Low-noise wireless microphone based on MEMS audio sensor and filtering method

Abstract

The invention relates to the technical field of voice signal processing, in particular to a low-noise wireless microphone based on an MEMS audio sensor and a filtering method, and the method comprises the steps: determining the waveform symmetry degree of each voice signal segment based on the correlation between all peak values and all valley values, and the change trend of all peak values in the time domain of each voice signal segment; obtaining the frequency spectrum of each voice signal segment, extracting all peak values in the frequency spectrum, and determining the peak saliency value of each voice signal segment based on the distribution degree of all the peak values and the difference between the maximum peak value and all the rest peak values; and based on the waveform symmetry degree and the peak saliency value, carrying out noise reduction on a voice signal in the microphone. The invention aims to efficiently identify and specifically eliminate the howling noise, and improve the quality of the audio output by the wireless microphone.

Description

Technical Field

[0001] The present application relates to the technical field of speech signal processing, and in particular to a low-noise wireless microphone and filtering method based on a MEMS audio sensor. Background Art

[0002] MEMS audio sensor is an audio sound-to-electric conversion element produced by MEMS (micro-electromechanical system) advanced semiconductor manufacturing process. It has the advantages of small size, low cost, high sensitivity and low power consumption, making it very suitable for portable devices such as wireless microphones. Wireless microphones are an audio device widely used in stage performances, conference speeches, public speeches and other occasions. It uses wireless technology (such as radio frequency, Bluetooth, etc.) to transmit audio signals, does not require physical connection, and provides greater flexibility and convenience.

[0003] Low-noise wireless microphones usually use filtering technology to reduce the noise of voice signals collected by MEMS audio sensors to reduce the interference of environmental noise and improve the clarity and quality of voice signals. In most cases, existing wireless microphones use a filtering method to reduce the noise of collected voice signals. This method is not targeted at the noise in the voice signal and can easily lead to certain types of noise in the voice signal not being effectively filtered out, such as the howling noise that often occurs in stage performances, conference speeches, public speeches, etc., which reduces the quality of the audio output by the wireless microphone. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of this application is to provide a low-noise wireless microphone and filtering method based on a MEMS audio sensor. The technical solutions adopted are as follows: In a first aspect, an embodiment of the present application provides a low-noise wireless microphone filtering method based on a MEMS audio sensor, the method comprising the following steps: Acquire a speech signal from a low-noise wireless microphone, and divide the speech signal into a plurality of speech signal segments; Extract all peaks and valleys in the time domain of each speech signal segment, and determine the fluctuation correlation of each speech signal segment based on the correlation between all peaks and all valleys; determine the trend fluctuation of each speech signal segment based on the change trend of all peaks in each speech signal segment, and determine the waveform symmetry of each speech signal segment in combination with the fluctuation correlation; Acquire the spectrum of each speech signal segment, extract all peaks in the spectrum, and determine the first sharpness of each speech signal segment based on the distribution degree of all peaks; determine the second sharpness of each speech signal segment based on the difference between the maximum peak and all other peaks in the spectrum of each speech signal segment, and determine the peak significance value of each speech signal segment in combination with the first sharpness; Based on the waveform symmetry and the peak significance value, the noise coefficient of each speech signal segment is determined, and the speech signal in the low-noise wireless microphone is denoised.

[0005] Preferably, the fluctuation correlation of each speech signal segment is the Pearson correlation coefficient between all peak values ​​and all valley values ​​in the time domain of each speech signal segment.

[0006] Preferably, the method for determining the trend fluctuation of each speech signal segment is: All peak values ​​in the time domain of each speech signal segment are used as input of the trend detection algorithm, and the trend statistics of each speech signal segment are output as the trend fluctuation degree of each speech signal segment.

[0007] Preferably, the expression of the waveform symmetry of each speech signal segment is: ; In the formula, Indicates the waveform symmetry of speech signal segment i; Indicates the trend fluctuation of speech signal segment i; represents the fluctuation correlation of speech signal segment i; exp( ) represents an exponential function with a natural constant as the base; norm[ ] represents a normalization function.

[0008] Preferably, the first sharpness of each speech signal segment is the kurtosis of all peaks in the frequency spectrum of each speech signal segment.

[0009] Preferably, the second sharpness of each speech signal segment is the average of the difference between the maximum peak value and all other peak values ​​in the frequency spectrum of each speech signal segment.

[0010] Preferably, the expression of the peak significance value of each speech signal segment is: ; In the formula, represents the peak significance value of speech signal segment i; represents the first sharpness of speech signal segment i; represents the second sharpness of speech signal segment i; norm() represents the normalization function.

[0011] Preferably, the expression of the noise coefficient of each speech signal segment is: ; represents the noise factor of speech signal segment i; Indicates the waveform symmetry of speech signal segment i; Represents the peak significance value of speech signal segment i.

[0012] Preferably, the noise reduction of the voice signal in the low-noise wireless microphone includes: The noise coefficients of all speech signal segments are used as the input of the threshold segmentation algorithm, and the segmentation threshold is output; If the noise coefficient of the speech signal segment is greater than the segmentation threshold, it is used as the input of the Kalman filtering algorithm and the denoised speech signal segment is output. Conversely, if the noise coefficient of the speech signal segment is less than or equal to the segmentation threshold, the speech signal segment is used as the input of the LMS adaptive filtering algorithm and the denoised speech signal segment is output.

[0013] In a second aspect, an embodiment of the present application also provides a low-noise wireless microphone based on a MEMS audio sensor, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, the steps of any one of the above-mentioned low-noise wireless microphone filtering methods based on a MEMS audio sensor are implemented.

[0014] This application has at least the following beneficial effects: The present application constructs waveform symmetry by dividing the voice signal in the microphone into multiple voice signal segments, and analyzing the correlation of the changing trends of the peaks and troughs in each voice signal segment, as well as the changing trends of the peak values. The beneficial effect of the present application is that the time domain waveform of the voice signal segment can be effectively evaluated to be a trumpet-shaped waveform opening to the right, which is a typical feature of howling noise. The distinction between the time domain waveform features of the human voice and the howling noise in the voice signal segment is improved, and possible howling segments are accurately identified, and noise is reduced in a targeted manner, thereby improving the quality of the audio output by the wireless microphone. The present application constructs a peak significance value by analyzing the changing trends and steepness of the peak values ​​of the voice signal segment in the frequency domain. The beneficial effect of the present application is that the peaks in the frequency domain can be effectively identified, which can improve the quality of the voice signal segment containing howling noise. Recognition accuracy, so as to more accurately locate the howling noise; the present application constructs a noise coefficient by integrating the analysis results of the speech signal segment in the time domain and the frequency domain, and its beneficial effect is that it can accurately identify the speech signal segment containing the howling noise, and adopt different filtering strategies for the speech signal segment containing the howling noise and the remaining speech signal segment, so as to optimize the filtering method and improve the quality of the audio output by the wireless microphone; the present application comprehensively analyzes the characteristics of the speech signal in the time domain and the frequency domain, efficiently identifies and specifically eliminates the howling noise, and performs Kalman filtering on the speech signal segment containing the howling noise, and performs LMS adaptive filtering on the speech signal segment without the howling noise, thereby reducing the possibility that the howling noise in the collected speech signal cannot be effectively filtered out, and improving the quality of the audio output by the wireless microphone. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 A flowchart of a low-noise wireless microphone filtering method based on a MEMS audio sensor provided in one embodiment of the present application; Figure 2 A time domain waveform diagram of a howling noise provided by one embodiment of the present application; Figure 3 A spectrum diagram of howling noise provided by one embodiment of the present application; Figure 4 A schematic diagram of a noise coefficient extraction process provided by an embodiment of the present application; Figure 5 A time domain waveform diagram after filtering out the howling noise provided by an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to further explain the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the specific implementation, structure, features and effects of the low-noise wireless microphone and filtering method based on the MEMS audio sensor proposed in the present application are described in detail below in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0018] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0019] The specific scheme of the low-noise wireless microphone and filtering method based on the MEMS audio sensor provided by the present application is described in detail below with reference to the accompanying drawings.

[0020] See also Figure 1 , which shows a flowchart of a low-noise wireless microphone filtering method based on a MEMS audio sensor provided by an embodiment of the present application, the method comprising the following steps: Step S1: Acquire a voice signal from a low-noise wireless microphone and divide the voice signal into a plurality of voice signal segments.

[0021] During large-scale events such as stage performances or public speeches, the natural voice of the performer may not be clearly heard by everyone, so the sound is amplified by using microphones and audio equipment to ensure that listeners in every corner can have a good listening experience. However, when the microphone and audio are used at the same time, the sound from the audio can be transmitted to the microphone through the space, which may cause acoustic feedback, that is, howling. Howling will affect the quality and stability of the sound, so in order to ensure the quality of the audio output by the wireless microphone, the howling noise needs to be filtered out.

[0022] The low-noise wireless microphone in this embodiment is composed of an input device, a transmitter, a receiver and a power supply, wherein the input device is composed of a MEMS audio sensor and a data processing unit. The MEMS audio sensor is used to collect the voice signal in the microphone and transmit it to the data processing unit, wherein the sampling frequency is set to f.

[0023] It should be noted that the value of the sampling frequency f is artificially set. In this embodiment, the value of the sampling frequency is 44.1 KHz. The implementer can also set it by himself according to the specific situation. This embodiment does not impose any special restrictions.

[0024] Furthermore, in order to ensure that the amplitudes in the speech signal are at the same level, the amplitudes of the speech signals are normalized. There are many commonly used normalization methods. In this embodiment, the maximum and minimum value normalization method is used to normalize the speech signal. The implementer may also combine other normalization methods such as the z-score normalization method. Regarding the selection of the normalization method, this embodiment does not impose any special restrictions.

[0025] Among them, the maximum and minimum value normalization method is a well-known technology, and the specific process of normalizing the data will not be described in detail.

[0026] Since a speech signal may contain howling, in order to suppress the howling in the speech signal in a targeted manner, the speech signal is divided into a plurality of speech signal segments.

[0027] It should be noted that there are many commonly used segmentation algorithms. In this embodiment, the voice signal is used as the input of the voice segmentation algorithm VAD, and multiple voice signal segments are output to achieve the segmentation of the voice signal. The implementer can also adopt other segmentation methods according to the specific situation. This embodiment does not impose any special restrictions on the selection of the segmentation algorithm.

[0028] Among them, the semantic segmentation algorithm VAD is a well-known technology, and its specific principle will not be repeated here.

[0029] Step S2: extract all peaks and valleys in the time domain of each speech signal segment, and determine the fluctuation correlation of each speech signal segment based on the correlation between all peaks and all valleys; determine the trend fluctuation of each speech signal segment based on the change trend of all peaks in each speech signal segment, and determine the waveform symmetry of each speech signal segment in combination with the fluctuation correlation.

[0030] In stage performances, conference speeches, public speeches and other occasions, environmental noise usually includes sounds from a variety of different sources, such as background conversations, air conditioning sounds, etc. The sound sources of these sounds are usually scattered and far away from the wireless microphone. When the user uses a wireless microphone to give a speech, the sound source of the human voice is relatively concentrated and close to the wireless microphone, and the howling noise generated in the wireless microphone is usually a high-volume and sharp chirping sound, which makes the human voice and howling noise in the collected voice signal have a higher volume than other types of environmental noise in the voice signal.

[0031] Generally speaking, the human voice signal does not have an obvious regular waveform in the time domain. This is because the human voice is a complex signal composed of multiple frequency components, including unvoiced and voiced sounds. The waveform of the human voice changes with the content and tone of the speech, and there is no fixed pattern. The howling noise in the wireless microphone usually presents a trumpet-shaped signal waveform opening to the right in the time domain. This is because howling is usually caused by acoustic feedback. When the sound is emitted from the speaker, it is picked up again by the wireless microphone, amplified by the amplifier, and then emitted from the speaker again. In this cycle, each cycle will increase the intensity of the sound of a specific frequency, forming positive feedback, resulting in an increasing volume, and finally forming a howling. This process is manifested in the time domain as the amplitude of the signal gradually increases in a symmetrical form.

[0032] Therefore, based on the difference in the waveform characteristics of human voice and howling signals in the time domain, the change regularity characteristics of each speech signal segment are analyzed to screen out the speech signal segments that may contain howling noise and filter out the howling noise in the speech signal. Specifically: (1) Extract all peaks and valleys in the time domain of each speech signal segment. There are many commonly used peak and valley extraction algorithms. In this embodiment, a peak-valley second-order difference recognition algorithm is used to obtain all peaks and valleys in the time domain of the speech signal segment. The implementer may also use other peak-valley extraction algorithms. This embodiment does not impose any special restrictions on the selection of the peak-valley extraction method.

[0033] Among them, the peak and trough second-order difference recognition algorithm is a well-known technology, and the specific process of extracting the peak and trough values ​​of the speech signal segment in the time domain will not be repeated here.

[0034] Preferably, the time domain waveform of the howling noise provided in this embodiment is as follows: Figure 2 shown.

[0035] (2) Further, the Pearson correlation coefficient between all peaks and all valleys in the time domain of each speech signal segment is used as the fluctuation correlation of each speech signal segment; The calculation steps of the Pearson correlation coefficient are well-known techniques, and the specific calculation process will not be described in detail.

[0036] (3) Further, all peak values ​​in the time domain of each speech signal segment are used as input of the trend detection algorithm, and the trend statistics of each speech signal segment are output as the trend fluctuation degree of each speech signal segment; It should be noted that there are many commonly used trend test algorithms. The Mann-Kendall trend test algorithm is used in this embodiment. The implementer can also use the Kendall trend test algorithm to test the change trend of the peak value. Regarding the selection of the trend test algorithm, this embodiment does not make any special restrictions.

[0037] Among them, the Mann-Kendall trend test algorithm is a well-known technology, and the specific process of testing the data change trend is not repeated here.

[0038] (4) Further, based on the trend fluctuation and fluctuation correlation of each speech signal segment, the waveform symmetry of each speech signal segment is determined, specifically: The waveform symmetry of speech signal segment i The expression is: ; In the formula, Indicates the trend fluctuation of speech signal segment i; represents the fluctuation correlation of i; exp( ) represents the exponential function with a natural constant as the base; norm[ ] represents the normalization function.

[0039] According to the waveform symmetry of each speech signal segment, it can be understood that the more the signal between the upper and lower parts of the time domain horizontal axis in the speech signal segment shows an opposite trend over time, the smaller the fluctuation correlation is, indicating that the more symmetrical the signal between the upper and lower parts of the time domain horizontal axis of the speech signal is, that is, it shows the characteristics of speaker symmetry, and the more likely it is that it is a speech signal segment containing howling noise; and the more the signal amplitude in the upper part of the time domain horizontal axis in the speech signal segment has an increasing trend over time, that is, the greater the trend fluctuation, the more the amplitude of the speech signal segment in the time domain has a symmetrical trend of gradually increasing, and the more likely it is that the time domain waveform of the speech signal segment is in the shape of a trumpet opening to the right, that is, the greater the waveform symmetry is, the more likely the corresponding signal segment is the speech signal segment where howling is located; conversely, the greater the fluctuation correlation of the speech signal segment and the smaller the trend fluctuation is, the smaller the waveform symmetry is, the smaller the possibility that the time domain waveform of the speech signal segment is in the shape of a trumpet opening to the right is, and the less likely the corresponding signal segment is to be a speech signal segment containing howling noise.

[0040] Step S3: Obtain the spectrum of each speech signal segment, and extract all peaks in the spectrum, and determine the first sharpness of each speech signal segment based on the distribution degree of all peaks; determine the second sharpness of each speech signal segment based on the difference between the maximum peak and all other peaks in the spectrum of each speech signal segment, and determine the peak significance value of each speech signal segment in combination with the first sharpness.

[0041] In the process of howling noise in a wireless microphone, the sound intensity of a certain frequency usually increases, forming a positive feedback, causing the volume to become louder and louder, which is perceived as a harsh chirping sound. The howling noise in the collected voice signal has a single frequency with a large frequency amplitude compared to the human voice in the frequency domain, thus presenting a frequency domain waveform with obvious peaks.

[0042] Therefore, by analyzing the distribution characteristics of the speech signal in the frequency domain, the speech signal segment where the howling may be located can be further determined, and noise reduction processing can be performed on it, thereby improving the quality of the audio output by the wireless microphone. Specifically: (1) Taking each speech signal segment as the input of the time-frequency conversion algorithm and outputting the spectrum of each speech signal segment; It should be understood that there are many commonly used time-frequency conversion algorithms. In this embodiment, fast Fourier transform is used to obtain the frequency spectrum of each speech signal segment. The implementer can also use other time-frequency conversion algorithms such as wavelet transform. This embodiment does not impose any special restrictions on the selection of time-frequency conversion algorithm.

[0043] Among them, fast Fourier transform is a well-known technology in the field of signal processing, and the specific process of converting time domain signals into frequency domain is not repeated here.

[0044] Preferably, the spectrum diagram of the howling noise provided in this embodiment is as follows: Figure 3 shown.

[0045] (2) further, extracting all peaks in the frequency spectrum of each speech signal segment, and taking the kurtosis of all peaks in the frequency spectrum of each speech signal segment as the first sharpness of each speech signal segment; The calculation steps of kurtosis are well-known techniques, and the specific calculation process will not be described in detail.

[0046] (3) Further, the difference between the maximum peak value and all other peak values ​​in the spectrum of each speech signal segment is averaged as the second sharpness of each speech signal segment; It should be noted that there are many methods for measuring the differences between data. In this embodiment, the absolute value of the difference between the maximum peak and the remaining peaks in the frequency spectrum of each speech signal segment is calculated as a method for measuring the difference between the maximum peak and the remaining peaks. The implementer may also use other methods for measuring the differences between data, such as ratios, based on specific circumstances. This embodiment does not impose any special restrictions on the selection of methods for measuring the differences between data.

[0047] (4) Further, based on the first sharpness and the second sharpness, a peak significance value of the speech signal segment is determined, thereby determining the speech signal segment where the howling is located, specifically: The peak significance value of speech signal segment i The expression is: ; In the formula, represents the first sharpness of speech signal segment i; represents the second sharpness of speech signal segment i; norm() represents the normalization function.

[0048] According to the peak significance value of each speech signal segment, it can be understood that if the steepness of all peaks on the spectrum is greater, that is, the first sharpness is greater, and the difference between the maximum peak of the speech signal segment in its frequency domain and the remaining peaks is greater, that is, the second sharpness is greater, then the frequency domain waveform with obvious peaks is more likely to appear in the spectrum of the speech signal segment, that is, the larger the peak significance value, the more likely the corresponding speech signal segment is to be a speech signal segment containing howling noise; conversely, if the steepness of all peaks on the spectrum is smaller, that is, the first sharpness is smaller, and the difference between the maximum peak of the speech signal segment in its frequency domain and the remaining peaks is smaller, that is, the second sharpness is smaller, then the frequency domain waveform with obvious peaks is more likely to appear in the spectrum of the speech signal segment, that is, the smaller the peak significance value, the less likely the corresponding speech signal segment is to be a speech signal segment containing howling noise.

[0049] Step S4: Based on the waveform symmetry and peak significance value of each speech signal segment, the noise coefficient of each speech signal segment is determined, and the speech signal in the low-noise wireless microphone is denoised.

[0050] The noise coefficient is determined by comprehensively considering the waveform symmetry and the peak significance value, so as to accurately determine the speech signal segment containing the howling noise, and perform noise reduction processing on the speech signal segment, specifically: Noise coefficient of speech signal segment i The expression is: ; Indicates the waveform symmetry of speech signal segment i; Represents the peak significance value of speech signal segment i.

[0051] Preferably, the noise coefficient extraction process schematic diagram provided in this embodiment is as follows Figure 4 shown.

[0052] Further, the noise coefficients of all speech signal segments are used as inputs of the threshold segmentation algorithm, and the segmentation threshold is output; If the noise coefficient of the speech signal segment is greater than the segmentation threshold, it is used as the input of the Kalman filtering algorithm and the denoised speech signal segment is output. Conversely, if the noise coefficient of the speech signal segment is less than or equal to the segmentation threshold, the speech signal segment is used as the input of the LMS adaptive filtering algorithm and the denoised speech signal segment is output.

[0053] It should be noted that there are many commonly used threshold segmentation algorithms. This embodiment adopts the maximum inter-class variance algorithm. The implementer may also adopt other threshold segmentation algorithms according to specific circumstances. This embodiment does not impose any special restrictions on the selection of the threshold segmentation algorithm.

[0054] Among them, the maximum inter-class variance algorithm is a well-known technology, and its specific principle will not be repeated here.

[0055] It should be noted that, since howling is a specific, periodic interference, and its changes have certain regularity, when processing howling with regular changes, the Kalman filter algorithm is usually used to filter and reduce noise, because the Kalman filter algorithm is mainly used for state estimation of linear Gaussian systems, it can provide the optimal estimate of the state, so that it can effectively identify and predict the frequency components of howling. Through accurate prediction, the Kalman filter algorithm can eliminate howling in a targeted manner without affecting other normal speech parts, thereby maintaining the clarity and audibility of the speech signal; For other parts of the speech signal, there may be other unexpected sounds such as environmental noise and equipment noise. They are usually time-varying and may have nonlinear characteristics. The LMS adaptive filtering algorithm is usually used for situations where the system model is unknown or changing. Therefore, the LMS adaptive filtering algorithm is used to adjust the filter parameters in real time to adapt to the constantly changing noise and interference, thereby effectively suppressing noise and improving the quality of the wireless microphone output audio.

[0056] Preferably, the time domain waveform after filtering out the howling noise provided in this embodiment is as follows: Figure 5 shown.

[0057] Based on the same inventive concept as the above method, an embodiment of the present application also provides a low-noise wireless microphone based on a MEMS audio sensor, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of any one of the above-mentioned low-noise wireless microphone filtering methods based on MEMS audio sensors are implemented.

[0058] It should be noted that the above sequence of the embodiments of the present application is for description only and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0059] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0060] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present application should be included in the protection scope of the present application.

Claims

1. A low-noise wireless microphone filtering method based on a MEMS audio sensor, characterized in that: The method comprises the following steps: Acquire a speech signal from a low-noise wireless microphone, and divide the speech signal into a plurality of speech signal segments; Extract all peaks and valleys in the time domain of each speech signal segment, and determine the fluctuation correlation of each speech signal segment based on the correlation between all peaks and all valleys; determine the trend fluctuation of each speech signal segment based on the change trend of all peaks in each speech signal segment, and determine the waveform symmetry of each speech signal segment in combination with the fluctuation correlation; Acquire the spectrum of each speech signal segment, extract all peaks in the spectrum, and determine the first sharpness of each speech signal segment based on the distribution degree of all peaks; determine the second sharpness of each speech signal segment based on the difference between the maximum peak and all other peaks in the spectrum of each speech signal segment, and determine the peak significance value of each speech signal segment in combination with the first sharpness; Based on the waveform symmetry and the peak significance value, the noise coefficient of each speech signal segment is determined, and the speech signal in the low-noise wireless microphone is denoised.

2. The low-noise wireless microphone filtering method based on MEMS audio sensor as claimed in claim 1, characterized in that: The fluctuation correlation of each speech signal segment is the Pearson correlation coefficient between all peak values ​​and all valley values ​​in the time domain of each speech signal segment.

3. The low-noise wireless microphone filtering method based on MEMS audio sensor as claimed in claim 1, characterized in that: The method for determining the trend fluctuation of each speech signal segment is as follows: All peak values ​​in the time domain of each speech signal segment are used as input of the trend detection algorithm, and the trend statistics of each speech signal segment are output as the trend fluctuation degree of each speech signal segment.

4. The low-noise wireless microphone filtering method based on MEMS audio sensor as claimed in claim 1, characterized in that: The expression of the waveform symmetry of each speech signal segment is: ; In the formula, Indicates the waveform symmetry of speech signal segment i; Indicates the trend fluctuation of speech signal segment i; represents the fluctuation correlation of speech signal segment i; exp( ) represents an exponential function with a natural constant as the base; norm[ ] represents a normalization function.

5. The low-noise wireless microphone filtering method based on MEMS audio sensor as claimed in claim 1, characterized in that: The first sharpness of each speech signal segment is the kurtosis of all peaks in the frequency spectrum of each speech signal segment.

6. The low-noise wireless microphone filtering method based on MEMS audio sensor as claimed in claim 1, characterized in that: The second sharpness of each speech signal segment is the average of the difference between the maximum peak value and all other peak values ​​in the frequency spectrum of each speech signal segment.

7. The low-noise wireless microphone filtering method based on MEMS audio sensor as claimed in claim 1, characterized in that: The expression of the peak significance value of each speech signal segment is: ; In the formula, represents the peak significance value of speech signal segment i; represents the first sharpness of speech signal segment i; represents the second sharpness of speech signal segment i; norm() represents the normalization function.

8. The low-noise wireless microphone filtering method based on MEMS audio sensor as claimed in claim 1, characterized in that: The expression of the noise coefficient of each speech signal segment is: ; represents the noise factor of speech signal segment i; Indicates the waveform symmetry of speech signal segment i; Represents the peak significance value of speech signal segment i.

9. The low-noise wireless microphone filtering method based on MEMS audio sensor as claimed in claim 1, characterized in that: The method of reducing the noise of a speech signal in a low-noise wireless microphone comprises: The noise coefficients of all speech signal segments are used as the input of the threshold segmentation algorithm, and the segmentation threshold is output; If the noise coefficient of the speech signal segment is greater than the segmentation threshold, it is used as the input of the Kalman filtering algorithm and the denoised speech signal segment is output. Conversely, if the noise coefficient of the speech signal segment is less than or equal to the segmentation threshold, the speech signal segment is used as the input of the LMS adaptive filtering algorithm and the denoised speech signal segment is output.

10. A low-noise wireless microphone based on a MEMS audio sensor, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the low-noise wireless microphone filtering method based on the MEMS audio sensor are implemented as described in any one of claims 1 to 9.

Images