Method for filtering breathing sound in mask using adaptive FIR with residual controller and breathing mask

Abstract

The present invention discloses a method for filtering out breathing sounds in a mask by adaptive FIR with a residual controller and a breathing mask. By processing each frame of audio signal, a decision mechanism and a residual control function are introduced on the basis of adaptive filtering, and the FIR filter coefficients are adaptively updated in the breathing sound signal, while the algorithm stops updating the coefficients for the voice signal. The breathing sounds and residual noises in the mask can be effectively filtered out, thereby greatly improving the communication quality without damaging the voice signal.

Description

Technical Field

[0001] The invention relates to the technical field of signal processing, and in particular to a method for filtering breathing sound in a mask by adaptive FIR with a residual controller and a breathing mask. Background Art

[0002] Closed mask respirators are widely used in various extreme and harsh environments such as firefighting and high-altitude flight. Usually, the inner wall of the mask is equipped with transmitters and receivers and communication equipment to communicate with the outside world or teammates. The inside of the mask is a closed environment. During communication, the breathing airflow will be focused around the transmitter, producing obvious breathing sounds, which seriously reduce the call quality. Breathing sounds have a certain periodicity and cannot be classified as noise from an acoustic point of view, but such sounds seriously reduce the call quality and should be filtered out.

[0003] At present, the methods of removing breathing sounds in masks are usually spectral subtraction, Wiener filtering, subspace decomposition, etc. Traditional speech enhancement algorithms treat them as background noise and performance processing. The methods of filtering out breathing sounds in traditional algorithms have the following two disadvantages:

[0004] (1) This type of algorithm can only filter out high signal-to-noise ratio noise, but the breathing sound in the respirator is relatively loud, and the voice signal-to-noise ratio is low, so it cannot be completely filtered out, and obvious residual music noise will appear. At this time, the algorithm fails and cannot achieve the effect of voice enhancement and sound quality improvement. Instead, the obvious music noise reduces the call quality and voice intelligibility.

[0005] (2) The algorithm assumes that the noise is steady-state noise and its energy does not change dramatically over time. However, the breathing sound inside the mask is not a background noise with constant power, but changes with the breathing sound. Changes in a person's movement state can also cause dramatic changes in the breathing sound.

[0006] Therefore, when traditional algorithms process this type of noise, they are unable to accurately update the breathing sound spectrum in a timely manner, and are likely to damage the voice during processing, reducing call quality and voice intelligibility. Summary of the invention

[0007] In view of this, it is necessary to provide a method and a breathing mask for filtering out breathing sounds in a mask using an adaptive FIR with a residual controller, which can filter out breathing sounds more effectively without damaging the voice.

[0008] A method for filtering breathing sounds in a mask using an adaptive FIR with a residual controller comprises the following specific steps:

[0009] Acquire an audio signal and perform frame processing on the audio signal;

[0010] The power coefficient and autocorrelation coefficient of each frame of noisy signal are obtained by fast Fourier transform;

[0011] The short-time frame energy is used as a judgment function to judge whether the signal contains a speech frame. If yes, the FIR filter coefficient is stopped from being updated. If no, the FIR filter coefficient is adaptively updated according to the error signal.

[0012] The breathing sound is filtered out using the FIR filter with updated coefficients, and a residual control function is introduced to eliminate the residual noise using the inter-frame randomness of the breathing sound to obtain the filtered signal;

[0013] The target output signal is obtained by performing inverse fast Fourier transform on the filtered signal.

[0014] Preferably, the calculation formula of the power coefficient E(X) of each frame of the noisy signal is as shown in formula (1):

[0015] (1);

[0016] Among them, |X(ω)| is the amplitude modulus of the signal frequency;

[0017] The autocorrelation coefficient of each frame of the noisy signal The calculation formula is shown in formula (2):

[0018] (2);

[0019] in, Indicates the time delay of the signal; Indicates the length of the signal; represents the signal value at time t; express The signal value at that moment.

[0020] Preferably, the step of using the short-time frame energy as a judgment function to judge whether the signal contains a speech frame specifically includes:

[0021] According to the power coefficient E(X) and autocorrelation coefficient of each frame of the noisy signal Calculate the judgment function , as shown in formula (3):

[0022] (3);

[0023] The average energy in the breathing sound samples is counted, and the threshold value TH is calculated according to the formula shown in formula (4):

[0024] (4);

[0025] in, is the adjustment factor, and its value range is 0.1~0.9; It is the average value of breathing sounds collected from multiple breathing sound samples, that is, the average energy of multiple breathing sounds;

[0026] When the judgment function is greater than the threshold, it is considered that the signal contains a speech frame, otherwise it does not contain a speech frame and the FIR filter coefficients need to be adaptively updated.

[0027] Preferably, the step of updating the FIR filter coefficients specifically includes:

[0028] Determine the input vector of the audio signal after frame processing , the initial weight vector of the FIR filter and controls the step size factor for updating the FIR filter coefficients ;

[0029] Calculate the actual output signal of the FIR filter , as shown in formula (5):

[0030] (5);

[0031] in, Represents the weight vector The transpose of

[0032] According to the expected output signal of the FIR filter and the actual output signal Calculate the error signal , as shown in formula (6):

[0033] (6);

[0034] According to the error signal and the input vector , calculate and update the weight vector , as shown in formula (7):

[0035] (7);

[0036] Repeat the above steps to adaptively update the FIR filter coefficients according to the error signal until the error signal approaches 0, and then stop updating the FIR filter coefficients.

[0037] Preferably, the step of introducing a residual control function and eliminating residual noise by utilizing the inter-frame randomness of the breathing sound specifically includes:

[0038] Convert the input signal into frequency domain ;

[0039] in, is the frequency domain signal of the speech signal, is the frequency domain signal of the breathing sound signal;

[0040] Calculate the residual control function and the posterior estimated probability , as shown in (8) and (9) respectively:

[0041] (8);

[0042] (9);

[0043] in, is the control function; is the a priori signal-to-noise ratio, is the actual signal-to-noise ratio of the kth spectral component; is the a posteriori signal-to-noise ratio, which is the signal-to-noise ratio measured for the kth spectral component after adding noise; represents the frequency domain component with breathing signal, l Represents the length of the signal;

[0044] The a priori signal-to-noise ratio is calculated by inputting a signal with residual noise and the a posteriori signal-to-noise ratio , as shown in (10) and (11) respectively:

[0045] (10);

[0046] (11);

[0047] in, Represents the variance of the kth spectral component of the speech signal spectrum; represents the variance of the kth spectral component of the breathing sound spectrum; represents the frequency domain component of the breathing signal;

[0048] Depend on Derived by decision-making , the calculation formula is shown in formula (12):

[0049] (12);

[0050] Where a is the smoothing coefficient;

[0051] The final output signal is obtained by filtering out the residual noise through the difference between the input signal and the residual control function. , as shown in formula (13):

[0052] (13).

[0053] Preferably, the specific step of performing frame processing on the audio signal includes:

[0054] Step 11, selecting a frame length to determine the number of samples contained in each frame;

[0055] Step 12, determining the frame shift, which is the time interval between two adjacent frames;

[0056] Step 13, loading a window function. In order to avoid discontinuity at the frame boundary, a window function is usually loaded on the samples of each frame;

[0057] Step 14, overlapping and adding, if the frame shift value is not an integer multiple of the frame length, overlapping and adding operations are required to ensure the continuity of the signal;

[0058] Step 15, repeat the process, repeating steps 11 to 14 for the entire audio signal until all samples are processed.

[0059] Preferably, the specific step of performing inverse fast Fourier transform on the filtered signal to obtain the target output signal comprises:

[0060] Reconstruct the signal spectrum and filter the signal Perform inverse fast Fourier transform to convert the frequency domain signal into time domain signal and output it.

[0061] And, a breathing mask, characterized in that it includes an FIR filter for filtering out the breathing sound in the mask using the method of adaptive FIR filtering out the breathing sound in the mask with a residual controller as described above.

[0062] In the above-mentioned adaptive FIR filtering method with residual controller for removing breathing sound in the mask and the breathing mask, each frame of audio signal is processed, and a decision mechanism and a residual control function are introduced on the basis of adaptive filtering, and the FIR filter coefficient is adaptively updated in the breathing sound signal, and the algorithm stops updating the coefficient in the voice signal segment, which can effectively filter out the breathing sound and residual noise in the mask, greatly improve the communication quality, and will not damage the voice signal. The invention has controllable cost and broad market prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flow chart of a method for filtering breathing sound in a mask using an adaptive FIR with a residual controller according to an embodiment of the present invention.

[0064] Figure 2 The present invention is a flowchart of performing frame processing on an audio signal in a method for filtering breathing sounds in a mask using an adaptive FIR with a residual controller according to an embodiment of the present invention.

[0065] Figure 3 It is a schematic diagram of a non-speech segment FIR estimator of a method for filtering breathing sounds in a mask using an adaptive FIR with a residual controller according to an embodiment of the present invention.

[0066] Figure 4 It is a schematic diagram of updating the speech segment stop coefficient of the method for filtering the breathing sound in the mask by using an adaptive FIR with a residual controller according to an embodiment of the present invention. DETAILED DESCRIPTION

[0067] This embodiment takes the method for filtering breathing sounds in a mask by adaptive FIR with a residual controller and a breathing mask as an example, and the present invention will be described in detail below in conjunction with specific embodiments and drawings.

[0068] See also Figure 1 , Figure 2 , Figure 3 and Figure 4 , showing a method for adaptive FIR filtering of breathing sounds in a mask with a residual controller provided by an embodiment of the present invention, comprising the following specific steps: acquiring an audio signal and performing frame processing on the audio signal; obtaining a power coefficient and an autocorrelation coefficient of each frame of the noisy signal by fast Fourier transform; using short-time frame energy as a judgment function to judge whether the signal contains a speech frame, if yes, stopping updating the FIR filter coefficients, if not, adaptively updating the FIR filter coefficients according to the error signal; using the FIR filter with updated coefficients to filter out the breathing sounds, and introducing a residual control function, using the inter-frame randomness of the breathing sounds to eliminate the residual noise, and obtaining a filtered signal; performing an inverse fast Fourier transform on the filtered signal to obtain a target output signal.

[0069] FIR (Finite Impulse Response) filter is a finite length unit impulse response filter, also known as a non-recursive filter. It is the most basic element in the digital signal processing system. It can have a strict linear phase-frequency characteristic while ensuring any amplitude-frequency characteristic. At the same time, its unit sampling response is finite length, so the filter is a stable system. Therefore, FIR filters are widely used in communications, image processing, pattern recognition and other fields.

[0070] The method for filtering breathing sounds in a mask using an adaptive FIR with a residual controller provided in an embodiment of the present invention comprises the following steps:

[0071] Step S10, acquiring an audio signal and performing frame processing on the audio signal.

[0072] The audio signal includes a breathing sound signal and a speech signal. The audio signal is divided into a plurality of continuous frames, and the frame length and the number of samples contained in each frame are determined according to signal characteristics and application requirements.

[0073] Specifically, the specific steps of performing frame processing on the audio signal include:

[0074] Step S11, selecting a frame length to determine the number of samples contained in each frame;

[0075] Step S12, determining the frame shift, which is the time interval between two adjacent frames;

[0076] Step S13, loading a window function. In order to avoid discontinuity at the frame boundary, a window function is usually loaded on the samples of each frame;

[0077] Step S14, overlapping and adding, if the frame shift value is not an integer multiple of the frame length, overlapping and adding operations are required to ensure the continuity of the signal;

[0078] Step S15, repeat processing: repeat steps S11 to S14 for the entire audio signal until all samples are processed.

[0079] By dividing the frames, the audio signals can be analyzed and processed more effectively to meet the needs of different application scenarios.

[0080] The audio signal is divided into multiple continuous frames, and the frame length and the number of samples contained in each frame are determined according to the signal characteristics and application requirements.

[0081] Specifically, the frame length is the duration of each frame signal, which is usually between 20 milliseconds and 100 milliseconds. After the frame length is determined, the number of samples contained in each frame is also related to the sampling rate, that is, the number of samples is obtained by multiplying the sampling rate by the frame length. For example, when the sampling rate is 8 kHz, a frame length of 22.5 milliseconds corresponds to 180 samples; when the sampling rate is 16 kHz, a frame length of 22.5 milliseconds corresponds to 360 samples. In this embodiment, the duration of each frame of the audio signal is 22.5 milliseconds, and each frame of the audio signal includes 180 samples.

[0082] The basis for segmenting audio signals includes multiple aspects: First, the characteristics of the signal. Audio signals may show different characteristics in different time periods, such as changes in phonemes in speech signals and changes in beats in music signals. Segmenting audio signals can help identify and analyze these changes. Second, the sampling rate. The sampling rate of the audio signal determines the maximum frame length. According to the Nyquist theorem, the frame length cannot exceed half of the sampling rate. Third, application requirements. Different audio processing applications may require different frame lengths and frame shift settings. Fourth, the non-stationarity of the signal. Audio signals are usually non-stationary. By segmenting, non-stationary signals can be converted into multiple short-term stationary signals for easy analysis.

[0083] Audio signal segmentation is a multi-factor decision-making process, and the most appropriate segmentation method needs to be determined based on specific application scenarios and requirements.

[0084] Specifically, framing the audio signal is beneficial for localized analysis of the signal. Audio signals are usually non-stationary, that is, their characteristics change over time. Framing can localize the non-stationary signal into a series of short-time stationary signals, which is convenient for spectral analysis. It can reduce boundary effects. The use of window functions can reduce discontinuities at frame boundaries and avoid signal distortion. Spectral analysis, feature extraction and audio encoding can also be performed. After framing, each frame can be subjected to a fast Fourier transform (FFT) to obtain the signal's spectral information, which is beneficial for extracting the features of the audio signal. In the audio encoding process, framing is used to convert continuous audio signals into discrete data suitable for encoding and transmission. Audio signals can also be processed in real time. Framing allows audio signals to be processed in real time in blocks, which is very important for real-time audio applications.

[0085] Step S20, obtaining the power coefficient and autocorrelation coefficient of each frame of the noisy signal through FFT transformation (Fast Fourier Transform, which can convert the time domain signal into the frequency domain signal for processing and analysis);

[0086] The power coefficient E(X) of a signal is the square of the signal frequency amplitude modulus:

[0087] (1);

[0088] Autocorrelation coefficient of the signal The expression formula is:

[0089] (2);

[0090] in, is the time delay, which indicates the time delay of the signal; Indicates the length of the signal, ensuring that it is The time value is 1; represents the signal value at time t; express The signal value at time t, that is, the signal shifts after time t The value after.

[0091] Step S30, using the short-time frame energy as a judgment function to judge whether the signal contains a speech frame, if it contains a speech frame, then stop updating the coefficients, otherwise, then adaptively update the FIR filter coefficients;

[0092] According to the results of (1) and (2), the judgment function can be obtained for:

[0093] (3);

[0094] in, is the autocorrelation coefficient of the signal, is the power coefficient of the signal.

[0095] Count the average energy of the breathing sound samples and set the following thresholds:

[0096] (4);

[0097] In the formula, TH represents the threshold value, To adjust the factor, adjust according to the specific scenario, Derived from the statistics of breathing sound sample set;

[0098] Specifically, The value depends on the size of the scene used for the algorithm. When used in small spaces, The value is 0.3. When used in a large confined space, The value is 0.7. In this embodiment, The preferred value is 0.3.

[0099] Specifically, the breathing sound sample set is a data set used to study and develop breathing sound analysis technology, and generally includes recordings of sounds produced by different individuals during breathing. In this embodiment, The average value of the breathing sounds in multiple breathing sound sample sets is used, that is, the average energy of multiple breathing sounds.

[0100] When the judgment function is greater than the threshold, it is considered that the signal contains a speech frame, otherwise it does not contain a speech frame and the FIR filter coefficients need to be adaptively updated.

[0101] The weight coefficients of the FIR filter are usually dynamically updated based on the predicted difference. The difference algorithm adopts the LMS (Least Mean Squares) algorithm, which minimizes the error between the actual output and the expected output by iteratively updating the filter weights.

[0102] The steps are:

[0103] (i) Initialize the FIR weight vector , usually set it to 0 or a small random number and choose an appropriate step factor ,The step size factor needs to be small enough to ensure the convergence of the algorithm, and can be selected as 0.06 in the mask environment;

[0104] (ii) For each input signal, form the input vector ;

[0105] (iii) Calculate the output of the filter : ;

[0106] (iv) Calculate the error signal , error signal It is an expected signal With filter Difference: ;

[0107] (v) Update the weight vector based on the error signal and the input vector : ;

[0108] (vi) Repeat (ii)-(v) and dynamically update the filter coefficients according to the error signal until the error signal When it approaches 0, stop updating the FIR filter coefficients.

[0109] Step S40, using the FIR filter with updated coefficients to filter out the breathing sound. At this time, residual noise still remains in the filtered sound. For this, a residual control function is introduced to eliminate the residual noise by using the inter-frame randomness of the breathing sound to obtain a filtered signal;

[0110] Since the statistical mean of the breathing sound is used to replace the breathing sound of the current frame when the FIR filter is used to filter out the breathing sound, residual spectral peaks remain after spectral subtraction at some random spectral peaks of the breathing sound of the current frame. At the valley of the breathing sound spectrum of the current frame, the residual breathing sound spectrum component is set to a value close to zero due to the use of half-wave rectification. In this way, the enhanced residual breathing sound becomes some discrete spectral peaks in the frequency domain. Generally, the short-time spectrum of breathing sound fluctuates randomly, which means that when the breathing sound of a certain frame is relatively large, the estimated breathing sound power spectrum is not enough to subtract the breathing sound of the current frame, resulting in residual breathing sound. After these residual noises are transformed into the time domain, they form rhythmic noise similar to music, that is, music noise, which presents a superposition of some sine-like signals in the time domain.

[0111] The reason for the generation of "music noise" is related to the characteristics of the noise signal, because it conforms to the Gaussian distribution and has a wide range of random amplitude changes. Therefore, when the statistical variance of the breathing sound during the speechless period is used to represent the breathing sound, a part of the larger breathing sound will remain due to the size of the breathing sound component of the signal frame, which appears as a random bulge on the spectrum. At this time, the residual music noise seriously affects the voice quality.

[0112] To solve this problem, the residual control function is introduced to eliminate the residual noise by using the randomness of breathing sound between frames. The input time domain signal is converted into a frequency domain signal, which is ,in, is the frequency domain signal of the speech signal, is the frequency domain signal of the breathing sound signal. The derived residual control function is formula (5), and the posterior estimated probability Formula (6):

[0113] (5);

[0114] (6);

[0115] in is the control function. There are two unknown quantities in this function, namely the prior signal-to-noise ratio (which can be regarded as the actual signal-to-noise ratio of the kth spectral component) and the a posteriori signal-to-noise ratio (It can be regarded as the signal-to-noise ratio measured for the kth spectral component after adding noise.) It can be obtained by inputting a signal with residual noise.

[0116] (7);

[0117] (8);

[0118] in represents the variance of the kth spectral component of the speech signal spectrum, where the speech signal is a speech signal without breathing sound but with residual noise. represents the variance of the kth spectral component of the breathing sound spectrum. From formula (7) and formula (8), only is known, that is, the frequency domain component with the breathing signal. The breathing sound is a stable signal, and the variance of the breathing sound can be continuously updated by detecting the non-speech segment using VAD (Voice Activation Detection). The variance of the pure signal needs to be derived through the decision-guided method, and the result is shown in formula (9):

[0119] (9);

[0120] Where a is the smoothing coefficient, which is generally taken as 0.98.

[0121] Finally, the residual noise is filtered out by the difference between the input signal and the residual control function to obtain the target output signal , that is, the signal after filtering out the residual noise, as shown in formula (10):

[0122] (10).

[0123] Step S50, reconstructing the signal spectrum: Perform IFFT transformation (Inverse Fast Fourier Transform, which can convert frequency domain signals into time domain signals) to obtain time domain signals and output them.

[0124] The method provided by the present invention is used in the form of a dynamic link library. The audio signal collected by the microphone in the mask is collected and processed in frames. After collecting one frame, the algorithm is called to filter out the breathing sound.

[0125] And, a breathing mask, comprising an FIR filter, for filtering out the breathing sound in the mask by using the method of adaptive FIR filtering out the breathing sound in the mask with a residual controller as described above.

[0126] In the above-mentioned adaptive FIR filtering method with residual controller for removing breathing sound in the mask and the breathing mask, each frame of audio signal is processed, and a decision mechanism and a residual control function are introduced on the basis of adaptive filtering, and the FIR filter coefficient is adaptively updated in the breathing sound signal, and the coefficient is stopped from being updated in the voice signal, which can effectively filter out the breathing sound and residual noise in the mask, greatly improve the communication quality, and will not damage the voice signal. The invention has controllable cost and broad market prospects.

[0127] It should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for filtering breathing sounds in a mask using an adaptive FIR filter with a residual controller, characterized in that: The specific steps include: Acquire an audio signal and perform frame processing on the audio signal; The power coefficient and autocorrelation coefficient of each frame of noisy signal are obtained by fast Fourier transform; The short-time frame energy is used as a judgment function to judge whether the signal contains a speech frame. If yes, the FIR filter coefficient is stopped from being updated. If no, the FIR filter coefficient is adaptively updated according to the error signal. The breathing sound is filtered out using the FIR filter with updated coefficients, and a residual control function is introduced to eliminate the residual noise using the inter-frame randomness of the breathing sound to obtain the filtered signal; Performing inverse fast Fourier transform on the filtered signal to obtain the target output signal; Wherein, the step of updating the FIR filter coefficients specifically includes: Determine the input vector of the audio signal after frame processing , the initial weight vector of the FIR filter and controls the step size factor for updating the FIR filter coefficients ; Calculate the actual output signal of the FIR filter , as shown in formula (5): （5）； in, Represents the weight vector The transpose of According to the expected output signal of the FIR filter and the actual output signal Calculate the error signal , as shown in formula (6): （6）； According to the error signal and the input vector , calculate and update the weight vector , as shown in formula (7): （7）； Repeat the above steps to adaptively update the FIR filter coefficients according to the error signal until the error signal approaches 0, and then stop updating the FIR filter coefficients.

2. The method for filtering breathing sounds in a mask using an adaptive FIR with a residual controller as claimed in claim 1, characterized in that: The calculation formula of the power coefficient E(X) of each frame of the noisy signal is shown in formula (1): （1）； Among them, |X(ω)| is the amplitude modulus of the signal frequency; The autocorrelation coefficient of each frame of the noisy signal The calculation formula is shown in formula (2): （2）； in, Indicates the time delay of the signal; Indicates the length of the signal; represents the signal value at time t; express The signal value at that moment.

3. The method for filtering breathing sound in a mask by adaptive FIR with a residual controller as claimed in claim 2, characterized in that: The step of using the short-time frame energy as a judgment function to judge whether the signal contains a speech frame specifically includes: According to the power coefficient E(X) and autocorrelation coefficient of each frame of the noisy signal Calculate the judgment function , as shown in formula (3): （3）； The average energy in the breathing sound samples is counted, and the threshold value TH is calculated according to the formula shown in formula (4): （4）； in, is the adjustment factor, and its value range is 0.1~0.9; It is the average value of breathing sounds collected from multiple breathing sound samples, that is, the average energy of multiple breathing sounds; When the judgment function is greater than the threshold, it is considered that the signal contains a speech frame, otherwise it does not contain a speech frame and the FIR filter coefficients need to be adaptively updated.

4. The method for filtering breathing sound in a mask by adaptive FIR with a residual controller as claimed in claim 3, characterized in that: The step of introducing a residual control function and eliminating residual noise by utilizing the inter-frame randomness of breathing sound specifically includes: Convert the input signal into frequency domain ; in, is the frequency domain signal of the speech signal, is the frequency domain signal of the breathing sound signal; Calculate the residual control function and the posterior estimated probability , as shown in (8) and (9) respectively: （8）； （9）； in, is the control function; is the a priori signal-to-noise ratio, is the actual signal-to-noise ratio of the kth spectral component; is the a posteriori signal-to-noise ratio, which is the signal-to-noise ratio measured for the kth spectral component after adding noise; represents the frequency domain component with breathing signal, l Represents the length of the signal; The priori signal-to-noise ratio is calculated by inputting a signal with residual noise and the a posteriori signal-to-noise ratio , as shown in (10) and (11) respectively: （10）； （11）； in, Represents the variance of the kth spectral component of the speech signal spectrum; represents the variance of the kth spectral component of the breathing sound spectrum; represents the frequency domain component of the breathing signal; Depend on Derived by decision-making , the calculation formula is shown in formula (12): （12）； Where a is the smoothing coefficient; The final output signal is obtained by filtering out the residual noise through the difference between the input signal and the residual control function. , as shown in formula (13): （13）。 5. The method for filtering breathing sound in a mask by adaptive FIR with residual controller as claimed in claim 1, characterized in that: The specific steps of performing frame processing on the audio signal include: Step 11, selecting a frame length to determine the number of samples contained in each frame; Step 12, determining the frame shift, which is the time interval between two adjacent frames; Step 13, loading a window function. In order to avoid discontinuity at the frame boundary, a window function is usually loaded on the samples of each frame; Step 14, overlapping and adding, if the frame shift value is not an integer multiple of the frame length, overlapping and adding operations are required to ensure the continuity of the signal; Step 15, repeat the process, repeating steps 11 to 14 for the entire audio signal until all samples are processed.

6. The method for filtering breathing sound in a mask by adaptive FIR with residual controller as claimed in claim 1, characterized in that: The specific steps of performing inverse fast Fourier transform on the filtered signal to obtain the target output signal include: Reconstruct the signal spectrum and filter the signal Perform inverse fast Fourier transform to convert the frequency domain signal into time domain signal and output it.

7. A breathing mask, characterized in that: It comprises an FIR filter, which is used to filter out the breathing sound in the mask by using the method of filtering out the breathing sound in the mask by using an adaptive FIR with a residual controller as described in any one of claims 1 to 6.

Images