailed Implementation

[0010] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0011] like Figure 1 As shown, this invention proposes an open circular array sound source localization method based on the same-order relative circular harmonic coefficients, comprising the following steps: Step 1: For a given uniform open circular microphone array structure, in designing the single-source localization algorithm for the reverberation chamber, consider that the added noise has zero mean and variance. The sound pressure signals discretely sampled by the microphone array elements are modeled in the short-time Fourier domain, and then decomposed into the circular harmonic domain using a set of orthogonal circular harmonic functions to obtain the characteristic beam.

[0012] Consider a An open, uniform circular array of microphones, with radius R, is formed, and the center of the array is located at the origin of the xy-plane. The azimuth angle of each microphone is ,in For an incident angle of... The far-field sound source, the first A microphone signal can be modeled in the short-time Fourier domain as: , in, It is the wave number. It is a time frame. , , It is the source signal. It is the first The zero mean and variance of the data received by each microphone are: Unrelated noise, From the source to the first The acoustic transfer function of each microphone. Assuming no spatial sampling error, using... The sound pressure field sampled by each microphone is obtained through circular harmonic transformation: , in, For characteristic beams, and They are and Circular harmonic transformation, , , For the largest order, Need to meet .

[0013] Step 2: The characteristic beam depends on the array characteristics (such as the number of microphone array elements, the array radius, and whether the array is mounted on a scatterer such as a rigid cylinder or an open circular array). In order to remove the dependence on array characteristics, the characteristic beam is subjected to mode intensity compensation to obtain the circular harmonic coefficients.

[0014] Circular harmonic coefficient This is achieved through mode intensity compensation calculation of the characteristic beam, namely: , in, It is the mode strength. , It is a Bessel function of the first kind.

[0015] Relative harmonic coefficients were initially defined in the spherical harmonic domain, using the ratio of the power spectrum and cross power spectrum of each order harmonic coefficient to the zeroth harmonic coefficient to characterize the spatial information of a sound source. Extending this concept to the circular harmonic domain, different-order relative circular harmonic coefficients (DO-RCHC) can be represented by harmonic coefficients that have removed array characteristic dependencies: , in, To represent the expected value, superscript This represents the conjugate complex number. DO-RCHC can also be represented by characteristic beams: , Due to the modal intensity term and The existence of It inherently exhibits strangeness.

[0016] To overcome the singularity problem of DO-RCHC, a new positioning feature, the Same-Order Relative Circular Harmonic Coefficient (SO-RCHC), is constructed. SO-RCHC is defined as the ratio of the power spectrum to the cross-power spectrum of the same-order circular harmonic coefficients. Circular harmonic coefficients are obtained by mode intensity compensation from the characteristic beam. When the mode intensity value is too small, it leads to abnormal amplification of the circular harmonic coefficients, also known as the singularity problem. In the circular harmonic domain, the mode intensity is consistent across the same order. When constructing the SO-RCHC feature, the mode intensity term can be simultaneously canceled out from the ratio of the power spectrum to the cross-power spectrum of the same-order circular harmonic coefficients, thus avoiding the singularity problem. The definition of SO-RCHC is as follows: , and All are composed of harmonic coefficients Construction, i.e. and It is also independent of array characteristics. Utilizing the consistency of intensity of modes of the same order, SO-RCHC can also be represented by characteristic beams: , It can be seen that due to the consistency of mode strength at the same order, the mode strength terms cancel each other out. Unlike That would be affected by the singularity problem caused by the mode strength being too low.

[0017] Step 3: When environmental noise is present, perform frequency smoothing on SO-RCHC. The higher the array order, the more characteristic beams can be acquired, and the higher the spatial resolution of the array.

[0018] In the presence of ambient noise, SO-RCHC can be rewritten as: , in, For signal power, , For noise power, .

[0019] In practical applications, it is expected that... Estimated by frame time averaging: , right Frequency smoothing using SO-RCHC: , For frequency-smooth SO-RCHC localization features, smaller weighting values This will reduce robustness to positioning. Weighted values ​​and Inversely proportional, in the low-frequency range, the order... The bigger The smaller the value, the more sensitive the SO-RCHC (Solar-Resonant Continuous Harmonized Beam) is to noise. Order selection filters out reliable, higher-order characteristic beams that are less affected by noise for source location estimation.

[0020] Step 4: To address the vulnerability of high-order SO-RCHC beams to noise interference, an order selection method is proposed. The aim is to select reliable high-order characteristic beams less affected by noise for source location estimation, utilizing beams from order 1 to the highest reliable order. The direction of SO-RCHC information calculation can maintain positioning robustness and further improve positioning accuracy.

[0021] The SO-RCHC of order 1 encodes the azimuth information of the source, and an estimated direction vector containing the source direction information can be constructed: , in, and Let represent the real and imaginary parts of the complex number, respectively. Applying frequency-smoothed SO-RCHC to the grid-search arrival direction estimator expands the angular search range. Discretized For the nth grid point, grid points , The nth order direction vector can be written as: , Due to the periodicity of trigonometric functions, For any n, we have Limit the search range to Therefore, the set of search angle points is ,in This represents rounding up by comparing direction vectors. and direction vector set Using similarity to estimate target angle : , in, It follows the L2 norm because of the inherent periodicity of the direction vector. Within the range, there is not a unique candidate angle that satisfies the above formula, expressed as: , in, The set of candidate angles is defined by the target direction and... It consists of several virtual source directions. The first-order estimate is least affected by noise, while higher-order estimates are more susceptible to noise. To ensure reliability, a [missing information - likely a parameter] is introduced. Consistency criterion between the first-order estimated target direction and the second-order estimated target direction. If satisfied: , in, For the threshold, The order estimate is considered reliable. The maximum reliable order can be obtained by examining each order successively. The new estimated angle is recalculated using the order selection. : .

[0022] Using 1st order to The source direction is calculated from the smoothed SO-RCHC information of the first order frequency. This ambiguity could be due to the target source direction or the angle opposite to it, i.e., the inherent angular ambiguity of the circular harmonic coefficients of the same order. This ambiguity can be addressed by... and The Delay Sum Beamforming (DSB) algorithm is used to solve this problem in both directions, selecting the direction with the maximum power as the target source direction. This process results in only a minimal increase in computational complexity. For the steering angle... The beam power can be expressed as: , By comparison and An angle with a higher value is determined as the direction of arrival angle of the source.

[0023] The following is a specific embodiment for illustration.

[0024] The room dimensions were set to 9.7 m × 7.05 m × 3 m, with the array located at the center of the room measuring 4.85 m × 3.525 m × 1.5 m. The array was an open, uniform circular array consisting of 8 microphone elements with a radius of 4 cm. Twenty different speakers were randomly selected from the TIMIT speech library to form the test speech library (10 male and 10 female voices). A 1-second segment was randomly extracted from each speech as the test speech signal, with a sampling frequency of 16 kHz. Room reverberation was generated using the indoor acoustic impulse response based on the mirror source method. Noise was zero-mean and had a variance of [missing value]. The noise used is additive white Gaussian noise, and the noise is uncorrelated between different array elements; the signal and noise are also uncorrelated. The short-time Fourier transform has a frame length of 256 and a frame shift of 50%. The sound source was placed at 36 different locations for testing, with the horizontal angle of the sound source varying. Incrementally placed in sequence In the experiment, the distance from the array center to the sound source was 1.5m within the specified range. Five experiments were conducted at each location, with different speech signals provided by a test speech library for each experiment. Therefore, a total of 180 Monte Carlo experiments were performed under different test environments. The performance metric used was the root mean square error (RMSE), which measures the localization accuracy of each sound source localization method. RMSE is defined as follows: .

[0025] The proposed SO-RCHC localization method and its compatibility with DSB, TF-CHB, and S were evaluated through simulation experiments. 2 The positioning performance of the H-MUSIC positioning method under different simulation conditions is evaluated. To avoid spatial aliasing errors during circular harmonic transformation, the maximum harmonic order is determined by different selection criteria. S 2 H-MUSIC: Maximum characteristic beam order The frequency range of the processed signal is set to 300Hz~1365Hz; Other positioning methods: Maximum characteristic beam order The frequency range of the processed signal is set to 300Hz~4000Hz.

[0026] The running time of the proposed sound source localization method is compared with that of the comparative methods, as shown in Table 1: Table 1

[0027] Figures 2(a), 2(b), and 2(c) illustrate the RMSE of various localization methods under different reverberation times at signal-to-noise ratios (SNR) of 0 dB, 5 dB, and 10 dB. It can be seen that the performance of all algorithms decreases with reverberation time. The performance of SO-RCHC decreases with increasing signal-to-noise ratio (SNR), but SO-RCHC exhibits stronger robustness. At a SNR of 0 dB, other localization methods perform poorly in reverberant environments, with errors significantly higher than the SO-RCHC method, especially SNR. 2 The H-MUSIC method achieves an RMSE exceeding [a certain threshold] when the signal-to-noise ratio (SNR) is 0 dB and the reverberation time (RT60) is 0.8 s. TF-CHB mitigates singularity issues by using a Tikhonov regularization filter but is sensitive to noise. DSB offers good noise immunity but poor reverberation immunity. SO-RCHC delivers the most accurate and stable positioning results even in challenging environments with low signal-to-noise ratios and high reverberation.

[0028] Figure 3 It is a comparison of the intensity amplitudes of different order modes. Figure 4 The first three amplitude comparisons of DO-RCHC and SO-RCHC under noise-free and echo-free conditions are presented. It can be seen that SO-RCHC effectively solves the singularity problem inherent in open circular arrays caused by excessively small mode intensity amplitudes.

[0029] Based on the same inventive concept, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the aforementioned sound source localization method based on the same-order relative circular harmonic coefficients.

[0030] Based on the same inventive concept, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the aforementioned sound source localization method based on the same-order relative circular harmonic coefficients.

[0031] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0032] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0033] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0034] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0035] The above embodiments are merely illustrative of the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solutions based on the technical concept proposed in this invention shall fall within the scope of protection of this invention.

Claims

1. A sound source localization method based on the same-order relative circular harmonic coefficients, characterized in that, Includes the following steps: Step 1: Perform a short-time Fourier transform on the sound pressure signal acquired by the uniform open circular microphone array to convert the sound pressure signal to the short-time Fourier domain. Then, use the orthogonal circular harmonic function to transform the sound pressure signal converted to the short-time Fourier domain to the circular harmonic domain to obtain the characteristic beam. Step 2: Perform mode intensity compensation calculation on the characteristic beam to obtain the circular harmonic coefficients. Construct relative circular harmonic coefficients of the same order based on the circular harmonic coefficients for source location estimation. The relative circular harmonic coefficients of the same order are defined as the ratio of the power spectrum and cross power spectrum of the circular harmonic coefficients of the same order. Step 3: In the presence of environmental noise, perform frequency smoothing on the same-order relative circular harmonic coefficients constructed in Step 2 to obtain frequency-smoothed same-order relative circular harmonic coefficients. Step 4: Based on the frequency smoothing of the same order relative circular harmonic coefficients, a number selection method is used to select some characteristic beams from the characteristic beams for sound source azimuth estimation, and a delay summation beamforming algorithm is used to determine the final sound source arrival direction angle.

2. The sound source localization method based on the same-order relative circular harmonic coefficients according to claim 1, characterized in that, In step 1, the product is... Sound pressure signals are sampled using a uniform, open, circular microphone array consisting of [number] microphones, with an array radius of [radius]. The center of the circular microphone array is located at the origin of the xy plane, and the first... Azimuth angle of each microphone for: ， in, ; For the angle of incidence is The far-field sound source, the first The sound pressure signals from each microphone are modeled in the short-time Fourier domain as follows: ， Where k is the wave number and t is the time frame. , K is the total wave number, and T is the total time frame. For the short-time Fourier domain, the first The sound pressure signal from each microphone, For the sound source from the far field to the first The sound transfer function of a microphone For far-field sound source signals, Indicates the first The mean and variance of the data received by each microphone are zero. Unrelated noise; use The characteristic beam is obtained by circular harmonic transformation of the sound pressure field sampled by each microphone: ， in, for First-order characteristic beam, and They are respectively and Circular harmonic transformation, , , For order, For the largest order, .

3. The sound source localization method based on the same-order relative circular harmonic coefficients according to claim 2, characterized in that, In step 2, the formula for calculating the circular harmonic coefficient is as follows: ， in, for Circular harmonic coefficients, For mode strength, , It is a Bessel function of the first kind; Constructing relative circular harmonic coefficients of the same order based on circular harmonic coefficients as follows: ， in, To represent the expected value, superscript To represent conjugate complex numbers, for Circular harmonic coefficient.

4. The sound source localization method based on the same-order relative circular harmonic coefficients according to claim 3, characterized in that, The specific process of step 3 is as follows: In the presence of ambient noise, the same-order relative circular harmonic coefficients constructed in step 2 can be rewritten as: ， in, These are the relative circular harmonic coefficients of the same order. For signal power, , For noise power, ; pass The relative circular harmonic coefficients of the same order are estimated by time averaging of frames, and then the estimated relative circular harmonic coefficients of the same order are frequency smoothed, as shown in the formula: ， ， in, To estimate the relative circular harmonic coefficients of the same order, These are the frequency-smoothed relative circular harmonic coefficients of the same order. for Characteristic beam.

5. The sound source localization method based on the same-order relative circular harmonic coefficients according to claim 4, characterized in that, The specific process of step 4 is as follows: Based on the frequency-smoothed relative circular harmonic coefficients of the same order, an estimated direction vector containing source direction information is constructed. : ， in, and These represent the real and imaginary parts of a complex number, respectively. Search range of angle points Discretized For the grid point, for the , grid points The nth-order direction vector is written as: ， For any n, we have Let the set of search angle points be , This represents rounding up by comparing direction vectors. and direction vector set Using similarity to estimate target angle : ， in, It is the L2 norm. Within the range, there is not a unique candidate angle that satisfies the above formula, expressed as: ， in, Represents the set of candidate angles. It is a set of integers; Based on the consistency criterion between the nth-order estimated target direction and the first-order estimated target direction, from the maximum order... Begin by sequentially checking whether the current order satisfies the condition. , If the threshold is met, the current order is taken as the maximum reliable order h, and a new estimated angle is recalculated. : ， If the condition is not met, decrease the current order by one and continue checking if the condition is met. until the maximum reliable order is found; New estimation angle Given the direction of arrival (DOA) of the sound source, or the angle opposite to the direction of arrival, the final DOA is determined using a delay-summation beamforming algorithm. ， in, or , for The beam power, by comparison and ,Will and The maximum value in As the final direction angle of arrival of the sound source.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the sound source localization method based on the same-order relative circular harmonic coefficients as described in any one of claims 1 to 5.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the sound source localization method based on the same-order relative circular harmonic coefficients as described in any one of claims 1 to 5.

Images