A biological radar speech enhancement method and system based on variational mode decomposition

Abstract

The application discloses a biological radar speech enhancement method and system based on variational mode decomposition. The method comprises the following steps: obtaining biological radar speech to obtain an original noisy radar speech signal; performing empirical mode decomposition on the intrinsic mode function feature of the original noisy radar speech signal; determining the number of layers of variational mode decomposition according to the intrinsic mode function feature; performing variational mode decomposition VMD on the original noisy radar speech according to the number of layers to obtain each order intrinsic mode decomposition; solving the Pearson coefficient of each order intrinsic mode decomposition and the original radar speech signal to determine useful decomposition and reconstruction modes; and performing denoising processing on the useful modes by using an improved threshold to reconstruct the enhanced radar speech. The application has strong adaptability, can significantly improve the quality and intelligibility of speech on the basis of eliminating noise in the radar speech, and therefore has strong use value and application prospect in the aspect of radar speech enhancement.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of speech enhancement, and particularly relates to a biological radar speech enhancement method and system based on variational mode decomposition. BACKGROUND

[0002] The biological radar speech detection technology overcomes the shortcomings of the traditional microphone speech detection device, such as being easily disturbed by environmental acoustic noise, the throat microphone and the bone conduction microphone needing to be in contact with the human skin, and the optical speech detection sensor being easily affected by environmental factors such as temperature and climate, and has the advantages of non-contact, non-invasive, safety, good directivity, high sensitivity, long detection distance, strong anti-interference ability and the like. The biological radar technology provides a new way for speech signal acquisition. At present, the 94GHz biological radar can provide a good compromise in terms of detection distance and detection sensitivity, but there is still a problem of short detection distance. The radar speech using electromagnetic waves as a medium is often superimposed with circuit noise, electromagnetic interference noise and harmonic noise. These noises greatly reduce the quality of the radar speech. Therefore, it is of great significance to study the radar speech enhancement method for the development of the new speech detection technology. SUMMARY

[0003] The present application aims to provide a 94GHz asymmetric antenna biological radar speech enhancement method, which determines the number of layers of variational mode decomposition according to empirical mode decomposition, then performs variational mode decomposition on the original radar speech signal according to the number of layers, and finally effectively removes the noise content in the radar speech by using an improved threshold strategy. The present application provides technical support for the development of biological radar speech detection technology in terms of speech denoising.

[0004] To achieve the above-mentioned purpose, the present application adopts the following technical solutions.

[0005] A biological radar speech enhancement method based on variational mode decomposition, comprising the following steps:

[0006] Obtaining a biological radar speech to obtain an original noisy radar speech signal;

[0007] Performing empirical mode decomposition on the intrinsic mode function feature of the original noisy radar speech signal to determine the number of layers of variational mode decomposition according to the intrinsic mode function feature;

[0008] Performing variational mode decomposition VMD on the original noisy radar speech according to the number of decomposition layers to obtain each order intrinsic mode decomposition;

[0009] Solving the Pearson coefficient of each order intrinsic mode decomposition and the original radar speech signal to determine the useful decomposition reconstruction mode;

[0010] The useful modal is denoised by using the improved threshold value, and the enhanced radar voice is reconstructed.

[0011] As a further improvement of the application, the biological radar voice is sampled by using an 8-channel Powerlab physiological recorder, and the sampling rate is 10 kHz.

[0012] As a further improvement of the application, the number of layers of the variational modal decomposition is determined according to the intrinsic modal function feature of the empirical mode decomposition, comprising:

[0013] Given a speech signal x(t), the original signal is adaptively decomposed into intrinsic modal functions by the screening process of the empirical mode decomposition method, and each intrinsic modal function IMF is a sub-frequency component of the original signal; by analyzing the amplitude characteristics of each IMF from high frequency to low frequency, when the amplitude value of the IMF is less than a set value, the mode is the speech detail information, and the remaining modes are the main information, and the number of main information modes is the number of layers of the variational modal decomposition;

[0014] The first IMF whose amplitude value is less than the set value from top to bottom after EMD decomposition is the number of main information modes, which is obtained by the following formula:

[0015]

[0016] In the formula, IMF i is each order intrinsic modal function after empirical mode decomposition, and i represents the order of the intrinsic modal function.

[0017] As a further improvement of the application, the original noisy radar voice is variational modal decomposed VMD to obtain each order intrinsic modal decomposition according to the number of decomposition layers, comprising:

[0018] Given the original radar voice signal x(t), the analytic signal of each mode u k is obtained by Hilbert transform, and a variational constraint problem is constructed:

[0019]

[0020]

[0021] x(t) is decomposed into K modes, wherein u k (k=1, 2, …, K) is the K modes of the VMD method, w k (k=1, 2, …, K) is the center frequency of each mode; a Lagrange multiplier operator is introduced:

[0022]

[0023] Wherein, alpha is the control bandwidth penalty factor, lambda (t) is the Lagrange multiplier operator, * represents convolution, delta (t) is the unit impulse function;

[0024] The multiplication operator alternating direction method and Parseval Fourier equidistant transformation are adopted, and each mode is converted from time domain to frequency domain by (2) formula:

[0025]

[0026] Wherein, the center frequency of each mode is:

[0027]

[0028] Wherein w k n+1 Is the power spectrum center of the kth mode.

[0029] As a further improvement of the application, the Pearson coefficient is solved for each order of the intrinsic mode decomposition and the original radar speech signal, comprising:

[0030]

[0031] Wherein, x t And y t Represent two random variables.

[0032] As a further improvement of the application, the improved threshold value is represented by the following formula:

[0033]

[0034] N j The length of the jth scale signal, sigma is the estimated noise of each mode signal, wherein the improved mode estimated noise can be calculated by the following formula:

[0035]

[0036] Wherein, L is the estimated length of the initial silence section of the radar speech signal.

[0037] As a further improvement of the application, the useful mode is denoised by using the improved threshold value, comprising:

[0038]

[0039] Wherein, m is a compensation factor.

[0040] As a further improvement of the application, the compensation factor m is 0.001.

[0041] As a further improvement of the present application, the reconstructed enhanced radar speech is the 94GHz radar speech after denoising by the improved threshold strategy, and the specific reconstruction is calculated according to the following formula:

[0042]

[0043] Wherein, IMF is K modes of VMD decomposition, and a is the number of useful modes determined by calculating the Pearson coefficient.

[0044] A biological radar speech enhancement system based on variational mode decomposition includes the following steps:

[0045] The acquisition module is used to acquire the biological radar speech to obtain an original noisy radar speech signal.

[0046] The decomposition module is used to perform empirical mode decomposition on the intrinsic mode function feature of the original noisy radar speech signal, determine the number of variational mode decomposition layers according to the intrinsic mode function feature, perform variational mode decomposition VMD on the original noisy radar speech according to the number of decomposition layers to obtain each order intrinsic mode decomposition, and solve the Pearson coefficient of each order intrinsic mode decomposition and the original radar speech signal to determine the useful decomposition and reconstruction mode.

[0047] The reconstruction module is used to denoise the useful mode by using the improved threshold value, and reconstruct the enhanced radar speech.

[0048] The present application has the following beneficial effects compared with the existing radar speech enhancement technology by using the above-mentioned technology:

[0049] The present application is through the empirical mode decomposition of the acquired 94GHz asymmetric antenna radar speech signal, determines the decomposition layer number, performs variational mode decomposition according to the decomposition layer number, then estimates the noise variance according to the speech silence section information, and effectively improves the quality and intelligibility of the speech without causing distortion of the speech signal, has strong adaptability and effectiveness. The examples of using this method show that this variational state decomposition method can effectively improve the quality and intelligibility of the radar speech, compared with the traditional speech enhancement method, this method can improve the score of the original radar speech as a whole. It is proved that the present application can effectively enhance the original radar speech signal. The present application can provide effective technical support for the use of biological radar to detect human speech in the future. Therefore, the present application has strong use value and application prospect in eliminating radar speech noise. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 It is a biological radar speech enhancement method flow chart based on variational mode decomposition and improved threshold strategy;

[0051] Figure 2is a radar speech material after empirical mode decomposition, wherein (a) is a raw radar speech signal containing noise; (b) is each intrinsic mode function after decomposition;

[0052] Figure 3 is a radar speech material after variational mode decomposition, wherein (a) is each intrinsic mode function after decomposition; (b) is a frequency spectrum of each intrinsic mode function;

[0053] Figure 4 is a Pearson coefficient of a raw radar speech signal and a mode after variational mode decomposition;

[0054] Figure 5 is a raw radar speech signal collected by a radar system, wherein (a) is a time-domain waveform of the radar speech signal; (b) is a spectrogram of the radar speech signal;

[0055] Figure 6 is a radar speech signal after enhancement by a wavelet soft threshold algorithm, wherein (a) is a time-domain waveform of the enhanced radar speech signal; (b) is a spectrogram of the enhanced radar speech signal;

[0056] Figure 7 is a radar speech signal after enhancement by an empirical mode decomposition algorithm, wherein (a) is a time-domain waveform of the enhanced radar speech signal; (b) is a spectrogram of the enhanced radar speech signal;

[0057] Figure 8 is a radar speech signal enhanced based on variational mode decomposition and an improved threshold strategy according to the present patent, wherein (a) is a time-domain waveform of the enhanced radar speech signal; (b) is a spectrogram of the enhanced radar speech signal. DETAILED DESCRIPTION

[0058] In order to enable persons skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in combination with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by persons skilled in the art without creative labor should belong to the protection scope of the present application.

[0059] It is to be understood that the terms "first", "second", and the like, used in the description and the claims of the application, as well as the above-described drawings, are used to distinguish between similar objects and are not necessarily used to describe a particular sequential or chronological order. It is to be understood that the use of the terms so-termed "first", "second", etc. can be interchanged, where appropriate, to refer to the same element in order to describe the embodiments of the application described herein, which can be implemented in an order other than that illustrated or described herein. Furthermore, the terms "comprise" and "have", as well as any variations thereof, are intended to cover non-exclusive inclusions, for example, a process, method, system, product or apparatus comprising a list of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to such a process, method, product or apparatus.

[0060] The application provides a biological radar speech enhancement method based on variational mode decomposition, comprising the following steps:

[0061] 1) Sampling the biological radar speech to obtain an original noisy radar speech signal;

[0062] In step 1), the Powerlab physiological recorder with 8 channels is used to sample the biological radar speech, and the sampling rate is 10 kHz.

[0063] 2) The collected original noisy radar speech signal is firstly subjected to empirical mode decomposition, and the number of decomposition layers is determined according to the characteristics of the intrinsic mode function after empirical mode decomposition;

[0064] In step 2), the number of decomposition layers determined by empirical mode decomposition is as follows:

[0065] Suppose that the given speech signal is x(t), the empirical mode decomposition method adaptively decomposes the original signal into intrinsic mode functions through a screening process, and each intrinsic mode function IMF is a sub-frequency component of the original signal. By analyzing the amplitude characteristics of each IMF from high frequency to low frequency, it is determined that when the amplitude value of the IMF is less than 0.1, the mode is the speech detail information, and the remaining modes are the main information, and the number of main information modes is the number of decomposition layers of the variational mode decomposition. Then, the IMF after empirical mode decomposition EMD is from top to bottom, and the first IMF whose amplitude value is less than 0.1 is the number of main information modes, which can be obtained by the following formula:

[0066]

[0067] 3) Decomposing the original noisy radar speech according to the number of decomposition layers;

[0068] Step 3) The variational mode decomposition (VMD) method is as follows:

[0069] Given the original radar speech signal x(t), first the analytic signal of each mode u k is obtained by Hilbert transform, and the variational constraint problem is constructed:

[0070]

[0071]

[0072] x(t) is decomposed into K modes, where u k (k=1, 2, …, K) is the K modes of VMD method, w k (k=1, 2, …, K) is the center frequency of each mode. The solution of each mode requires the constraint problem of equation (1) to be transformed into a non-constrained variational problem. The Lagrange multiplier operator is introduced:

[0073]

[0074] where α is the penalty factor controlling the bandwidth, λ(t) is the Lagrange multiplier, * represents convolution, and δ(t) is the unit impulse function.

[0075] The alternating direction method of multipliers and Parseval Fourier equidistant transform are used to convert each mode from time domain to frequency domain by equation (2):

[0076]

[0077] where the center frequency of each mode is:

[0078]

[0079] where w k n+1 is the power spectrum center of the kth mode.

[0080] 4) Calculate the Pearson coefficient between each order eigenmode decomposition of variational mode decomposition and the original radar speech signal; to determine the useful decomposition and reconstruction mode;

[0081] 5) Calculate the Pearson coefficient between each order eigenmode decomposition of variational mode decomposition and the original radar speech signal; to determine the Pearson coefficient calculation method of the original radar speech signal and VMD decomposition mode in step 3) as follows:

[0082]

[0083] x t and y t represent two random variables. For radar speech enhancement, experiments show that when the correlation coefficient is greater than 0.1, it is the threshold for dividing whether it is a useful mode.

[0084] The improved threshold strategy method is as follows, and the adaptive soft threshold of each mode is represented by the following formula:

[0085]

[0086] N j The length of the jth scale signal, and sigma is the estimated noise of each mode signal. The improved noise of each mode can be calculated by the following formula:

[0087]

[0088] Wherein, L is the estimated length of the initial silent section of the radar speech signal.

[0089] The improved soft threshold function of the improved threshold strategy is used to denoise the useful mode according to the following formula:

[0090]

[0091] Wherein, m is a compensation factor, which prevents the past noise from causing distortion of the speech signal.

[0092] The value of the compensation factor m is 0.001.

[0093] 6) The improved threshold is used to denoise the useful mode, and then the enhanced radar speech is reconstructed.

[0094] Wherein, the 94GHz radar speech denoised by the improved threshold strategy is reconstructed to obtain the enhanced speech, and the specific reconstruction is calculated according to the following formula:

[0095]

[0096] Wherein, IMF is K each mode of VMD decomposition, and a is the number of useful modes determined by calculating the Pearson coefficient.

[0097] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings, but the present application is not limited to this embodiment. In order to make the public have a thorough understanding of the present application, the specific details are described in detail in the following preferred examples of the present application.

[0098] Referring to Figure 1 The basic principle of the present application based on the variational mode decomposition and the improved threshold strategy method is as follows: first, the collected radar speech signal is decomposed by empirical mode decomposition to determine the number of decomposition layers; then, according to the number of decomposition layers, the collected radar speech signal is decomposed by empirical mode decomposition; the Pearson coefficient of the original radar speech signal and each order mode function is calculated to determine the useful mode; the useful mode is denoised by the improved threshold strategy; and the processed useful mode is reconstructed to obtain the enhanced speech.

[0099] The processing results are compared, please refer to Figure 6 、 Figure 7 、 Figure 8 Figures (a) are time-domain waveforms of the radar speech signals, and figures (b) are spectrograms of the radar speech signals.

[0100] As can be seen from Figure 5 , the original radar speech signal has noise in the entire frequency band, and there is also single-band noise interference, which leads to low quality of the collected radar speech. Figure 6 is the radar speech signal enhanced by the wavelet soft threshold algorithm. As can be seen from the figure, the wavelet soft threshold effectively removes the noise in the radar speech. However, observing the spectrogram, the speech horizontal stripes have a certain deformation, and new noise is introduced, in addition, there is still residual noise in the low frequency band.

[0101] Figure 7 is the radar speech signal enhanced by empirical mode decomposition. As can be seen from the figure, the noise in the original radar speech is greatly suppressed, and whether in the time-domain waveform or in the spectrogram, the quality of the speech is greatly improved. However, observing the spectrogram, the clarity between the horizontal stripes is not high, so there is still a deficiency in radar speech enhancement.

[0102] Figure 8 is the radar speech signal enhanced by the method of variational mode decomposition and improved threshold strategy. As can be seen from the figure, the noise in the entire frequency band of the radar speech signal enhanced by this method is weakened, and the objective evaluation method experiment shows that the method of the present application can effectively improve the quality and intelligibility of the radar speech without distorting the original radar speech signal.

[0103] Although the speech enhancement method based on variational mode decomposition and improved threshold strategy discussed in the present application is targeted and especially suitable for 94GHz asymmetric antenna biological radar speech signals, the scope of use of the present application is not limited to 94GHz asymmetric antenna radar speech signals. It also has important guiding significance and reference value for other millimeter wave, centimeter wave radar speech signals, and some acoustic speech signals collected in the same environment. The embodiments of the present application are described in detail above in combination with the drawings, which are not intended to limit the present application in any way. Any simple modification, change and equivalent structural change made according to the technical essence of the present application to the above embodiments are still within the protection scope of the technical solution of the present application.

[0104] In order to ensure the consistency of the sound source, the recording material "1-2-3-4-5-6" of a male experimenter was selected, and played by a sound box at a distance of 10 meters from the 94GHz asymmetric antenna biological radar system.

[0105] The radar speech is enhanced in the following steps:

[0106] 1) The radar speech is sampled by using an 8-channel Powerlab physiological recorder, and the sampling rate is 10 kHz; the original noisy radar speech signal is obtained after sampling;

[0107] 2) The noisy radar speech signal is decomposed into intrinsic mode functions by empirical mode decomposition, and the main information layer is determined;

[0108] 3) The noisy radar speech signal is decomposed by variational mode decomposition according to the determined decomposition layer;

[0109] 4) The Pearson correlation coefficients of the decomposed modes and the original radar speech signal are calculated, and the useful modes are determined;

[0110] 5) The useful modes are denoised by using the improved threshold strategy;

[0111] 6) The denoised useful modes are reconstructed to obtain the enhanced radar speech.

[0112] Comparison of processing results:

[0113] From Figure 5 it can be seen that the original radar speech signal has noise in the entire frequency band, and there is also single frequency band noise interference, which leads to low quality of the collected radar speech. Figure 6 is the radar speech signal enhanced by the wavelet soft threshold algorithm. From the figure, it can be seen that the wavelet soft threshold effectively removes the noise in the radar speech. However, observing the spectrogram, the speech horizontal stripes have a certain deformation, and new noise is introduced, in addition, there is still residual noise in the low frequency band. Figure 7 is the radar speech signal enhanced by empirical mode decomposition. From the figure, it can be seen that the noise in the original radar speech is greatly suppressed, and whether in the time domain waveform or in the spectrogram, the quality of the speech is greatly improved, but observing the spectrogram, the clarity between the horizontal stripes is not high, so there is still a deficiency in radar speech enhancement. Figure 8 is the radar speech signal enhanced by the variational mode decomposition and improved threshold strategy method. From the figure, it can be seen that the noise in the radar speech enhanced by this method is weakened in the entire frequency band, and the objective evaluation method experiment shows that the method described in the patent can effectively improve the quality and intelligibility of the radar speech without causing distortion of the original radar speech signal. This is mainly because the method can effectively decompose the original radar speech signal into mode components of different frequencies, effectively prevent over-decomposition and under-decomposition, and suppress mode aliasing.

[0114] In addition, the improved threshold strategy of the present application is based on the average energy of the initial silent section to calculate the estimated value of the modal noise after the variational modal decomposition, so the accuracy is higher. In addition, the original soft threshold function is modified by using a compensation factor to modify the soft threshold function, so that not only the excessive damage to the speech signal can be effectively avoided, but also the residual noise in the noise radar speech can be effectively suppressed.

[0115] The present application also provides a biological radar speech enhancement system based on variational modal decomposition, comprising the following steps:

[0116] The acquisition module is used to acquire the biological radar speech to obtain an original noisy radar speech signal.

[0117] The decomposition module is used to perform empirical mode decomposition on the intrinsic mode function feature of the original noisy radar speech signal, determine the number of variational modal decomposition layers according to the intrinsic mode function feature, perform variational modal decomposition VMD on the original noisy radar speech according to the number of decomposition layers to obtain each order intrinsic modal decomposition, and solve the Pearson coefficient of each order intrinsic modal decomposition and the original radar speech signal to determine a useful decomposition and reconstruction mode.

[0118] The reconstruction module is used to perform denoising processing on the useful mode by using the improved threshold to reconstruct an enhanced radar speech.

[0119] The present application provides an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the biological radar speech enhancement method based on variational modal decomposition when executing the computer program.

[0120] The biological radar speech enhancement method based on variational modal decomposition comprises:

[0121] The biological radar speech is acquired to obtain an original noisy radar speech signal.

[0122] The intrinsic mode function feature of the original noisy radar speech signal is subjected to empirical mode decomposition, and the number of variational modal decomposition layers is determined according to the intrinsic mode function feature.

[0123] The original noisy radar speech is subjected to variational modal decomposition VMD according to the number of decomposition layers to obtain each order intrinsic modal decomposition.

[0124] The Pearson coefficient of each order intrinsic modal decomposition and the original radar speech signal is solved to determine a useful decomposition and reconstruction mode.

[0125] The useful mode is subjected to denoising processing by using the improved threshold to reconstruct an enhanced radar speech.

[0126] The application further provides a computer readable storage medium storing a computer program, the computer program being executed by a processor to implement steps of the biological radar speech enhancement method based on variational modal decomposition.

[0127] The biological radar speech enhancement method based on variational modal decomposition comprises the following steps:

[0128] Obtaining biological radar speech to obtain an original noisy radar speech signal;

[0129] Performing intrinsic mode function feature of empirical mode decomposition on the original noisy radar speech signal, and determining a variational modal decomposition layer number according to the intrinsic mode function feature;

[0130] Performing variational modal decomposition VMD on the original noisy radar speech according to the decomposition layer number to obtain each order intrinsic modal decomposition;

[0131] Solving a Pearson coefficient of each order intrinsic modal decomposition and the original radar speech signal to determine a useful decomposition reconstruction mode;

[0132] Performing denoising processing on the useful mode by using an improved threshold value to reconstruct an enhanced radar speech.

[0133] Those skilled in the art should understand that embodiments of the application can be provided as a method, a system, or a computer program product. Therefore, the application can adopt a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can adopt a computer program product in the form of being implemented on one or more computer usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer usable program code.

[0134] The application is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks

[0135] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 The functions of a flow or multiple flows and / or a block or multiple blocks in conjunction with the disclosed methods. Figure 1

[0136] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flow Figure 1 The functions of a flow or multiple flows and / or a block or multiple blocks in conjunction with the disclosed methods. Figure 1 The functions of a flow or multiple flows and / or a block or multiple blocks in conjunction with the disclosed methods.

[0137] Finally, it should be noted that the above-mentioned embodiments are merely used to illustrate the technical solutions of the present application, rather than limiting the same. Even though the present application has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced equivalently, and any modification or replacement without departing from the spirit and scope of the present application should be covered within the protection scope of the claims of the present application.​

Claims

1. A method for enhancing speech in bio-radar based on variational mode decomposition, characterized in that, Includes the following steps: Acquire the 94GHz asymmetric antenna radar voice signal to obtain the original noisy radar voice signal; Empirical mode decomposition is performed on the original noisy radar speech signal to obtain intrinsic mode function features, and the number of variational mode decomposition layers is determined based on the intrinsic mode function features; Based on the number of decomposition layers, variational mode decomposition (VMD) is performed on the original noisy radar speech to obtain the eigenmode decompositions of each order. Pearson coefficients are calculated from the decomposed intrinsic modes of each order and the original radar speech signal to determine the useful decomposed and reconstructed modes. The useful modalities are denoised using an improved threshold, and the enhanced radar speech is reconstructed. The step of determining the number of variational mode decomposition layers based on the characteristics of intrinsic mode functions includes: Given a speech signal x ( t The empirical mode decomposition method adaptively decomposes the original signal into local oscillator mode functions through a sieving process, and each local oscillator mode function (IMF) is a sub-frequency component of the original signal. By analyzing the amplitude characteristics of each IMF from high frequency to low frequency, when the amplitude value of the IMF is less than a set value, the mode is speech detail information, and the other modes are main information. The number of main information modes is the number of variational mode decomposition layers. The order of the first IMF (Information Mode Factorization) with an amplitude value less than a set value after Empirical Mode Decomposition (EMD) is the primary information mode number, obtained from the following formula: In the formula, IMF i These are the eigenmode functions of each order after empirical mode decomposition. i Represents the order of the eigenmode function; The step of performing variational mode decomposition (VMD) on the original noisy radar speech according to the number of decomposition layers to obtain eigenmode decompositions of each order includes: Given the original radar voice signal as x ( t The modes are obtained through Hilbert transform. u k The analytical signal is obtained, and a variational constraint problem is constructed: Will x ( t The decomposition is divided into K modes, where u k ( k =1,2,…,K) represent the K modes of the VMD method. w k (k=1,2,…,K) represents the center frequency of each mode; the Lagrange multiplier operator is introduced: in, α It is a penalty factor for controlling bandwidth. λ ( t ) represents the Lagrange multiplication operator, * represents convolution. δ ( t Unit impulse function; Using the alternating direction method of multiplication operators and Parseval Fourier isometric transform, the modes are transformed from the time domain to the frequency domain by equation (2): The center frequencies of each mode are: in wkn +1 is the power spectrum center of the k-th mode; The process of decomposing the intrinsic mode shapes of each order and solving for the Pearson coefficients from the original radar speech signal includes: in, x t and y t Represents two random variables; The improved threshold is expressed by the following formula: N j The length of the j-th scale signal, σ, is the estimated noise of each modal signal; where the improved estimated noise of each modal can be calculated by the following formula: Where L is the estimated length of the initial silent segment of the radar voice signal during the period when there is no voice activity on the 94GHz radar. Denoising of useful modes using improved thresholding includes: in, m This is a compensation factor.

2. The method for enhancing speech in bio-radar based on variational mode decomposition according to claim 1, characterized in that: The acquisition of the 94GHz asymmetric antenna radar voice signal was achieved by sampling the bio-radar voice using an 8-channel Powerlab physiological recorder at a sampling rate of 10kHz.

3. The method for enhancing speech in bio-radar based on variational mode decomposition according to claim 1, characterized in that: The compensation factor has a value of 0.

001.

4. The method for enhancing speech in bio-radar based on variational mode decomposition according to claim 2, characterized in that: The reconstructed enhanced radar speech is the enhanced speech obtained by reconstructing the 94GHz radar speech after denoising using an improved threshold strategy. The specific reconstruction is calculated using the following formula: Wherein, IMF represents the K modes of VMD decomposition. a The number of useful modes is determined to calculate the Pearson coefficient.

5. A bio-radar speech enhancement system based on variational mode decomposition, based on the bio-radar speech enhancement method based on variational mode decomposition as described in any one of claims 1-4, characterized in that, Includes the following steps: The acquisition module is used to acquire the bio-radar speech and obtain the original noisy radar speech signal. The decomposition module is used to perform empirical mode decomposition on the original noisy radar speech signal to obtain intrinsic mode function features, determine the number of variational mode decomposition layers based on the intrinsic mode function features, perform variational mode decomposition (VMD) on the original noisy radar speech signal according to the number of decomposition layers to obtain intrinsic mode decompositions of each order, and solve for Pearson coefficients on each order of intrinsic mode decomposition and the original radar speech signal to determine useful decomposition and reconstruction modes. The reconstruction module is used to denoise the useful modalities using an improved threshold, and reconstruct the enhanced radar speech.

Images