Double-microphone wind noise detection method and system

Abstract

The invention relates to the technical field of sound detection, and discloses a double-microphone wind noise detection method and system, and the method comprises the steps: collecting a time domain audio signal for deploying double microphones, and carrying out the signal iteration processing of the time domain audio signal, and obtaining a target signal; performing short-time segmentation and short-time frequency domain transformation on the target signal to obtain a time-frequency spectrogram, identifying a background noise frequency spectrum in the time-frequency spectrogram, performing smooth correction to obtain a corrected frequency spectrum, identifying frequency spectrum components in the time-frequency spectrogram, and obtaining a target frequency spectrum; performing signal independent component analysis on the target frequency spectrum to identify a wind noise signal, identifying a total wind noise signal of the target frequency spectrum, calculating the signal intensity of the total wind noise signal, and analyzing the frequency characteristic of the total wind noise signal; and performing time-space correlation analysis on the double microphones to obtain a wind noise propagation rule, and performing real-time wind noise detection on the double microphones by using signal strength, frequency characteristics and the wind noise propagation rule. According to the invention, the overall wind noise detection effect of the double microphones can be improved.

Description

Technical Field

[0001] The invention relates to a dual-microphone wind noise detection method and system, belonging to the technical field of sound detection. Background Art

[0002] In the field of audio acquisition and processing, wind noise is a key interference factor that affects sound quality. Accurately detecting wind noise is of great significance for improving the clarity, intelligibility and accuracy of subsequent processing of audio signals. Dual-microphone technology has shown great potential in wind noise detection due to its unique architecture and working principle. The wind noise detection method using dual microphones can effectively utilize the signal differences collected by the two microphones to accurately identify the wind noise components in the environment. By analyzing and comparing the signals received by the dual microphones, the dynamic changes of wind noise can be tracked in real time, thereby providing accurate wind noise information for the audio processing system, so as to take targeted noise reduction measures and significantly improve the audio quality. This not only helps to improve the effect of voice communication, such as making the sound clearer and more natural in scenarios such as teleconferencing and voice assistants, but also provides pure audio signals for applications such as audio recording and monitoring, and enhances the performance of related equipment and systems.

[0003] At present, common wind noise detection methods are mostly based on a single microphone or a simple signal processing algorithm. When a single microphone detects wind noise, it is difficult to accurately distinguish wind noise from complex environmental noise due to the lack of comparative reference, resulting in low detection accuracy. Some simple signal processing algorithms have poor adaptability when faced with changeable and complex wind noise characteristics, and cannot meet the requirements for wind noise detection accuracy and real-time performance in practical applications. Especially in strong wind environments or when there are multiple types of noise in the audio signal at the same time, these traditional methods are often difficult to work effectively, resulting in less than ideal wind noise detection results, which in turn affects the overall effect of audio processing. Summary of the invention

[0004] The present invention provides a dual-microphone wind noise detection method and system, the main purpose of which is to improve the overall effect of the dual-microphone wind noise detection.

[0005] To achieve the above object, the present invention provides a dual-microphone wind noise detection method, comprising: Acquire dual microphones for wind noise detection, perform acoustic spatial deployment on the dual microphones to obtain deployed dual microphones, collect time domain audio signals of the deployed dual microphones using a preconfigured phase-locked loop synchronization technology, analyze expected signals of the time domain audio signals, and perform signal iterative processing on the time domain audio signals based on the expected signals to obtain target signals; The target signal is segmented into short time periods to obtain segmented signals, the segmented signals are transformed into short-time frequency domain to obtain a time-frequency spectrum of the segmented signal, a silent segment spectrum in the time-frequency spectrum is identified, a spectrum feature of the silent segment spectrum is analyzed, a background noise spectrum of the time-frequency spectrum is determined based on the spectrum feature, the background noise spectrum is smoothed and corrected to obtain a corrected spectrum, and a spectrum component in the time-frequency spectrum is identified based on the corrected spectrum to obtain a target spectrum; Calculating a linear transformation matrix of the target spectrum, performing signal independent component analysis on the target spectrum based on the linear transformation matrix to identify a wind noise signal in the target spectrum, analyzing a wind noise signal feature corresponding to the wind noise signal, identifying a full wind noise signal of the target spectrum based on the wind noise signal feature, calculating a signal strength of the full wind noise signal, and analyzing a frequency characteristic of the full wind noise signal; The spatial positions and sampling times of the dual microphones are queried, and based on the spatial positions and the sampling times, spatiotemporal correlation analysis is performed on the dual microphones to obtain a wind noise propagation law, and a wind noise dynamic threshold of the dual microphones is constructed by using the signal strength, the frequency characteristics, and the wind noise propagation law, and real-time wind noise detection is performed on the dual microphones based on the wind noise dynamic threshold.

[0006] Optionally, the performing acoustic spatial deployment on the dual microphones to obtain deployed dual microphones includes: Querying the sound signal wavelength and expected spatial resolution of the dual microphones to determine the arrangement spacing of the dual microphones; Identifying the sound source direction and sound change range of the dual microphones to determine the arrangement angle of the dual microphones; Based on the arrangement spacing and the arrangement angle, the dual microphones are preliminarily spatially arranged to obtain initial arranged microphones; Signal synchronization verification is performed on the initially deployed microphones, and when a synchronization result of the signal synchronization verification satisfies a preset synchronization value, dual microphones are deployed.

[0007] Optionally, analyzing the expected signal of the time-domain audio signal includes: Removing a DC offset of the time-domain audio signal to obtain a preliminary processed signal; Performing frequency screening on the preliminary processed signal to obtain a screened signal; Performing noise estimation on the screening signal to obtain estimated noise; The estimated noise is removed by using an adaptive filtering method to obtain a clean signal; analyzing a clean signal feature of the clean signal; Based on the clean signal characteristics, a desired signal of the time-domain audio signal is determined.

[0008] Optionally, performing a short-time frequency domain transform on the segmented signal to obtain a time-frequency spectrum diagram of the segmented signal includes: The segmented signal is transformed into a frequency domain in a short time period using the following formula to obtain a frequency domain transformed signal: in, represents the frequency domain transformed signal, represents the segmentation signal, w() represents the window function, N represents the length of the window in the window function, m represents the time index, R represents the sliding step of the window, Y represents the frequency index, j represents the imaginary unit, and n represents the time index of the segmentation signal in the window function; The frequency domain transformed signals are arranged in time sequence to obtain a time-frequency spectrum diagram of the segmented signals.

[0009] Optionally, the identifying of the silent segment spectrum in the time-spectrogram includes: The spectrum energy of the time window in the time-spectrogram is calculated using the following formula: in, represents the spectrum energy, represents the i-th window in the time window, represents the kth frequency in the time-spectrum diagram, The spectrum diagram is and The spectrum amplitude at , K represents the number of frequencies in the time spectrum graph; constructing a silent energy threshold of the time-spectrogram; Based on the silence energy threshold and the spectrum energy, a silence segment spectrum in the time-spectrogram is determined.

[0010] Optionally, analyzing the spectrum characteristics of the silent segment spectrum includes: Constructing a spectrum curve of the silent segment spectrum; Analyzing the spectrum shape characteristics of the silent segment spectrum using the spectrum curve; Identifying the energy distribution state of the silent segment spectrum on the frequency axis; Based on the energy distribution state, identifying the frequency distribution characteristics of the silent segment spectrum; Calculating the statistical value of the silent segment spectrum; Analyze the spectrum dispersion of the silent segment spectrum based on the statistical value; Identifying the spectrum statistical features of the silent segment spectrum based on the spectrum dispersion; Calculating the average value of the absolute value of the amplitude difference between adjacent frequency points in the silent segment spectrum to analyze the spectrum smoothness characteristics of the silent segment spectrum, The spectrum characteristics of the silent segment spectrum are determined based on the spectrum shape characteristics, the frequency distribution characteristics, the spectrum statistical characteristics and the spectrum smoothness characteristics.

[0011] Optionally, the identifying the full wind noise signal of the target spectrum based on the wind noise signal feature includes: Querying the time domain features, frequency domain features and high-order statistical features of the wind noise signal to obtain comprehensive features; Perform wind noise feature matching on the target spectrum based on the comprehensive feature to obtain a matching feature; Constructing a wind noise judgment rule of the matching features by using a weighting method; Based on the wind noise judgment rule, a preliminary judgment is made on the full wind noise signal of the target spectrum to obtain a preliminary full wind noise signal; The preliminary full wind noise signal is subjected to continuous regional integration analysis to determine the full wind noise signal of the target spectrum.

[0012] Optionally, calculating the signal strength of the full wind noise signal includes: The mean square value of the total wind noise signal is calculated using the following formula:

[0013] Among them, RMS represents the mean square value, M represents the sequence number of the full wind noise signal, represents the bth signal in the full wind noise signal, Analyzing the time domain signal strength of the full wind noise signal by using the mean square value; The sum of the squares of the spectrum amplitudes of the full wind noise signal is calculated using the following formula:

[0014] Among them, V represents the sum of squares of spectrum amplitude, H represents the number of frequency points of the full wind noise signal, represents the spectrum amplitude at frequency point h, Analyzing the frequency domain signal strength of the full wind noise signal by using the sum of squares of the spectrum amplitude; The signal strength of the full wind noise signal is determined according to the time domain signal strength and the frequency domain signal strength.

[0015] Optionally, performing spatiotemporal correlation analysis on the dual microphones based on the spatial position and the sampling time to obtain a wind noise propagation law includes: Query the signal sequence of the dual microphones based on the sampling time; Aligning the signal sequences on the time axis to obtain an alignment signal; analyzing a signal delay of the alignment signal; Calculating the propagation distance of the dual microphones based on the spatial position; Analyzing the propagation speed and propagation direction of the dual microphones based on the signal delay and the propagation distance; Query the signal strength of the dual microphones; Analyzing the signal attenuation trend of the dual microphones based on the signal strength and the propagation speed; The wind noise propagation law of the dual microphones is determined based on the propagation speed, the propagation direction and the signal attenuation trend.

[0016] In order to solve the above problem, the present invention also provides a dual-microphone wind noise detection system, the system comprising: A signal processing module is used to obtain dual microphones to be detected for wind noise, perform acoustic spatial deployment on the dual microphones to obtain deployed dual microphones, collect time domain audio signals of the deployed dual microphones using a preconfigured phase-locked loop synchronization technology, analyze an expected signal of the time domain audio signal, and perform signal iterative processing on the time domain audio signal based on the expected signal to obtain a target signal; a spectrum component identification module, for performing short-time segmentation on the target signal to obtain segmented signals, performing short-time frequency domain transformation on the segmented signals to obtain a time-frequency spectrum of the segmented signals, identifying a silent segment spectrum in the time-frequency spectrum, analyzing a spectrum feature of the silent segment spectrum, determining a background noise spectrum of the time-frequency spectrum based on the spectrum feature, performing smoothing correction on the background noise spectrum to obtain a corrected spectrum, identifying spectrum components in the time-frequency spectrum based on the corrected spectrum to obtain a target spectrum; a signal analysis module, configured to calculate a linear transformation matrix of the target spectrum, perform signal independent component analysis on the target spectrum based on the linear transformation matrix to identify a wind noise signal in the target spectrum, analyze a wind noise signal feature corresponding to the wind noise signal, identify a full wind noise signal of the target spectrum based on the wind noise signal feature, calculate a signal strength of the full wind noise signal, and analyze a frequency characteristic of the full wind noise signal; The wind noise detection module is used to query the spatial position and sampling time of the dual microphones, perform spatiotemporal correlation analysis on the dual microphones based on the spatial position and the sampling time, obtain the wind noise propagation law, use the signal strength, the frequency characteristics and the wind noise propagation law to construct the wind noise dynamic threshold of the dual microphones, and perform real-time wind noise detection on the dual microphones based on the wind noise dynamic threshold.

[0017] Compared with the problems described in the background technology, the embodiment of the present invention performs acoustic spatial deployment of dual microphones, determines the deployment spacing by querying the sound signal wavelength and the expected spatial resolution, clarifies the deployment angle according to the direction of the sound source and the range of change, and performs signal synchronization verification on the initially deployed microphones after the initial deployment is completed, to ensure that the dual microphones can effectively capture information such as the time difference and intensity difference of the sound; further, the embodiment of the present invention performs short-time frequency domain transformation on the segmented signal, converts the segmented signal into a frequency domain transformation signal using a specific formula, and then arranges it in chronological order to obtain a time-frequency spectrum diagram, which helps to discover the change law of the signal frequency component over time, and analyzes the spectrum characteristics of the silent segment, quantifies the spectrum shape, frequency distribution, spectrum discreteness, spectrum smoothness, etc., to provide a basis for estimating the background noise spectrum; further, the present invention calculates the linear transformation matrix of the target spectrum (which can be obtained by the ICA algorithm), decomposes the target spectrum signal into different independent components, creates conditions for identifying the wind noise signal, and calculates the signal strength of the full wind noise signal, and calculates the mean square value and the time-frequency domain signal strength respectively by formulas, thereby providing quantitative indicators for subsequent noise reduction strategy formulation and system performance evaluation. At the same time, the full wind noise signal is Fourier transformed, the spectrum is observed and the energy proportion of each frequency interval is calculated, and its frequency characteristics are analyzed to provide a basis for designing noise reduction technology; further, the present invention provides geometric and time dimension information for wind noise propagation analysis by querying the spatial position and sampling time of the dual microphones, and performs spatiotemporal correlation analysis on the dual microphones to obtain the propagation speed and direction of the wind noise, analyzes the signal attenuation trend in combination with the signal strength, and determines the law of wind noise propagation to help users understand the propagation characteristics of wind noise in the actual environment, providing a physical basis for subsequent wind noise processing and detection, and performing real-time wind noise detection on the dual microphones based on the dynamic threshold of wind noise, adjusting the detection standard according to the real-time changes of wind noise, improving the accuracy and real-time performance of the detection, thereby improving the overall effect of wind noise detection by the dual microphones. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic flow chart of a dual-microphone wind noise detection method provided by an embodiment of the present invention; Figure 2 A schematic diagram of functional modules for implementing a dual-microphone wind noise detection system provided in an embodiment of the present invention.

[0019] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0020] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0021] The embodiment of the present application provides a dual-microphone wind noise detection method. The execution subject of the dual-microphone wind noise detection method includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the dual-microphone wind noise detection method can be executed by software or hardware installed in a terminal device or a server device. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0022] Embodiment 1:

[0023] Reference Figure 1 FIG. 1 is a flow chart of a dual-microphone wind noise detection method provided in an embodiment of the present invention. In this embodiment, the dual-microphone wind noise detection method includes: S1. Acquire dual microphones for wind noise detection, perform acoustic spatial deployment on the dual microphones to obtain deployed dual microphones, collect time domain audio signals of the deployed dual microphones using a preconfigured phase-locked loop synchronization technology, analyze expected signals of the time domain audio signals, and perform signal iterative processing on the time domain audio signals based on the expected signals to obtain target signals.

[0024] The embodiment of the present invention can determine the hardware device used for wind noise detection by obtaining the dual microphones to be detected for wind noise, providing a basis for subsequent detection work. The dual microphones refer to a combination of two microphones used in an audio acquisition system. The two microphones are independent of each other in space, but work together in function to collect sound signals from the surrounding environment. They can be microphones of the same model, or they can be a combination of different types of microphones selected according to specific application scenarios and user requirements.

[0025] Furthermore, the embodiment of the present invention performs acoustic spatial deployment of the dual microphones to obtain information such as the time difference and intensity difference of sound arrival, which is helpful for subsequent analysis of the direction and characteristics of the sound source.

[0026] As an embodiment of the present invention, the acoustic spatial deployment of the dual microphones to obtain deployed dual microphones includes: querying the sound signal wavelength and the expected spatial resolution of the dual microphones to determine the arrangement spacing of the dual microphones, identifying the sound source direction and the sound change range of the dual microphones to determine the arrangement angle of the dual microphones, based on the arrangement spacing and the arrangement angle, performing preliminary spatial deployment of the dual microphones to obtain initial deployed microphones, performing signal synchronization verification on the initial deployed microphones, and when the synchronization result of the signal synchronization verification meets the preset synchronization value, the dual microphones are deployed.

[0027] The sound signal wavelength refers to a mechanical wave that has a certain wavelength when propagating in a medium, and the expected spatial resolution refers to the minimum spatial interval or angular range in which we expect to be able to distinguish the wind noise source in wind noise detection.

[0028] Optionally, the process of querying the sound signal wavelength and expected spatial resolution of the dual microphones to determine the arrangement spacing of the dual microphones is: according to the frequency range of the wind noise signal expected to be collected by the dual microphones, the sound signal wavelength is calculated using the sound wave calculation formula, and then, in combination with the expected spatial resolution, the arrangement spacing is set to a certain proportion of the corresponding wavelength (such as 1 / 4 or 1 / 3 wavelength) according to the required detection accuracy. The process of identifying the sound source direction and sound change range of the dual microphones to determine the arrangement angle of the dual microphones is: according to the direction from which the wind may blow, such as whether it is omnidirectional or a specific direction range, and the possible change range of sound intensity and frequency, the arrangement angle of the dual microphones is adjusted to an angle that can effectively capture wind noise signals from different directions. For example, for omnidirectional wind sources, it can be set to an angle of 90° or 120°. The signal synchronization verification of the initially deployed microphones can collect the dual microphone collection signals through signal collection technology, and determine the signal synchronization value by comparing the time difference or phase difference of the two microphone collection signals. When the synchronization result of the signal synchronization verification meets the preset synchronization value, it means that the time difference or the phase difference is within the preset nanosecond synchronization value, then it is considered that the synchronization result meets the requirement, and the dual microphone at this time is the deployed dual microphone.

[0029] The embodiment of the present invention can ensure the temporal consistency of the collected signal by collecting the time domain audio signal of the deployed dual microphones using the preconfigured phase-locked loop synchronization technology, avoid errors introduced by sampling asynchrony, and make subsequent signal analysis and processing more accurate.

[0030] Among them, the phase-locked loop synchronization technology refers to a technology widely used in electronic systems for frequency synthesis and signal synchronization. In the context of dual-microphone wind noise detection, it is mainly used to ensure that the audio signals collected by the two microphones are highly synchronized in time.

[0031] The embodiment of the present invention can determine the direction and target of subsequent signal processing by analyzing the expected signal of the time-domain audio signal, that is, it is clear in which direction the original time-domain audio signal is to be processed.

[0032] The expected signal refers to an audio signal that meets the user's expectations and has been ideally processed.

[0033] As an embodiment of the present invention, the analysis of the expected signal of the time domain audio signal includes: removing the DC offset of the time domain audio signal to obtain a preliminary processed signal, performing frequency screening on the preliminary processed signal to obtain a screened signal, performing noise estimation on the screened signal to obtain an estimated noise, removing the estimated noise by using an adaptive filtering method to obtain a clean signal, analyzing the clean signal characteristics of the clean signal, and determining the expected signal of the time domain audio signal based on the clean signal characteristics.

[0034] Optionally, the removal of the DC offset of the time domain audio signal to obtain the preliminary processed signal can be achieved by calculating the average value of the time domain audio signal and subtracting the average value from each sample value of the original signal so that the signal fluctuates around the zero mean; the frequency screening of the preliminary processed signal to obtain the screened signal can be achieved by using a filter (such as a low-pass, high-pass or band-pass filter); the noise estimation of the screened signal to obtain the estimated noise can be achieved by performing a statistical analysis on the silent segments in the screened signal (which can be judged based on the signal energy), such as calculating the mean, variance and other statistics of these silent segments, or using an adaptive method such as the least mean square (LMS) algorithm to estimate the noise component in the signal; the analysis of the clean signal characteristics of the clean signal can be achieved by extracting the characteristics of the clean signal from the time domain and frequency domain, such as the signal envelope and pulse duration in the time domain, and the spectrum peak and bandwidth in the frequency domain; the expected signal can determine the clean signal with the above-mentioned signal envelope, pulse duration, spectrum peak and bandwidth in the frequency domain as the expected signal.

[0035] Furthermore, the embodiment of the present invention can gradually improve the signal quality by performing iterative signal processing on the time domain audio signal based on the expected signal to obtain a target signal, so that the final target signal better meets the requirements of wind noise detection.

[0036] Optionally, the process of performing iterative signal processing on the time domain audio signal based on the expected signal to obtain the target signal is as follows: first, according to the characteristics of the expected signal, initial parameters are set for the signal processing algorithm (such as an adaptive filtering algorithm). For example, when using the least mean square (LMS) algorithm, the filter coefficients are initialized to small random values, and a suitable step size factor is set. Then, the time domain audio signal is used as input and iterative processing is performed using the selected signal processing algorithm (such as the LMS algorithm). In each iteration, the processing parameters are updated according to the difference between the expected signal and the current processing result, the processing method is continuously adjusted, and the signal is gradually optimized until a preset convergence condition is met or a predetermined number of iterations is reached, and finally the target signal is obtained.

[0037] S2. Segment the target signal into short time periods to obtain segmented signals, perform short-time frequency domain transformation on the segmented signals to obtain a time-frequency spectrum of the segmented signal, identify the silent segment spectrum in the time-frequency spectrum, analyze the spectral features of the silent segment spectrum, determine the background noise spectrum of the time-frequency spectrum based on the spectral features, perform smoothing correction on the background noise spectrum to obtain a corrected spectrum, identify the spectral components in the time-frequency spectrum based on the corrected spectrum, and obtain the target spectrum.

[0038] In the embodiment of the present invention, by performing short-time segmentation on the target signal to obtain segmented signals, the target signal can be divided into a series of short-time sub-signals, so as to subsequently perform detailed analysis on signal characteristics at different moments.

[0039] Optionally, the segmented signal can be obtained by selecting a suitable window length (such as 256 or 512 sample points) and a window function (such as a Hamming window), starting from the starting position of the target signal, and intercepting the signal in sequence according to the window length. Adjacent windows can be set to a certain overlapping part (such as 50% overlap).

[0040] In the embodiment of the present invention, by performing short-time frequency domain transformation on the segmented signal to obtain a time-frequency spectrum diagram of the segmented signal, the time-frequency spectrum diagram can help the user discover the changing rules of the signal frequency components at different times.

[0041] As an embodiment of the present invention, performing a short-time frequency domain transformation on the segmented signal to obtain a time-frequency spectrum of the segmented signal includes: performing a short-time frequency domain transformation on the segmented signal using the following formula to obtain a frequency domain transformation signal:

[0042] in, represents the frequency domain transformed signal, represents the segmentation signal, w() represents the window function, N represents the length of the window in the window function, m represents the time index, R represents the sliding step of the window, Y represents the frequency index, j represents the imaginary unit, and n represents the time index of the segmentation signal in the window function; The frequency domain transformed signals are arranged in time sequence to obtain a time-frequency spectrum diagram of the segmented signals.

[0043] Furthermore, the embodiment of the present invention can provide a key data source for extracting background noise information by identifying the silent segment spectrum in the time-spectrogram, and can provide a basis for subsequent noise processing and signal purification, thereby achieving effective noise reduction and signal separation.

[0044] As an embodiment of the present invention, the identifying of the silent segment spectrum in the time-spectrogram includes: calculating the spectrum energy of the time window in the time-spectrogram using the following formula:

[0045] in, represents the spectrum energy, represents the i-th window in the time window, represents the kth frequency in the time-spectrum diagram, The spectrum diagram is and The spectrum amplitude at , K represents the number of frequencies in the time spectrum graph; A silence energy threshold of the time-spectrogram is constructed, and based on the silence energy threshold and the spectrum energy, a silence segment spectrum in the time-spectrogram is determined.

[0046] Optionally, the silent energy threshold determines a suitable silent energy threshold by analyzing the overall energy distribution of the time-frequency spectrum or based on past experience (specific settings need to be combined with actual applications). The silent segment spectrum in the time-frequency spectrum graph is determined based on the silent energy threshold and the spectrum energy. The spectrum energy calculated for each time window can be compared with the silent energy threshold. The spectrum corresponding to the time window with energy lower than the threshold is determined as the silent segment spectrum.

[0047] The embodiment of the present invention can grasp the basic characteristics of the background noise, such as the main frequency range of the background noise and the energy levels at different frequencies, by analyzing the spectral characteristics of the silent segment spectrum, so as to provide specific quantitative indicators for the subsequent estimation of the background noise spectrum, and help to accurately describe the nature of the background noise.

[0048] As an embodiment of the present invention, the analysis of the spectral characteristics of the silent segment spectrum includes: constructing a spectral curve of the silent segment spectrum, using the spectral curve to analyze the spectral shape characteristics of the silent segment spectrum, identifying the energy distribution state of the silent segment spectrum on the frequency axis, identifying the frequency distribution characteristics of the silent segment spectrum based on the energy distribution state, calculating the statistical value of the silent segment spectrum, analyzing the spectral discreteness of the silent segment spectrum based on the statistical value, identifying the spectral statistical characteristics of the silent segment spectrum based on the spectral discreteness, calculating the average value of the absolute value of the amplitude difference between adjacent frequency points in the silent segment spectrum to analyze the spectral smoothness characteristics of the silent segment spectrum, and determining the spectral characteristics of the silent segment spectrum based on the spectral shape characteristics, the frequency distribution characteristics, the spectral statistical characteristics and the spectral smoothness characteristics.

[0049] Among them, the spectrum curve refers to a visual representation, which is a curve used to show the amplitude change of the silent segment spectrum in the frequency dimension; the energy distribution state refers to the distribution of the energy of the silent segment spectrum at different frequencies; and the spectrum dispersion refers to an indicator used to measure the dispersion degree of the amplitude change of the silent segment spectrum at different frequency points.

[0050] Optionally, the spectrum curve can be formed by connecting the data points with the frequency as the horizontal axis and the amplitude of the silent segment spectrum at each frequency point (which can be the average amplitude, peak amplitude, etc.) as the vertical axis; the spectrum shape feature can be determined by observing the shape of the spectrum curve to determine whether it is flat (which may be white noise), has a peak (which may be narrowband noise), or other irregular shapes, and analyzing the ups and downs of the curve and the position and number of peaks, etc. to determine the spectrum shape feature; the energy distribution state can calculate the energy of each frequency point or frequency interval (such as the sum of squares of the amplitude), observe the high and low distribution of energy on the frequency axis, and determine which frequency ranges the energy is mainly concentrated in and which frequency ranges have lower energy; the frequency distribution feature can find the frequency area with higher energy as the main frequency distribution range based on the energy distribution. If the energy is concentrated near a specific frequency, it can be determined. Narrowband frequency distribution, if the energy is relatively evenly distributed in a wider frequency range, it is a broadband frequency distribution; the spectrum dispersion can be obtained by calculating the statistical values ​​of the spectrum amplitude, such as variance and standard deviation, wherein the larger the variance or standard deviation, the higher the spectrum dispersion, indicating that the greater the fluctuation of the spectrum amplitude at different frequency points, and vice versa; the spectrum statistical features can be identified based on the value of the spectrum dispersion, combined with other factors such as the spectrum shape, statistical features, for example, high dispersion and irregular spectrum shape may indicate complex and changeable noise, low dispersion and relatively regular spectrum shape may indicate relatively stable noise; the spectrum smoothness feature can be obtained by successively calculating the absolute value of the amplitude difference of adjacent frequency points, and then calculating the average value for analysis, wherein a smaller average value indicates high spectrum smoothness and small amplitude change of adjacent frequency points, and a larger average value indicates low spectrum smoothness and large amplitude change.

[0051] The embodiment of the present invention can help remove noise from noisy signals and lay a foundation for extracting useful signals by determining the background noise spectrum of the time-spectrogram based on the spectral features, thereby being able to more accurately separate target signals (such as wind noise, speech, etc.) from mixed signals.

[0052] Optionally, the background noise spectrum of the time-frequency spectrum diagram can be determined based on the spectrum characteristics of the silent segment spectrum obtained by analyzing the spectrum characteristics of the silent segment spectrum, and the amplitude of the silent segment spectrum is statistically averaged in frequency. For each frequency point, the spectrum amplitudes of all silent segments at the frequency point are added, and then divided by the number of silent segments to obtain the background noise spectrum amplitude of the frequency point, and the background noise spectrum of the entire time-frequency spectrum diagram is determined by analogy.

[0053] The embodiment of the present invention can make the background noise spectrum more stable and continuous by smoothing and correcting the background noise spectrum to obtain a corrected spectrum, improve the reliability of noise estimation, reduce the spectrum estimation deviation caused by local noise or measurement error, and further improve the accuracy and stability of subsequent processing.

[0054] Optionally, the smoothing correction of the background noise spectrum to obtain the corrected spectrum may be achieved by utilizing a median filtering method.

[0055] The embodiment of the present invention obtains the target spectrum by identifying the spectral components in the time-spectrogram based on the corrected spectrum, so that the spectrum of the target signal (such as a wind noise signal, a speech signal, etc.) can be presented more clearly and the spectral characteristics of the target signal can be reflected more clearly.

[0056] Optionally, the identifying of the spectrum components in the time-spectrogram based on the modified spectrum to obtain the target spectrum may be achieved by spectral subtraction.

[0057] S3. Calculate a linear transformation matrix of the target spectrum, perform signal independent component analysis on the target spectrum based on the linear transformation matrix to identify a wind noise signal in the target spectrum, analyze wind noise signal characteristics corresponding to the wind noise signal, identify a full wind noise signal of the target spectrum based on the wind noise signal characteristics, calculate a signal strength of the full wind noise signal, and analyze a frequency characteristic of the full wind noise signal.

[0058] The embodiment of the present invention can find a new set of basis vectors by calculating the linear transformation matrix of the target spectrum, decompose the signal in the target spectrum into different independent components, and then decompose the complex signal into relatively independent parts, creating conditions for finding the wind noise signal. The linear transformation matrix refers to a matrix used to convert the original signal representation from one space to another space.

[0059] Optionally, the linear transformation matrix can be calculated by an ICA algorithm.

[0060] The embodiment of the present invention can help users extract wind noise signals from the target spectrum by performing signal independent component analysis on the target spectrum based on the linear transformation matrix to identify wind noise signals in the target spectrum, provide a purer wind noise signal source for subsequent wind noise analysis and processing, and improve the identifiability and analyzability of wind noise signals.

[0061] Optionally, the process of performing signal independent component analysis on the target spectrum based on the linear transformation matrix to identify the wind noise signal in the target spectrum is as follows: first, multiplying the target spectrum data by the linear transformation matrix for linear transformation to obtain a transformed signal matrix; then, calculating the statistical characteristics (such as kurtosis) of each independent component, and comparing them with known wind noise characteristics to screen wind noise signals; finally, verifying the results by cross-validation and other methods. If they are inaccurate, the indicator threshold can be adjusted or the recognition process can be improved to optimize wind noise signal recognition.

[0062] The embodiment of the present invention helps the user to more accurately grasp the nature of the wind noise signal by analyzing the wind noise signal characteristics corresponding to the wind noise signal.

[0063] Optionally, the wind noise signal characteristics can be obtained through deep learning model analysis.

[0064] The embodiment of the present invention can expand the locally identified wind noise signal to the entire target spectrum by identifying the full wind noise signal of the target spectrum based on the wind noise signal characteristics, and obtain a complete wind noise signal distribution, thereby fully understanding the wind noise situation in the target spectrum.

[0065] As an embodiment of the present invention, identifying the full wind noise signal of the target spectrum based on the wind noise signal characteristics includes: querying the time domain characteristics, frequency domain characteristics and high-order statistical characteristics of the wind noise signal to obtain comprehensive characteristics, matching the wind noise characteristics of the target spectrum based on the comprehensive characteristics to obtain matching characteristics, constructing a wind noise judgment rule for the matching characteristics using a weighting method, performing a preliminary judgment on the full wind noise signal of the target spectrum based on the wind noise judgment rule to obtain a preliminary full wind noise signal, and performing continuous regional integration analysis on the preliminary full wind noise signal to determine the full wind noise signal of the target spectrum.

[0066] Optionally, the matching feature can be obtained by comparing the features of each part of the target spectrum with the comprehensive features one by one, judging the similarity, and determining the degree of conformity of each part with the wind noise feature; the wind noise judgment rule can be obtained by assigning weights to the matching features corresponding to the time domain, frequency domain and high-order statistics, such as frequency domain weight 0.5, time domain 0.3, and high-order statistics 0.2, calculating the comprehensive score according to the weights, and setting the score threshold as the wind noise judgment rule; the preliminary full wind noise signal can be obtained by calculating the comprehensive score for each part of the target spectrum according to the wind noise judgment rule, and the part with a score higher than the threshold is judged as a wind noise signal.

[0067] The embodiment of the present invention can provide a quantitative index for subsequent noise reduction strategy formulation and system performance evaluation by calculating the signal strength of the full wind noise signal.

[0068] As an embodiment of the present invention, the calculating the signal strength of the full wind noise signal includes: calculating the mean square value of the full wind noise signal using the following formula:

[0069] Among them, RMS represents the mean square value, M represents the sequence number of the full wind noise signal, represents the bth signal in the full wind noise signal, Analyzing the time domain signal strength of the full wind noise signal by using the mean square value; The sum of the squares of the spectrum amplitudes of the full wind noise signal is calculated using the following formula:

[0070] Among them, V represents the sum of squares of spectrum amplitude, H represents the number of frequency points of the full wind noise signal, represents the spectrum amplitude at frequency point h, Analyzing the frequency domain signal strength of the full wind noise signal by using the sum of squares of the spectrum amplitude; The signal strength of the full wind noise signal is determined according to the time domain signal strength and the frequency domain signal strength.

[0071] Furthermore, the embodiment of the present invention can further understand the generation mechanism and propagation characteristics of wind noise by analyzing the frequency characteristics of the full wind noise signal, provide a basis for designing more effective wind noise suppression filters and other noise reduction technologies, and also provide more detailed information for the classification and evaluation of wind noise.

[0072] Optionally, the analysis of the frequency characteristics of the full wind noise signal can be performed by performing Fourier transform on the full wind noise signal to obtain a spectrum diagram, and observing its shape, frequency range and spectrum peak position; secondly, dividing the frequency axis into multiple intervals, and calculating the energy proportion of each interval to determine.

[0073] S4. Query the spatial positions and sampling times of the dual microphones, perform spatiotemporal correlation analysis on the dual microphones based on the spatial positions and the sampling times to obtain a wind noise propagation law, construct a wind noise dynamic threshold of the dual microphones by using the signal strength, the frequency characteristics and the wind noise propagation law, and perform real-time wind noise detection on the dual microphones based on the wind noise dynamic threshold.

[0074] The embodiment of the present invention can provide geometric information for subsequent wind noise propagation analysis by querying the spatial positions and sampling moments of the dual microphones, understand their relative positions in space, and ensure accurate correspondence of signal data at each moment in subsequent analysis, thereby providing a time dimension basis for spatiotemporal correlation analysis.

[0075] Furthermore, the embodiment of the present invention performs spatiotemporal correlation analysis on the dual microphones based on the spatial position and the sampling time to obtain the wind noise propagation law, which can help users understand the propagation characteristics of wind noise in an actual environment and provide a physical basis for subsequent wind noise processing and detection.

[0076] As an embodiment of the present invention, the dual microphones are subjected to spatiotemporal correlation analysis based on the spatial position and the sampling time to obtain a wind noise propagation law, including: querying the signal sequence of the dual microphones based on the sampling time, aligning the signal sequence on the time axis to obtain an aligned signal, analyzing the signal delay of the aligned signal, calculating the propagation distance of the dual microphones based on the spatial position, analyzing the propagation speed and propagation direction of the dual microphones based on the signal delay and the propagation distance, querying the signal strength of the dual microphones, analyzing the signal attenuation trend of the dual microphones based on the signal strength and the propagation speed, and determining the wind noise propagation law of the dual microphones based on the propagation speed, the propagation direction and the signal attenuation trend.

[0077] Optionally, the signal sequences are aligned on the time axis to obtain the aligned signal by finding signal feature points (such as peak points, zero crossing points) and calibrating the two signal sequences on the time axis; the signal delay of the analysis of the aligned signal can be calculated and analyzed using a cross-correlation function (the time offset corresponding to the peak of the cross-correlation function is the time difference between the wind noise signal reaching the dual microphones, that is, the signal delay); the propagation distance can be obtained based on the spatial position coordinates of the dual microphones using a spatial distance formula; the propagation speed can be obtained by dividing the propagation distance by the signal delay, and the propagation direction can be determined based on the order of the dual microphone positions and the signal delay. It should be noted that the side of the microphone that receives the signal first is the incoming direction; the signal attenuation trend can be determined by obtaining the signal strength at the dual microphones, combining the propagation speed, and comparing the intensity changes at different distances using a fitting function (such as an inverse square or exponential function).

[0078] The embodiment of the present invention utilizes the signal strength, the frequency characteristics and the wind noise propagation law to construct the wind noise dynamic threshold of the dual microphones. The wind noise dynamic threshold can be used to more accurately reflect the actual level of wind noise, avoid misjudgment that may be caused by a fixed threshold, and provide a flexible standard for more accurate wind noise detection.

[0079] The wind noise dynamic threshold refers to a judgment standard that changes dynamically according to various characteristics of wind noise (such as signal strength, frequency characteristics, propagation law, etc.). It is not a fixed value, but will be adaptively adjusted as the wind noise changes under different time and space conditions.

[0080] Optionally, the process of constructing the dynamic threshold of wind noise of the dual microphones by using the signal strength, the frequency characteristics and the wind noise propagation law is as follows: first, according to the wind noise propagation law, if the wind noise propagates rapidly from a specific direction and the signal strength is large, it means that the wind noise has a serious impact, and the dynamic threshold can be appropriately increased; secondly, considering the frequency characteristics, if the wind noise is concentrated in the high frequency band, a lower threshold is set for the high frequency signal, and a relatively high threshold is set for the low frequency signal. For example, when the wind noise signal strength is significantly enhanced in a certain period of time, and the high frequency component increases, it is judged according to the propagation law that the wind noise is approaching. At this time, the dynamic threshold is greatly reduced in the high frequency band and slightly reduced in the low frequency band to accurately detect the wind noise.

[0081] Furthermore, the embodiment of the present invention can adjust the detection standard according to the real-time characteristics of the wind noise by performing real-time wind noise detection on the dual microphones based on the wind noise dynamic threshold, thereby improving the accuracy and real-time performance of wind noise detection, ensuring the stability and reliability of the system in different environments and at different times, improving system performance, and reducing the adverse effects of wind noise on subsequent signal processing and applications.

[0082] Embodiment 2: like Figure 2 , which is a schematic diagram of functional modules of a dual-microphone wind noise detection system of the present invention.

[0083] The dual-microphone wind noise detection system 200 of the present invention can be installed in an electronic device. According to the functions to be implemented, the dual-microphone wind noise detection system can include a signal processing module 201, a spectrum component identification module 202, a signal analysis module 203 and a wind noise detection module 204. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0084] In the embodiment of the present invention, the functions of each module / unit are as follows: The signal processing module 201 is used to obtain dual microphones to be detected for wind noise, perform acoustic spatial deployment on the dual microphones to obtain deployed dual microphones, collect time domain audio signals of the deployed dual microphones using a preconfigured phase-locked loop synchronization technology, analyze expected signals of the time domain audio signals, and perform signal iterative processing on the time domain audio signals based on the expected signals to obtain target signals; The spectral component identification module 202 is used to segment the target signal into short time periods to obtain segmented signals, perform short-time frequency domain transformation on the segmented signals to obtain a time-frequency spectrum of the segmented signals, identify the silent segment spectrum in the time-frequency spectrum, analyze the spectral characteristics of the silent segment spectrum, determine the background noise spectrum of the time-frequency spectrum based on the spectral characteristics, perform smoothing correction on the background noise spectrum to obtain a corrected spectrum, identify the spectral components in the time-frequency spectrum based on the corrected spectrum, and obtain the target spectrum; The signal analysis module 203 is used to calculate the linear transformation matrix of the target spectrum, perform signal independent component analysis on the target spectrum based on the linear transformation matrix to identify the wind noise signal in the target spectrum, analyze the wind noise signal characteristics corresponding to the wind noise signal, identify the full wind noise signal of the target spectrum based on the wind noise signal characteristics, calculate the signal strength of the full wind noise signal, and analyze the frequency characteristics of the full wind noise signal; The wind noise detection module 204 is used to query the spatial position and sampling time of the dual microphones, perform spatiotemporal correlation analysis on the dual microphones based on the spatial position and the sampling time, obtain the wind noise propagation law, use the signal strength, the frequency characteristics and the wind noise propagation law to construct the wind noise dynamic threshold of the dual microphones, and perform real-time wind noise detection on the dual microphones based on the wind noise dynamic threshold.

[0085] In detail, the modules in the dual-microphone wind noise detection system 200 described in the embodiment of the present invention are used in the same manner as described above. Figure 1 The same technical means are used as the wind noise detection method using dual microphones described in the specification and can produce the same technical effects, which will not be repeated here.

[0086] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0087] 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 spirit and scope of the technical solution of the present invention.

Claims

1. A dual-microphone wind noise detection method, characterized in that: The method comprises: Acquire dual microphones for wind noise detection, perform acoustic spatial deployment on the dual microphones to obtain deployed dual microphones, collect time domain audio signals of the deployed dual microphones using a preconfigured phase-locked loop synchronization technology, analyze expected signals of the time domain audio signals, and perform signal iterative processing on the time domain audio signals based on the expected signals to obtain target signals; The target signal is segmented into short time periods to obtain segmented signals, the segmented signals are transformed into short-time frequency domain to obtain a time-frequency spectrum of the segmented signal, a silent segment spectrum in the time-frequency spectrum is identified, a spectrum feature of the silent segment spectrum is analyzed, a background noise spectrum of the time-frequency spectrum is determined based on the spectrum feature, the background noise spectrum is smoothed and corrected to obtain a corrected spectrum, and a spectrum component in the time-frequency spectrum is identified based on the corrected spectrum to obtain a target spectrum; Calculating a linear transformation matrix of the target spectrum, performing signal independent component analysis on the target spectrum based on the linear transformation matrix to identify a wind noise signal in the target spectrum, analyzing a wind noise signal feature corresponding to the wind noise signal, identifying a full wind noise signal of the target spectrum based on the wind noise signal feature, calculating a signal strength of the full wind noise signal, and analyzing a frequency characteristic of the full wind noise signal; The spatial positions and sampling times of the dual microphones are queried, and based on the spatial positions and the sampling times, spatiotemporal correlation analysis is performed on the dual microphones to obtain a wind noise propagation law, and a wind noise dynamic threshold of the dual microphones is constructed by using the signal strength, the frequency characteristics, and the wind noise propagation law, and real-time wind noise detection is performed on the dual microphones based on the wind noise dynamic threshold.

2. A dual-microphone wind noise detection method as claimed in claim 1, characterized in that: The step of performing acoustic spatial deployment on the dual microphones to obtain deployed dual microphones includes: Querying the sound signal wavelength and expected spatial resolution of the dual microphones to determine the arrangement spacing of the dual microphones; Identifying the sound source direction and sound change range of the dual microphones to determine the arrangement angle of the dual microphones; Based on the arrangement spacing and the arrangement angle, the dual microphones are preliminarily spatially arranged to obtain initial arranged microphones; Signal synchronization verification is performed on the initially deployed microphones, and when a synchronization result of the signal synchronization verification satisfies a preset synchronization value, dual microphones are deployed.

3. The dual-microphone wind noise detection method according to claim 1, characterized in that: The analyzing the expected signal of the time-domain audio signal comprises: Removing a DC offset of the time-domain audio signal to obtain a preliminary processed signal; Performing frequency screening on the preliminary processed signal to obtain a screened signal; Performing noise estimation on the screening signal to obtain estimated noise; The estimated noise is removed by using an adaptive filtering method to obtain a clean signal; analyzing a clean signal feature of the clean signal; Based on the clean signal characteristics, a desired signal of the time-domain audio signal is determined.

4. The dual-microphone wind noise detection method according to claim 1, characterized in that: The performing short-time frequency domain transformation on the segmented signal to obtain a time-frequency spectrum diagram of the segmented signal includes: The segmented signal is transformed into a frequency domain in a short time period using the following formula to obtain a frequency domain transformed signal: in, represents the frequency domain transformed signal, represents the segmentation signal, w() represents the window function, N represents the length of the window in the window function, m represents the time index, R represents the sliding step of the window, Y represents the frequency index, j represents the imaginary unit, and n represents the time index of the segmentation signal in the window function; The frequency domain transformed signals are arranged in time sequence to obtain a time-frequency spectrum diagram of the segmented signals.

5. The dual-microphone wind noise detection method according to claim 1, characterized in that: The step of identifying the silent segment spectrum in the time-spectrogram comprises: The spectrum energy of the time window in the time-spectrogram is calculated using the following formula: in, represents the spectrum energy, represents the i-th window in the time window, represents the kth frequency in the time-spectrum diagram, The time spectrum diagram is and The spectrum amplitude at , K represents the number of frequencies in the time spectrum graph; constructing a silent energy threshold of the time-spectrogram; Based on the silence energy threshold and the spectrum energy, a silence segment spectrum in the time-spectrogram is determined.

6. The dual-microphone wind noise detection method according to claim 1, characterized in that: The analyzing the spectrum characteristics of the silent segment spectrum includes: Constructing a spectrum curve of the silent segment spectrum; Analyzing the spectrum shape characteristics of the silent segment spectrum using the spectrum curve; Identifying the energy distribution state of the silent segment spectrum on the frequency axis; Based on the energy distribution state, identifying the frequency distribution characteristics of the silent segment spectrum; Calculating the statistical value of the silent segment spectrum; Analyze the spectrum dispersion of the silent segment spectrum based on the statistical value; Identifying the spectrum statistical features of the silent segment spectrum based on the spectrum dispersion; Calculating the average value of the absolute value of the amplitude difference between adjacent frequency points in the silent segment spectrum to analyze the spectrum smoothness characteristics of the silent segment spectrum; The spectrum characteristics of the silent segment spectrum are determined based on the spectrum shape characteristics, the frequency distribution characteristics, the spectrum statistical characteristics and the spectrum smoothness characteristics.

7. The dual-microphone wind noise detection method according to claim 1, characterized in that: The identifying the full wind noise signal of the target spectrum based on the wind noise signal feature includes: Querying the time domain features, frequency domain features and high-order statistical features of the wind noise signal to obtain comprehensive features; Perform wind noise feature matching on the target spectrum based on the comprehensive feature to obtain a matching feature; Constructing a wind noise judgment rule of the matching features by using a weighting method; Based on the wind noise judgment rule, a preliminary judgment is made on the full wind noise signal of the target spectrum to obtain a preliminary full wind noise signal; The preliminary full wind noise signal is subjected to continuous regional integration analysis to determine the full wind noise signal of the target spectrum.

8. The dual-microphone wind noise detection method according to claim 1, characterized in that: The calculating the signal strength of the full wind noise signal includes: The mean square value of the total wind noise signal is calculated using the following formula: Among them, RMS represents the mean square value, M represents the sequence number of the full wind noise signal, represents the bth signal in the full wind noise signal, Analyzing the time domain signal strength of the full wind noise signal by using the mean square value; The sum of the squares of the spectrum amplitudes of the full wind noise signal is calculated using the following formula: Among them, V represents the sum of squares of spectrum amplitude, H represents the number of frequency points of the full wind noise signal, represents the spectrum amplitude at frequency point h, Analyzing the frequency domain signal strength of the full wind noise signal by using the sum of squares of the spectrum amplitude; The signal strength of the full wind noise signal is determined according to the time domain signal strength and the frequency domain signal strength.

9. The dual-microphone wind noise detection method according to claim 1, characterized in that: The performing of spatiotemporal correlation analysis on the dual microphones based on the spatial position and the sampling time to obtain the wind noise propagation law includes: Query the signal sequence of the dual microphones based on the sampling time; Aligning the signal sequences on the time axis to obtain an alignment signal; analyzing a signal delay of the alignment signal; Calculating the propagation distance of the dual microphones based on the spatial position; Analyzing the propagation speed and propagation direction of the dual microphones based on the signal delay and the propagation distance; Query the signal strength of the dual microphones; Analyzing the signal attenuation trend of the dual microphones based on the signal strength and the propagation speed; The wind noise propagation law of the dual microphones is determined based on the propagation speed, the propagation direction and the signal attenuation trend.

10. A dual-microphone wind noise detection system, characterized in that: The system comprises: A signal processing module is used to obtain dual microphones to be detected for wind noise, perform acoustic spatial deployment on the dual microphones to obtain deployed dual microphones, collect time domain audio signals of the deployed dual microphones using a preconfigured phase-locked loop synchronization technology, analyze an expected signal of the time domain audio signal, and perform signal iterative processing on the time domain audio signal based on the expected signal to obtain a target signal; a spectrum component identification module, for performing short-time segmentation on the target signal to obtain segmented signals, performing short-time frequency domain transformation on the segmented signals to obtain a time-frequency spectrum of the segmented signals, identifying a silent segment spectrum in the time-frequency spectrum, analyzing a spectrum feature of the silent segment spectrum, determining a background noise spectrum of the time-frequency spectrum based on the spectrum feature, performing smoothing correction on the background noise spectrum to obtain a corrected spectrum, identifying spectrum components in the time-frequency spectrum based on the corrected spectrum to obtain a target spectrum; a signal analysis module, configured to calculate a linear transformation matrix of the target spectrum, perform signal independent component analysis on the target spectrum based on the linear transformation matrix to identify a wind noise signal in the target spectrum, analyze a wind noise signal feature corresponding to the wind noise signal, identify a full wind noise signal of the target spectrum based on the wind noise signal feature, calculate a signal strength of the full wind noise signal, and analyze a frequency characteristic of the full wind noise signal; The wind noise detection module is used to query the spatial position and sampling time of the dual microphones, perform spatiotemporal correlation analysis on the dual microphones based on the spatial position and the sampling time, obtain the wind noise propagation law, use the signal strength, the frequency characteristics and the wind noise propagation law to construct the wind noise dynamic threshold of the dual microphones, and perform real-time wind noise detection on the dual microphones based on the wind noise dynamic threshold.

Images