Biological audio data processing method and device and medium
By constructing the RGB image of respiratory audio and using convolutional neural network to obtain the filter threshold combination matrix, and combining dynamic filters for noise reduction processing, the instability problem of the fixed threshold filtering method is solved, and a more efficient respiratory audio signal processing effect is achieved.
Patent Information
- Application Number
- CN202510627891.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-15
- Publication Date
- 2025-08-05
AI Technical Summary
The frequency domain filtering method based on fixed thresholds in the prior art cannot adapt to the dynamic changes of the respiratory audio signal, resulting in unstable noise reduction effect, which may lead to loss of effective signals or noise residues.
The dynamic filtering method based on the convolutional neural network is adopted to construct the target image by obtaining the RGB value of the respiratory audio signal, and the filter threshold combination matrix is obtained using the target convolutional neural network model, and noise reduction processing is performed in combination with the dynamic filter, and the filter threshold is dynamically adjusted to adapt to the local characteristics of the signal.
Improve the accuracy of breathing audio noise reduction, avoid the loss of effective signals and noise residues, and improve the clarity and analyticity of the signal.
Smart Images
Figure CN120431964A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to a method, device and medium for processing biological audio data. Background Art
[0002] Biological breathing audio often contains a variety of interference signals, such as vibration signals caused by body movement or device contact, coupled noise from electrocardiogram signals, etc. These interference signals affect the clarity and analyzability of breathing audio. Therefore, it is necessary to perform noise reduction processing on breathing audio to make it clearer, easier to analyze and identify. In the existing technology, noise reduction is usually performed based on frequency domain filtering with a manually set fixed threshold.
[0003] However, the above method also has the following technical problems:
[0004] The characteristics of the signal in breathing audio (such as frequency and amplitude) change dynamically over time and do not remain constant. Fixed-threshold frequency-domain filtering is usually based on global spectrum statistics and cannot adjust to the local characteristics of each frame of the signal, thus failing to adapt to such dynamic changes. Therefore, noise reduction based on the above method is unstable and may result in loss of effective signal or residual noise. Summary of the Invention
[0005] In view of the above technical problems, the technical solution adopted by the present invention is:
[0006] According to a first aspect of the present invention, a method for processing biological audio data is provided, the method comprising the following steps:
[0007] S1. Obtain the biological audio data set S = {B, C, D} corresponding to the breathing audio A of the target organism, where B is a list of breathing audio signal values, C is a list of vibration signal values, and D is a list of electrocardiogram signal values. B includes the signal value B of the breathing audio signal of the i-th frame in A. i , i ranges from 1 to m, m is the total number of frames of the breathing audio signal in A, C includes i The signal value C of the vibration signal collected synchronously i , D includes B i The signal value D of the synchronously collected ECG signal i .
[0008] S2, respectively map all signal values in B, all signal values in C, and all signal values in D to the numerical interval [0, 255] to obtain B i The corresponding quantized value B 1 i 、C i The corresponding quantized value C 1 i and Di The corresponding quantization value D 1 i .
[0009] S3. Get B i Corresponding RGB value F i , where B 1 i As F i The intensity value of the red channel, C 1 i As F i The intensity value of the green channel, D 1 i As F i The intensity value of the mid-blue channel.
[0010] S4, based on F i Construct the target image T corresponding to A and input T into the target convolutional neural network model to obtain the filter threshold combination matrix J. The filter threshold combination in J corresponds one-to-one to the pixel points in T. The filter threshold combination includes a low-frequency cutoff frequency and a high-frequency cutoff frequency.
[0011] S5. Obtain B based on the filter threshold combination in J i The target filter threshold combination P corresponding to the corresponding breathing audio signal i .
[0012] S6, based on dynamic filter and P1, P2, ..., P i ,……,P m Perform noise reduction processing on A to obtain the target audio M corresponding to A.
[0013] According to a second aspect of the present invention, there is provided a device for processing biological audio data, the device comprising:
[0014] The biological audio data acquisition module is used to obtain the biological audio data set S = {B, C, D} corresponding to the breathing audio A of the target organism, where B is a list of breathing audio signal values, C is a list of vibration signal values, and D is a list of electrocardiogram signal values. B includes the signal value B of the breathing audio signal of the i-th frame in A. i , i ranges from 1 to m, m is the total number of frames of the breathing audio signal in A, C includes i The signal value C of the vibration signal collected synchronously i , D includes B i The signal value D of the synchronously collected ECG signal i .
[0015] The signal value quantization module is used to map all signal values in B, all signal values in C, and all signal values in D to the numerical interval [0, 255] to obtain Bi The corresponding quantized value B 1 i 、C i The corresponding quantized value C 1 i and D i The corresponding quantization value D 1 i .
[0016] RGB value acquisition module, used to obtain B i Corresponding RGB value F i , where B 1 i As F i The intensity value of the red channel, C 1 i As F i The intensity value of the green channel, D 1 i As F i The intensity value of the mid-blue channel.
[0017] Filter threshold combination matrix acquisition module, used for i Construct the target image T corresponding to A and input T into the target convolutional neural network model to obtain the filter threshold combination matrix J. The filter threshold combination in J corresponds one-to-one to the pixel points in T. The filter threshold combination includes a low-frequency cutoff frequency and a high-frequency cutoff frequency.
[0018] The target filter threshold combination acquisition module is used to obtain B based on the filter threshold combination in J i The target filter threshold combination P corresponding to the corresponding breathing audio signal i .
[0019] Noise reduction processing module for dynamic filter and P1, P2, ..., P i ,……,P m Perform noise reduction processing on A to obtain the target audio M corresponding to A.
[0020] According to a third aspect of the present invention, a non-transitory computer-readable storage medium is provided, in which a computer program is stored. The computer program is loaded and executed by a processor to implement the aforementioned method.
[0021] The present invention has at least the following beneficial effects:
[0022] The present invention provides a method, device and medium for processing biological audio data. The method can obtain a biological audio data set corresponding to the breathing audio of a target organism, process the data in the biological audio data set, obtain the RGB value corresponding to the signal value of each frame of the breathing audio signal in the breathing audio, construct a target image corresponding to the breathing audio based on the RGB value corresponding to the signal value of each frame of the breathing audio signal, and input the target image into a target convolutional neural network model to obtain a filter threshold combination matrix, obtain a target filter threshold combination corresponding to each frame of the breathing audio signal based on the filter threshold combination in the filter threshold combination matrix, wherein the filter threshold combination includes a low-frequency cutoff frequency and a high-frequency cutoff frequency, and perform noise reduction processing on the breathing audio based on a dynamic filter and the target filter threshold combination corresponding to each frame of the breathing audio signal to obtain the target audio corresponding to the breathing audio. It can be seen that the present invention can obtain the target image corresponding to the breathing audio based on the data in the biological audio data set corresponding to the breathing audio, obtain the filter threshold combination matrix based on the target image and the target convolutional neural network model, and further obtain the target filter threshold combination corresponding to each frame of the breathing audio signal, so that each frame of the breathing audio signal corresponds to a unique target filter threshold combination, and perform noise reduction processing on the breathing audio based on the dynamic filter and the unique target filter threshold combination corresponding to each frame of the breathing audio signal to obtain the target audio corresponding to the breathing audio, which can improve the accuracy of noise reduction, thereby avoiding the loss of effective signals or the occurrence of noise residues. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0024] Figure 1 A flowchart of a method for processing biological audio data provided by an embodiment of the present invention;
[0025] Figure 2 A schematic diagram of the structure of a biological audio data processing device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.
[0027] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar tasks and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0028] An embodiment of the present invention provides a method for processing biological audio data, the method comprising the following steps: Figure 1 As shown:
[0029] S1. Obtain the biological audio data set S = {B, C, D} corresponding to the breathing audio A of the target organism, where B is a list of breathing audio signal values, C is a list of vibration signal values, and D is a list of electrocardiogram signal values, where B includes the signal value B of the breathing audio signal of the i-th frame in A. i , i ranges from 1 to m, m is the total number of frames of the breathing audio signal in A, that is, B={B1,B2,……,B i ,……,B m}, C includes B i The signal value C of the vibration signal collected synchronously i , that is, C={C1,C2,……,C i ,……,C m}, D includes B i The signal value D of the synchronously collected ECG signal i , that is, D={D1,D2,……,D i ,……,D m}.
[0030] Specifically, B1, B2, ..., B i ,……,B m The corresponding breathing audio signal is a group of time-continuous breathing audio signals, which together constitute the complete breathing audio A.
[0031] Specifically, collect B i 、C i and D i The time point is the same.
[0032] Specifically, the target organism is an organism predetermined by those skilled in the art based on actual needs.
[0033] In a specific embodiment, an electronic stethoscope is used to collect B i and C i , using an electrocardiograph to collect D i , among which, electronic stethoscope can be understood as digital stethoscope.
[0034] Specifically, the measurement unit of the signal value of the breathing audio signal is decibel, the measurement unit of the signal value of the vibration signal is acceleration, and the measurement unit of the signal value of the electrocardiogram signal is millivolt.
[0035] Specifically, if the sampling frequency of the vibration signal is inconsistent with the sampling frequency of the breathing audio signal, the missing vibration signal value is generated by interpolation to complete C; if the sampling frequency of the ECG signal is inconsistent with the sampling frequency of the breathing audio signal, the missing ECG signal value is generated by interpolation to complete D. This can eliminate the time domain dislocation caused by different sampling frequencies, and lay the foundation for the subsequent acquisition of the RGB value corresponding to the signal value corresponding to each frame of the breathing audio signal.
[0036] S2, respectively map all signal values in B, all signal values in C, and all signal values in D to the numerical interval [0, 255] to obtain B i The corresponding quantized value B 1 i 、C i The corresponding quantized value C 1 i and D i The corresponding quantization value D 1 i Among them, those skilled in the art can adopt different mapping methods to obtain quantization values according to actual conditions, for example: using the logarithmic compression method to map the signal value of the breathing audio signal and the signal value of the vibration signal, and using the linear mapping method to map the signal value of the electrocardiogram signal, which will not be repeated here.
[0037] Specifically, max(B1, B2, ..., B i ,……,B m ) corresponds to a quantization value of 255, min(B1, B2, ..., B i ,……,B m ) corresponds to a quantized value of 0, max(C1, C2, ..., C i ,……,C m ) corresponds to a quantization value of 255, min(C1, C2, ..., C i ,……,C m ) corresponds to a quantized value of 0, max(D1, D2, ..., D i ,……,D m) corresponds to a quantization value of 255, min(D1, D2, ..., D i ,……,D m ) corresponds to a quantization value of 0, max() is a maximum value acquisition function, and min() is a minimum value acquisition function.
[0038] S3. Get B i Corresponding RGB value F i , where B 1 i As F i The intensity value of the red channel, C 1 i As F i The intensity value of the green channel, D 1 i As F i The intensity value of the mid-blue channel.
[0039] S4, based on F i Construct the target image T corresponding to A and input T into the target convolutional neural network model to obtain the filter threshold combination matrix J, where the filter threshold combination in J corresponds one-to-one to the pixel points in T; it can be understood that J includes the filter threshold combination corresponding to each pixel point in T.
[0040] Specifically, J is a two-dimensional matrix, and the position of the filter threshold combination in J is the same as the position of its corresponding pixel point in T.
[0041] Specifically, the filter threshold combination includes a low-frequency cutoff frequency and a high-frequency cutoff frequency.
[0042] Specifically, the target convolutional neural network model is a model obtained by training a preset convolutional neural network for a filter threshold combination prediction task, and the model architecture of the preset convolutional neural network is a U-Net architecture.
[0043] Through the above steps, the signal values corresponding to the respiratory signal, the vibration signal, and the electrocardiogram signal are quantized into interval values within 0-255, and encoded into RGB values, thereby realizing the spatiotemporal alignment and feature fusion of multimodal signals. A target image is constructed based on the RGB values corresponding to the signal values corresponding to each frame of the respiratory audio signal, and the target image is input into the target convolutional neural network model to obtain a filter threshold combination matrix. Compared with manually setting the filter threshold combination, the accuracy of the obtained filter threshold combination can be improved, and labor costs can be saved, which is conducive to improving efficiency.
[0044] S5. Obtain B based on the filter threshold combination in J i The target filter threshold combination P corresponding to the corresponding breathing audio signal i.
[0045] Specifically, P i The frequency range includes a low frequency cutoff and a high frequency cutoff.
[0046] S6, based on dynamic filter and P1, P2, ..., P i ,……,P m Perform noise reduction processing on A to obtain the target audio M corresponding to A.
[0047] In some embodiments, the above steps S1-S6 can be implemented by the following code:
[0048]
[0049]
[0050]
[0051]
[0052] Through the above steps, a biological audio dataset corresponding to the breathing audio of the target organism is obtained, the data in the biological audio dataset is processed, the RGB value corresponding to the signal value of each frame of the breathing audio signal in the breathing audio is obtained, the target image corresponding to the breathing audio is constructed based on the RGB value corresponding to the signal value of each frame of the breathing audio signal, and the target image is input into the target convolutional neural network model to obtain a filter threshold combination matrix, and based on the filter threshold combination in the filter threshold combination matrix, a target filter threshold combination corresponding to each frame of the breathing audio signal is obtained, wherein the filter threshold combination includes a low-frequency cutoff frequency and a high-frequency cutoff frequency, and the breathing audio is subjected to noise reduction processing based on the dynamic filter and the target filter threshold combination corresponding to each frame of the breathing audio signal to obtain the target audio corresponding to the breathing audio; the target image corresponding to the breathing audio can be obtained based on the data in the biological audio dataset corresponding to the breathing audio, the filter threshold combination matrix is obtained based on the target image and the target convolutional neural network model, and the target filter threshold combination corresponding to each frame of the breathing audio signal is further obtained, so that each frame of the breathing audio signal corresponds to a unique target filter threshold combination, and the breathing audio is subjected to noise reduction processing based on the dynamic filter and the unique target filter threshold combination corresponding to each frame of the breathing audio signal to obtain the target audio corresponding to the breathing audio, which can improve the accuracy of noise reduction, thereby avoiding the loss of effective signals or the occurrence of noise residuals.
[0053] Specifically, in step S4, the following steps are also included to obtain T:
[0054] S41, evenly divide A into n sub-audio numbers to obtain a sub-audio list E corresponding to A = {E1, E2, ..., E j,……,E n}, E j is the j-th sub-audio corresponding to A, where j ranges from 1 to n, and n is the preset number of sub-audios.
[0055] Specifically, the duration of each sub-audio is the same and there is no overlap between any two sub-audios.
[0056] Specifically, the duration of each sub-audio is t=T / n, where T is the duration of A.
[0057] Specifically, when j=1, E j The starting time point of A is the starting time point of E. When j≠1, j The starting time point and E j-1 The time interval between the starting time points is t, E j-1 is the j-1th sub-audio corresponding to A.
[0058] S42, if B i The corresponding breathing audio signal and E j The e-th frame breathing audio signal in the same frame breathing audio signal, then F i As E j The RGB value G corresponding to the e-th frame breathing audio signal je , to obtain E j Corresponding RGB value list G j ={G j1 , G j2 ,……,G je ,……,G jf}, where the value of e ranges from 1 to f, and f is the total number of frames of the breathing audio signal in the sub-audio.
[0059] Specifically, n×f=m.
[0060] S43, G je The RGB value of the pixel at the jth row and the eth column in T is used to obtain T, where the image resolution of T is f×n; that is, the number of horizontal pixels of T is f and the number of vertical pixels is n.
[0061] Concrete, E j The e-th frame of breathing audio signal in corresponds to the pixel point on the j-th row and e-th column in T.
[0062] Specifically, in step S5, if the breathing audio signal corresponding to a certain filter threshold combination in J is i The corresponding breathing audio signal is the same frame breathing audio signal, then the filtering threshold combination is used as P i , the breathing audio signal corresponding to the filter threshold combination is the breathing audio signal corresponding to the pixel point corresponding to the filter threshold combination.
[0063] Through the above steps, the breathing audio is evenly divided into n sub-audios to avoid repeated calculations. Based on the correspondence between each frame of breathing audio signals in each sub-audio and each frame of breathing audio signals in the breathing audio, and the RGB values corresponding to the signal values of each frame of breathing audio signals in the breathing audio, a target image is constructed, so that the pixels in the target image correspond one-to-one with each frame of breathing audio signals in each sub-audio. It can also be understood that the pixels in the target image correspond one-to-one with each frame of breathing audio signals in the breathing audio, so that the filter threshold combination matrix obtained by inputting the target image into the target convolutional neural network model includes the filter threshold combination corresponding to each pixel in the target image. Furthermore, the target filter threshold combination corresponding to each frame of breathing audio signals in the breathing audio can be obtained. The breathing audio is denoised based on the dynamic filter and the unique target filter threshold combination corresponding to each frame of breathing audio signals to obtain the target audio corresponding to the breathing audio, which can improve the accuracy of noise reduction, thereby avoiding the loss of effective signals or the occurrence of noise residue.
[0064] The present invention also provides a specific embodiment, which differs from the above embodiment in that step S4 includes the following steps to obtain T:
[0065] S401, divide A into h intermediate audios to obtain an intermediate audio list K corresponding to A = {K1, K2, ..., K g ,……,K h}, K g is the gth intermediate audio corresponding to A, where g ranges from 1 to h, and h is the number of intermediate audios.
[0066] Specifically, the duration of each intermediate audio is t1, where t1 is a preset intermediate audio length.
[0067] Specifically, when g=1, K g The starting time point is the starting time point of A. When g≠1, K g The starting time point and K g-1 The time interval between the starting time points is t2, K g-1 is the g-1th intermediate audio corresponding to A, and t2 is the preset non-overlapping duration.
[0068] Specifically, t2<t1.
[0069] S402, if B i The corresponding breathing audio signal and K g The y-th frame breathing audio signal in the same frame breathing audio signal, then F i As K gThe RGB value L corresponding to the breathing audio signal of the yth frame gy , to obtain K g Corresponding RGB value list L g ={L g1 , L g2 ,……,L gy ,……,L gq}, where y ranges from 1 to q, and q is the total number of frames of the breathing audio signal in the intermediate audio.
[0070] S403, L gy The RGB value of the pixel at the g-th row and y-th column in T is used to obtain T, where the image resolution of T is q×h; that is, the number of horizontal pixels of T is q and the number of vertical pixels is h.
[0071] Specifically, K g The y-th frame breathing audio signal in corresponds to the pixel point on the g-th row and y-th column in T.
[0072] Through the above steps, the breathing audio is divided into h intermediate audios, each of which has the same duration, and there is an overlapping part between two adjacent intermediate audios. Based on the correspondence between each frame of breathing audio signal in each intermediate audio and each frame of breathing audio signal in the breathing audio, as well as the RGB value corresponding to the signal value of each frame of breathing audio signal in the breathing audio, a target image is constructed, which can avoid destroying the continuity between the breathing audio signals. Further, it is beneficial to improve the accuracy of the filter threshold combination matrix output by the target convolutional neural network model.
[0073] Specifically, before step S401, the following steps are further included to obtain t2:
[0074] S10. Obtain the average respiratory frequency HX corresponding to A. Those skilled in the art know that any method for obtaining the average respiratory frequency corresponding to the respiratory audio in the prior art falls within the scope of protection of the present invention and will not be described in detail here.
[0075] S20, obtain the preset overlap mapping list YS={YS1, YS2, ..., YS x ,……,YS p}, YS x ={YS x1 , YS x2}, where YS x is the xth preset overlap mapping combination, x ranges from 1 to p, p is the number of preset overlap mapping combinations, YS x1 For YS x Preset respiratory rate in YS x2 For YS x1The corresponding preset overlap.
[0076] Specifically, the greater the preset respiratory frequency, the smaller the corresponding preset overlap.
[0077] S30, when HX=YS x1 When t2 is obtained, t2 meets the following conditions:
[0078] t2=t1×(1-YS x2 ).
[0079] Through the above steps, compared with setting a fixed preset non-overlapping time length, the preset non-overlapping time length is obtained based on the average respiratory frequency in the respiratory audio, the preset respiratory frequency and its corresponding preset overlap degree, wherein the larger the preset respiratory frequency, the smaller the corresponding preset overlap degree, the more adaptable it is, and the preset non-overlapping time length can be flexibly determined according to actual conditions. Further, the impact caused by the unreasonable setting of the fixed preset non-overlapping time length can be reduced.
[0080] Specifically, step S5 includes the following steps:
[0081] S51. Flatten J into a one-dimensional matrix in a row-first order to obtain the intermediate filter threshold combination matrix Q corresponding to J = [Q1, Q2, ..., Q r ,……,Q s ], Q r =(Q r1 , Q r2 ), Q r is the rth intermediate filter threshold combination in Q, r ranges from 1 to s, s is the number of intermediate filter threshold combinations in Q, Q r1 Q r The low-frequency cutoff frequency, Q r2 Q r The high-frequency cutoff frequency in ; it can be understood as connecting each row in J in sequence to form Q.
[0082] S52, traverse Q, if Q r The corresponding breathing audio signal and Q a The corresponding breathing audio signal is the same frame breathing audio signal and a>r, then let Q r1 =(Q r1 +Q a1 ) / 2, Q r2 =(Q r2 +Q a2 ) / 2, Q a Delete from Q so that Q is updated and Q after all updates is used as the key filter threshold combination matrix U, where Q ais the ath intermediate filter threshold combination in Q, where a ranges from 1 to s and a≠r, Q a1 Q a The low-frequency cutoff frequency, Q a2 Q a The high frequency cutoff frequency in, U=[U1,U2,……,U b ,……,U d ],U b =(U b1 , U b2 ), U b is the bth key filter threshold combination, b ranges from 1 to d, d is the number of key filter threshold combinations, U b1 For U b The low-frequency cutoff frequency, U b2 For U b The high frequency cutoff frequency.
[0083] Specifically, the breathing audio signal corresponding to the intermediate filtering threshold combination is the breathing audio signal corresponding to the pixel point corresponding to the intermediate filtering threshold combination.
[0084] Specifically, d<s.
[0085] S53, if U b The corresponding breathing audio signal and B i The corresponding breathing audio signal is the same frame breathing audio signal, then U b As P i .
[0086] Through the above steps, since there is overlap between adjacent intermediate audios, there are two intermediate filter threshold combinations in the intermediate filter threshold combination matrix corresponding to the same frame of respiratory audio signal. The data in the two intermediate filter threshold combinations are averaged, and redundant data is deleted so that the intermediate filter threshold combination matrix is updated and the intermediate filter threshold combination matrix after all updates is used as the key filter threshold combination matrix, which is beneficial to improving the accuracy of the key filter threshold combination, and further, is beneficial to improving the accuracy of obtaining the target filter threshold combination corresponding to each frame of respiratory audio signal in the respiratory audio.
[0087] In a specific embodiment, the following steps are included after step S52:
[0088] S521. Obtain a count value z and a preset window length v, wherein z is initially 0. The preset window length is set by those skilled in the art according to actual needs, for example, 5 or 10, and will not be repeated here.
[0089] S522. When z+v<d, obtain Y 1 (z，v) , Y2 (z，v) , W 1 (z，v) and W 2 (z，v) Then go to step S523, when z+v≥d, get Y 1 (z，d) , Y 2 (z，d) , W 1 (z，d) and W 2 (z，d) And go to step S524; Y 1 (z，v) For U z+1 , U z+2 ,……,U z+v The average value of all low-frequency cutoff frequencies in Y 2 (z，v) For U z+1 , U z+2 ,……,U z+v The variance of all low-frequency cutoff frequencies in , W 1 (z，v) For U z+1 , U z+2 ,……,U z+v The average value of all high frequency cutoff frequencies in W 2 (z，v) For U z+1 , U z+2 ,……,U z+v The variance of all high frequency cutoff frequencies in , Y 1 (z，d) For U z+1 , U z+2 ,……,U d The average value of all low-frequency cutoff frequencies in Y 2 (z，d) For U z+1 , U z+2 ,……,U d The variance of all low-frequency cutoff frequencies in , W 1 (z，d) For U z+1 , U z+2 ,……,U d The average value of all high frequency cutoff frequencies in W 2 (z，d) For U z+1 , U z+2 ,……,U d The variance of all high frequency cutoff frequencies in , U z+1 The z+1th key filter threshold combination, U z+2 The z+2th key filter threshold combination, U z+vThe z+vth key filter threshold combination.
[0090] S523, if W 1 (z，v) ≤W 0 And W 2 (z，v) ≤W 0 , then let z=z+1 and go to step S522; if W 1 (z，v) >W 0 And W 2 (z，v) ≤
[0091] W 0 , then Y 1 (z，v) As U z+1 , U z+2 ,……,U z+v The low frequency cutoff frequency in the quartz crystal is used to update U, and z=z+v is set and the process goes to step S522; if W 1 (z，v) ≤W 0 And W 2 (z，v) >W 0 , then Y 2 (z，v) As U z+1 , U z+2 ,……,U z+v The high frequency cutoff frequency in the quartz crystal is used to update U, and z=z+v is set and the process goes to step S522; if W 1 (z，v) >W 0 And W 2 (z，v) >W 0 , then Y 1 (z，v) As U z+1 , U z+2 ,……,U z+v The low frequency cutoff frequency in Y 2 (z，v) As U z+1 , U z+2 ,……,U z+v The high frequency cutoff frequency in the quartz crystal is used to update U, and z=z+v is set to proceed to step S522, W 0 is the preset variance.
[0092] S524, if W 1 (z，d) ≤W 0 And W 2 (z，d) ≤W 0, then go to step S525; if W 1 (z，d) >W 0 And W 2 (z，d) ≤W 0 , then Y 1 (z，d) As U z+1 , U z+2 ,……,U d The low frequency cutoff frequency in the U is updated and the process goes to step S525; if W 1 (z，d) ≤W 0 And W 2 (z，d) >W 0 , then Y 2 (z，d) As U z+1 , U z+2 ,……,U d The high frequency cutoff frequency in the CMOS is used to update U and proceed to step S525; if W 1 (z，d) >W 0 And W 2 (z，d) >W 0 , then Y 1 (z，d) As U z+1 , U z+2 ,……,U d The low frequency cutoff frequency in Y 2 (z，d) As U z+1 , U z+2 ,……,U d The high frequency cutoff frequency in is used to update U and enter step S525.
[0093] S525, if U b The corresponding breathing audio signal and B i The corresponding breathing audio signal is the same frame breathing audio signal, then U b As P i .
[0094] Through the above steps, if the variance of the low-frequency cutoff frequencies in all key filter threshold combinations in the window and the variance of the high-frequency cutoff frequencies in all key filter threshold combinations are not greater than the preset variance, it means that the data fluctuations of the low-frequency cutoff frequencies in all key filter threshold combinations and the low-frequency cutoff frequencies in all key filter threshold combinations are small. At this time, the window is slid by one step to continue processing. If the variance of the low-frequency cutoff frequencies in all key filter threshold combinations in the window is greater than the preset variance, it means that the data fluctuations of the low-frequency cutoff frequencies in all key filter threshold combinations are large. At this time, the average value of the low-frequency cutoff frequencies in all key filter threshold combinations is used as the low-frequency cutoff frequencies in all key filter threshold combinations. Similarly, if the variance of the high-frequency cutoff frequencies in all key filter threshold combinations in the window is greater than the preset variance, it means that the data fluctuations of the high-frequency cutoff frequencies in all key filter threshold combinations are large. At this time, all the relevant The average value of the high-frequency cutoff frequency in the key filter threshold combination is used as the high-frequency cutoff frequency in all the key filter threshold combinations, so that the key filter threshold combination is updated to reduce random errors. After the key filter threshold combination in the window is updated, the window is slid with a step size equal to the number of key filter threshold combinations in the window to continue processing, and the key filter threshold combination matrix is continued to be updated. Based on the key filter threshold matrix in the key filter threshold combination matrix after all updates are completed, the target filter threshold combination corresponding to each frame of the breathing audio signal in the breathing audio is obtained, which is conducive to improving the accuracy of obtaining the target filter threshold combination corresponding to each frame of the breathing audio signal in the breathing audio. Furthermore, based on the dynamic filter and the unique target filter threshold combination corresponding to each frame of the breathing audio signal, the breathing audio is denoised to obtain the target audio corresponding to the breathing audio, which can improve the accuracy of noise reduction, thereby avoiding the loss of effective signals or the occurrence of noise residues.
[0095] The embodiment of the present invention also provides a biological audio data processing device, such as Figure 2 As shown, the device includes:
[0096] The biological audio data acquisition module 1 is used to obtain the biological audio data set S = {B, C, D} corresponding to the breathing audio A of the target organism, where B is a list of breathing audio signal values, C is a list of vibration signal values, and D is a list of electrocardiogram signal values, wherein B includes the signal value B of the breathing audio signal of the i-th frame in A. i , i ranges from 1 to m, m is the total number of frames of the breathing audio signal in A, that is, B={B1,B2,……,B i ,……,B m}, C includes B i The signal value C of the vibration signal collected synchronously i , that is, C={C1,C2,……,Ci ,……,C m}, D includes B i The signal value D of the synchronously collected ECG signal i , that is, D={D1,D2,……,D i ,……,D m}.
[0097] Specifically, B1, B2, ..., B i ,……,B m The corresponding breathing audio signal is a group of time-continuous breathing audio signals, which together constitute the complete breathing audio A.
[0098] Specifically, collect B i 、C i and D i The time point is the same.
[0099] Specifically, the target organism is an organism predetermined by those skilled in the art based on actual needs.
[0100] In a specific embodiment, an electronic stethoscope is used to collect B i and C i , using an electrocardiograph to collect D i , among which, electronic stethoscope can be understood as digital stethoscope.
[0101] Specifically, the measurement unit of the signal value of the breathing audio signal is decibel, the measurement unit of the signal value of the vibration signal is acceleration, and the measurement unit of the signal value of the electrocardiogram signal is millivolt.
[0102] Specifically, if the sampling frequency of the vibration signal is inconsistent with the sampling frequency of the breathing audio signal, the missing vibration signal value is generated by interpolation to complete C; if the sampling frequency of the ECG signal is inconsistent with the sampling frequency of the breathing audio signal, the missing ECG signal value is generated by interpolation to complete D. This can eliminate the time domain dislocation caused by different sampling frequencies, and lay the foundation for the subsequent acquisition of the RGB value corresponding to the signal value corresponding to each frame of the breathing audio signal.
[0103] Signal value quantization module 2 is used to map all signal values in B, all signal values in C and all signal values in D to the numerical range [0, 255] to obtain B i The corresponding quantized value B 1 i 、C i The corresponding quantized value C 1 i and D i The corresponding quantization value D 1 iAmong them, those skilled in the art can adopt different mapping methods to obtain quantization values according to actual conditions, for example: using the logarithmic compression method to map the signal value of the breathing audio signal and the signal value of the vibration signal, and using the linear mapping method to map the signal value of the electrocardiogram signal, which will not be repeated here.
[0104] Specifically, max(B1, B2, ..., B i ,……,B m ) corresponds to a quantization value of 255, min(B1, B2, ..., B i ,……,B m ) corresponds to a quantized value of 0, max(C1, C2, ..., C i ,……,C m ) corresponds to a quantization value of 255, min(C1, C2, ..., C i ,……,C m ) corresponds to a quantized value of 0, max(D1, D2, ..., D i ,……,D m ) corresponds to a quantization value of 255, min(D1, D2, ..., D i ,……,D m ) corresponds to a quantization value of 0, max() is a maximum value acquisition function, and min() is a minimum value acquisition function.
[0105] RGB value acquisition module 3, used to obtain B i Corresponding RGB value F i , where B 1 i As F i The intensity value of the red channel, C 1 i As F i The intensity value of the green channel, D 1 i As F i The intensity value of the mid-blue channel.
[0106] Filter threshold combination matrix acquisition module 4 is used to obtain the value of the filter threshold combination matrix based on F i Construct the target image T corresponding to A and input T into the target convolutional neural network model to obtain the filter threshold combination matrix J, where the filter threshold combination in J corresponds one-to-one to the pixel points in T; it can be understood that J includes the filter threshold combination corresponding to each pixel point in T.
[0107] Specifically, J is a two-dimensional matrix, and the position of the filter threshold combination in J is the same as the position of its corresponding pixel point in T.
[0108] Specifically, the filter threshold combination includes a low-frequency cutoff frequency and a high-frequency cutoff frequency.
[0109] Specifically, the target convolutional neural network model is a model obtained by training a preset convolutional neural network for a filter threshold combination prediction task, and the model architecture of the preset convolutional neural network is a U-Net architecture.
[0110] Target filter threshold combination acquisition module 5, used to obtain B based on the filter threshold combination in J i The target filter threshold combination P corresponding to the corresponding breathing audio signal i .
[0111] Specifically, P i The frequency range includes a low frequency cutoff and a high frequency cutoff.
[0112] Noise reduction processing module 6 is used for noise reduction based on dynamic filter and P1, P2, ..., P i ,……,P m Perform noise reduction processing on A to obtain the target audio M corresponding to A.
[0113] Specifically, the filter threshold combination matrix acquisition module 4 includes:
[0114] The sub-audio list obtaining unit is used to divide A into n sub-audio evenly to obtain a sub-audio list E corresponding to A = {E1, E2, ..., E j ,……,E n}, E j is the j-th sub-audio corresponding to A, where j ranges from 1 to n, and n is the preset number of sub-audios.
[0115] Specifically, the duration of each sub-audio is the same and there is no overlap between any two sub-audios.
[0116] Specifically, the duration of each sub-audio is t=T / n, where T is the duration of A.
[0117] Specifically, when j=1, E j The starting time point of A is the starting time point of E. When j≠1, j The starting time point and E j-1 The time interval between the starting time points is t, E j-1 is the j-1th sub-audio corresponding to A.
[0118] The first RGB value matching unit is used for i The corresponding breathing audio signal and E j The e-th frame breathing audio signal in the same frame breathing audio signal, then F i As E jThe RGB value G corresponding to the e-th frame breathing audio signal je , to obtain E j Corresponding RGB value list G j ={G j1 , G j2 ,……,G je ,……,G jf}, where the value of e ranges from 1 to f, and f is the total number of frames of the breathing audio signal in the sub-audio.
[0119] Specifically, n×f=m.
[0120] The first target image construction unit is used to convert G je The RGB value of the pixel at the jth row and the eth column in T is used to obtain T, where the image resolution of T is f×n; that is, the number of horizontal pixels of T is f and the number of vertical pixels is n.
[0121] Concrete, E j The e-th frame of breathing audio signal in corresponds to the pixel point on the j-th row and e-th column in T.
[0122] Specifically, in the target filter threshold combination acquisition module 5, if the breathing audio signal corresponding to a filter threshold combination in J is i The corresponding breathing audio signal is the same frame breathing audio signal, then the filtering threshold combination is used as P i , the breathing audio signal corresponding to the filter threshold combination is the breathing audio signal corresponding to the pixel point corresponding to the filter threshold combination.
[0123] The present invention also provides a specific embodiment, which differs from the above embodiment in that the filter threshold combination matrix acquisition module 4 includes:
[0124] The intermediate audio list acquisition module is used to divide A into h intermediate audios to obtain the intermediate audio list K corresponding to A = {K1, K2, ..., K g ,……,K h}, K g is the gth intermediate audio corresponding to A, where g ranges from 1 to h, and h is the number of intermediate audios.
[0125] Specifically, the duration of each intermediate audio is t1, where t1 is a preset intermediate audio length.
[0126] Specifically, when g=1, K g The starting time point is the starting time point of A. When g≠1, K g The starting time point and K g-1 The time interval between the starting time points is t2, K g-1is the g-1th intermediate audio corresponding to A, and t2 is the preset non-overlapping duration.
[0127] Specifically, t2<t1.
[0128] The second RGB value matching unit is used for i The corresponding breathing audio signal and K g The y-th frame breathing audio signal in the same frame breathing audio signal, then F i As K g The RGB value L corresponding to the breathing audio signal of the yth frame gy , to obtain K g Corresponding RGB value list L g ={L g1 , L g2 ,……,L gy ,……,L gq}, where y ranges from 1 to q, and q is the total number of frames of the breathing audio signal in the intermediate audio.
[0129] The second target image construction unit is used to convert L gy The RGB value of the pixel at the g-th row and y-th column in T is used to obtain T, where the image resolution of T is q×h; that is, the number of horizontal pixels of T is q and the number of vertical pixels is h.
[0130] Specifically, K g The y-th frame breathing audio signal in corresponds to the pixel point on the g-th row and y-th column in T.
[0131] Specifically, the filter threshold combination matrix acquisition module 4 further includes:
[0132] The average respiratory frequency acquisition unit is used to obtain the average respiratory frequency HX corresponding to A. Those skilled in the art know that any method of obtaining the average respiratory frequency corresponding to the respiratory audio in the prior art falls within the scope of protection of the present invention and will not be described in detail here.
[0133] The preset overlap mapping list acquisition unit is used to acquire the preset overlap mapping list YS={YS1, YS2, ..., YS x ,……,YS p}, YS x ={YS x1 , YS x2}, where YS x is the xth preset overlap mapping combination, x ranges from 1 to p, p is the number of preset overlap mapping combinations, YS x1 For YS x Preset respiratory rate in YS x2 For YS x1The corresponding preset overlap.
[0134] Specifically, the greater the preset respiratory frequency, the smaller the corresponding preset overlap.
[0135] Preset non-overlapping duration acquisition unit for when HX=YS x1 When t2 is obtained, t2 meets the following conditions:
[0136] t2=t1×(1-YS x2 ).
[0137] Specifically, the target filtering threshold combination acquisition module 5 includes:
[0138] The intermediate filter threshold combination matrix acquisition unit is used to flatten J into a one-dimensional matrix in a row-first order to obtain the intermediate filter threshold combination matrix Q corresponding to J = [Q1, Q2, ..., Q r ,……,Q s ], Q r =(Q r1 , Q r2 ), Q r is the rth intermediate filter threshold combination in Q, r ranges from 1 to s, s is the number of intermediate filter threshold combinations in Q, Q r1 Q r The low-frequency cutoff frequency, Q r2 Q r The high-frequency cutoff frequency in ; it can be understood as connecting each row in J in sequence to form Q.
[0139] Key filter threshold combination matrix acquisition unit, used to traverse Q, if Q r The corresponding breathing audio signal and Q a The corresponding breathing audio signal is the same frame breathing audio signal and a>r, then let Q r1 =(Q r1 +Q a1 ) / 2, Q r2 =(Q r2 +Q a2 ) / 2, Q a Delete from Q so that Q is updated and Q after all updates is used as the key filter threshold combination matrix U, where Q a is the ath intermediate filter threshold combination in Q, where a ranges from 1 to s and a≠r, Q a1 Q a The low-frequency cutoff frequency, Q a2 Q a The high frequency cutoff frequency in, U=[U1,U2,……,U b ,……,U d ],U b=(U b1 , U b2 ), U b is the bth key filter threshold combination, b ranges from 1 to d, d is the number of key filter threshold combinations, U b1 For U b The low-frequency cutoff frequency, U b2 For U b The high frequency cutoff frequency.
[0140] Specifically, the breathing audio signal corresponding to the intermediate filtering threshold combination is the breathing audio signal corresponding to the pixel point corresponding to the intermediate filtering threshold combination.
[0141] Specifically, d<s.
[0142] Target filter threshold combination acquisition unit, used for b The corresponding breathing audio signal and B i The corresponding breathing audio signal is the same frame breathing audio signal, then U b As P i .
[0143] In a specific embodiment, the target filtering threshold combination acquisition module 5 further includes:
[0144] The basic value acquisition unit is used to obtain the count value z and the preset window length v, wherein z is initially 0, and the preset window length is set by those skilled in the art according to actual needs, for example: 5, 10, which will not be repeated here.
[0145] Judgment unit, used to obtain Y when z+v<d 1 (z，v) , Y 2 (z，v) , W 1 (z，v) and W 2 (z，v) And enter the first key filter threshold combination matrix update unit, when z+v≥d, obtain Y 1 (z，d) , Y 2 (z，d) , W 1 (z，d) and W 2 (z，d) And enter the second key filter threshold combination matrix update unit; Y 1 (z，v) For U z+1 , U z+2 ,……,U z+v The average value of all low-frequency cutoff frequencies in Y 2 (z，v) For U z+1 , Uz+2 ,……,U z+v The variance of all low-frequency cutoff frequencies in , W 1 (z，v) For U z+1 , U z+2 ,……,U z+v The average value of all high frequency cutoff frequencies in W 2 (z，v) For U z+1 , U z+2 ,……,U z+v The variance of all high frequency cutoff frequencies in , Y 1 (z，d) For U z+1 , U z+2 ,……,U d The average value of all low-frequency cutoff frequencies in Y 2 (z，d) For U z+1 , U z+2 ,……,U d The variance of all low-frequency cutoff frequencies in , W 1 (z，d) For U z+1 , U z+2 ,……,U d The average value of all high frequency cutoff frequencies in W 2 (z，d) For U z+1 , U z+2 ,……,U d The variance of all high frequency cutoff frequencies in , U z+1 The z+1th key filter threshold combination, U z+2 The z+2th key filter threshold combination, U z+v The z+vth key filter threshold combination.
[0146] The first key filter threshold combination matrix update unit is used if W 1 (z，v) ≤W 0 And W 2 (z，v) ≤W 0 , then let z=z+1 and enter the judgment unit; if W 1 (z，v) >W 0 And W 2 (z，v) ≤W 0 , then Y 1 (z，v) As U z+1 , U z+2 ,……,U z+vThe low-frequency cutoff frequency in the filter is used to update U, set z = z + v and enter the judgment unit; if W 1 (z，v) ≤W 0 And W 2 (z，v) >W 0 , then Y 2 (z，v) As U z+1 , U z+2 ,……,U z+v The high frequency cutoff frequency in the filter is used to update U, set z = z + v and enter the judgment unit; if W 1 (z，v) >W 0 And W 2 (z，v) >W 0 , then Y 1 (z，v) As U z+1 , U z+2 ,……,U z+v The low frequency cutoff frequency in Y 2 (z，v) As U z+1 , U z+2 ,……,U z+v The high frequency cutoff frequency in the filter is used to update U, set z = z + v and enter the judgment unit, W 0 is the preset variance.
[0147] The first key filter threshold combination matrix update unit is used if W 1 (z，d) ≤W 0 And W 2 (z，d) ≤W 0 , then enter the target filter threshold combination acquisition unit; if W 1 (z，d) >W 0 And W 2 (z，d) ≤W 0 , then Y 1 (z，d) As U z+1 , U z+2 ,……,U d The low-frequency cutoff frequency in the filter is used to update U and enter the target filter threshold combination acquisition unit; if W 1 (z，d) ≤W 0 And W 2 (z，d) >
[0148] W 0 , then Y2 (z，d) As U z+1 , U z+2 ,……,U d The high frequency cutoff frequency in the filter is used to update U and enter the target filter threshold combination acquisition unit; if W 1 (z，d) >W 0 And W 2 (z，d) >W 0 , then Y 1 (z，d) As U z+1 , U z+2 ,……,U d The low frequency cutoff frequency in Y 2 (z，d) As U z+1 , U z+2 ,……,U d The high frequency cutoff frequency in is used to update U and enter the target filter threshold combination acquisition unit.
[0149] Target filter threshold combination acquisition unit, used for b The corresponding breathing audio signal and B i The corresponding breathing audio signal is the same frame breathing audio signal, then U b As P i .
[0150] An embodiment of the present invention also provides a non-transitory computer-readable storage medium, which can be set in an electronic device to store a computer program related to a method in the method embodiment. The computer program is loaded and executed by the processor to implement the method provided in the above embodiment.
[0151] An embodiment of the present invention further provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method provided in the above embodiment when executing the computer program.
[0152] An embodiment of the present invention further provides a computer program product comprising program code. When the program product is run on an electronic device, the program code is used to enable the electronic device to execute the steps of the method according to various exemplary embodiments of the present invention described above in this specification.
[0153] The present invention provides a method, device and medium for processing biological audio data. The method can obtain a biological audio data set corresponding to the breathing audio of a target organism, process the data in the biological audio data set, obtain the RGB value corresponding to the signal value of each frame of the breathing audio signal in the breathing audio, construct a target image corresponding to the breathing audio based on the RGB value corresponding to the signal value of each frame of the breathing audio signal, and input the target image into a target convolutional neural network model to obtain a filter threshold combination matrix, obtain a target filter threshold combination corresponding to each frame of the breathing audio signal based on the filter threshold combination in the filter threshold combination matrix, wherein the filter threshold combination includes a low-frequency cutoff frequency and a high-frequency cutoff frequency, and perform noise reduction processing on the breathing audio based on a dynamic filter and the target filter threshold combination corresponding to each frame of the breathing audio signal to obtain the target audio corresponding to the breathing audio. It can be seen that the present invention can obtain the target image corresponding to the breathing audio based on the data in the biological audio data set corresponding to the breathing audio, obtain the filter threshold combination matrix based on the target image and the target convolutional neural network model, and further obtain the target filter threshold combination corresponding to each frame of the breathing audio signal, so that each frame of the breathing audio signal corresponds to a unique target filter threshold combination, and perform noise reduction processing on the breathing audio based on the dynamic filter and the unique target filter threshold combination corresponding to each frame of the breathing audio signal to obtain the target audio corresponding to the breathing audio, which can improve the accuracy of noise reduction, thereby avoiding the loss of effective signals or the occurrence of noise residues.
[0154] Although some specific embodiments of the present invention have been described in detail by way of examples, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present invention. It should also be understood by those skilled in the art that various modifications may be made to the embodiments without departing from the scope and spirit of the present invention.
Claims
1. A method for processing biological audio data, characterized in that: The method comprises the following steps: S1. Obtain the biological audio data set S = {B, C, D} corresponding to the breathing audio A of the target organism, where B is a list of breathing audio signal values, C is a list of vibration signal values, and D is a list of electrocardiogram signal values. B includes the signal value B of the breathing audio signal of the i-th frame in A. i , i ranges from 1 to m, m is the total number of frames of the breathing audio signal in A, C includes i The signal value C of the vibration signal collected synchronously i , D includes B i The signal value D of the synchronously collected ECG signal i ; S2, respectively map all signal values in B, all signal values in C, and all signal values in D to the numerical interval [0, 255] to obtain B i The corresponding quantized value B 1 i 、C i The corresponding quantized value C 1 i and D i The corresponding quantized value D 1 i ; S3. Get B i Corresponding RGB value F i , where B 1 i As F i The intensity value of the red channel, C 1 i As F i The intensity value of the green channel, D 1 i As F i The intensity value of the blue channel; S4, based on F i Construct the target image T corresponding to A and input T into the target convolutional neural network model to obtain the filter threshold combination matrix J. The filter threshold combination in J corresponds one-to-one to the pixel points in T. The filter threshold combination includes a low-frequency cutoff frequency and a high-frequency cutoff frequency. S5. Obtain B based on the filter threshold combination in J i The target filter threshold combination P corresponding to the corresponding breathing audio signal i ; S6, based on dynamic filter and P1, P2, ..., P i ,……,P m Perform noise reduction processing on A to obtain the target audio M corresponding to A.
2. The method for processing biological audio data according to claim 1, characterized in that: In step S4, the following steps are also included to obtain T: S41, evenly divide A into n sub-audio numbers to obtain a sub-audio list E corresponding to A = {E1, E2, ..., E j ,……,E n }, E j The jth sub-audio corresponding to A, where j ranges from 1 to n and n is the number of preset sub-audios; S42, if B i The corresponding breathing audio signal and E j The e-th frame breathing audio signal in the same frame breathing audio signal, then F i As E j The RGB value G corresponding to the e-th frame breathing audio signal je , to obtain E j Corresponding RGB value list G j ={G j1 , G j2 ,……,G je ,……,G jf }, where the value of e ranges from 1 to f, and f is the total number of frames of the breathing audio signal in the sub-audio; S43, G je The RGB value of the pixel at the jth row and the eth column in T is used to obtain T, where the image resolution of T is f×n.
3. The method for processing biological audio data according to claim 2, characterized in that: The duration of each sub-audio is the same and there is no overlap between any two sub-audios.
4. The method for processing biological audio data according to claim 2, wherein: E j The e-th frame of breathing audio signal in corresponds to the pixel point on the j-th row and e-th column in T.
5. The method for processing biological audio data according to claim 4, characterized in that: In step S5, if the breathing audio signal corresponding to a certain filter threshold combination in J is i The corresponding breathing audio signal is the same frame breathing audio signal, then the filtering threshold combination is used as P i , the breathing audio signal corresponding to the filter threshold combination is the breathing audio signal corresponding to the pixel point corresponding to the filter threshold combination.
6. The method for processing biological audio data according to claim 1, wherein: Step S4 includes the following steps to obtain T: S401, divide A into h intermediate audios to obtain an intermediate audio list K corresponding to A = {K1, K2, ..., K g ,……,K h }, K g is the gth intermediate audio corresponding to A, the value of g ranges from 1 to h, h is the number of intermediate audios, the duration of each intermediate audio is t1, t1 is the preset intermediate audio length, when g = 1, K g The starting time point is the starting time point of A. When g≠1, K g The starting time point and K g-1 The time interval between the starting time points is t2, K g-1 is the g-1th intermediate audio corresponding to A, t2 is the preset non-overlapping duration and t2<t1; S402, if B i The corresponding breathing audio signal and K g The y-th frame breathing audio signal in the same frame breathing audio signal, then F i As K g The RGB value L corresponding to the breathing audio signal of the yth frame gy , to obtain K g Corresponding RGB value list L g ={L g1 , L g2 ,……,L gy ,……,L gq }, where y ranges from 1 to q, and q is the total number of frames of the breathing audio signal in the intermediate audio; S403, L gy The RGB value of the pixel at the g-th row and y-th column in T is used to obtain T, where the image resolution of T is q×h.
7. The method for processing biological audio data according to claim 6, characterized in that: K g The y-th frame breathing audio signal in corresponds to the pixel point on the g-th row and y-th column in T.
8. The method for processing biological audio data according to claim 7, characterized in that: Step S5 includes the following steps: S51. Flatten J into a one-dimensional matrix in a row-first order to obtain the intermediate filter threshold combination matrix Q corresponding to J = [Q1, Q2, ..., Q r ,……,Q s ], Q r =(Q r1 , Q r2 ), Q r is the rth intermediate filter threshold combination in Q, r ranges from 1 to s, s is the number of intermediate filter threshold combinations in Q, Q r1 Q r The low-frequency cutoff frequency, Q r2 Q r The high frequency cutoff frequency in ; S52, traverse Q, if Q r The corresponding breathing audio signal and Q a The corresponding breathing audio signal is the same frame breathing audio signal and a>r, then let Q r1 =(Q r1 +Q a1 ) / 2, Q r2 =(Q r2 +Q a2 ) / 2, Q a Delete from Q so that Q is updated and Q after all updates is used as the key filter threshold combination matrix U, where Q a is the ath intermediate filter threshold combination in Q, where a ranges from 1 to s and a≠r, Q a1 Q a The low-frequency cutoff frequency, Q a2 Q a The high frequency cutoff frequency in, U=[U1,U2,……,U b ,……,U d ],U b =(U b1 , U b2 ), U b is the bth key filter threshold combination, b ranges from 1 to d, d is the number of key filter threshold combinations, U b1 For U b The low-frequency cutoff frequency, U b2 For U b The high frequency cutoff frequency in ; S53, if U b The corresponding breathing audio signal and B i The corresponding breathing audio signal is the same frame breathing audio signal, then U b As P i .
9. A biological audio data processing device, characterized in that: The device comprises: The biological audio data acquisition module is used to obtain the biological audio data set S = {B, C, D} corresponding to the breathing audio A of the target organism, where B is a list of breathing audio signal values, C is a list of vibration signal values, and D is a list of electrocardiogram signal values. B includes the signal value B of the breathing audio signal of the i-th frame in A. i , i ranges from 1 to m, m is the total number of frames of the breathing audio signal in A, C includes i The signal value C of the vibration signal collected synchronously i , D includes B i The signal value D of the synchronously collected ECG signal i ; The signal value quantization module is used to map all signal values in B, all signal values in C, and all signal values in D to the numerical interval [0, 255] to obtain B i The corresponding quantized value B 1 i 、C i The corresponding quantized value C 1 i and D i The corresponding quantized value D 1 i ; RGB value acquisition module, used to obtain B i Corresponding RGB value F i , where B 1 i As F i The intensity value of the red channel, C 1 i As F i The intensity value of the green channel, D 1 i As F i The intensity value of the blue channel; Filter threshold combination matrix acquisition module, used for i Construct the target image T corresponding to A and input T into the target convolutional neural network model to obtain the filter threshold combination matrix J. The filter threshold combination in J corresponds one-to-one to the pixel points in T. The filter threshold combination includes a low-frequency cutoff frequency and a high-frequency cutoff frequency. The target filter threshold combination acquisition module is used to obtain B based on the filter threshold combination in J i The target filter threshold combination P corresponding to the corresponding breathing audio signal i ; Noise reduction processing module for dynamic filter and P1, P2, ..., P i ,……,P m Perform noise reduction processing on A to obtain the target audio M corresponding to A.
10. A non-transitory computer-readable storage medium, characterized in that The storage medium stores a computer program, which is loaded and executed by a processor to implement the method for processing biological audio data according to any one of claims 1 to 8.