A method for extracting weak sound features from moving sound sources

Abstract

The present invention discloses a method for extracting weak sound features from a moving sound source. The method comprises the following steps: step 1, processing a time domain signal of a moving sound source received by a receiver to obtain energy time history data of each frequency / frequency band; step 2, selecting one frequency / frequency band, using a filtering algorithm to remove pulse interference from the energy time history data of the frequency / frequency band, and performing a sliding average process; step 3, performing an N-order Fourier series decomposition on the energy time history data of the frequency / frequency band; step 4, retaining the first M orders of the Fourier series amplitude, setting the amplitudes of the remaining orders to zero, and then performing an N-order inverse Fourier transform; and step 5, removing the fixed background noise component according to a criterion for removing the fixed background noise component from the received signal, and obtaining the maximum signal energy at the positive horizontal moment of the frequency / frequency band. The present invention can remove the fixed background noise component and pulse interference from the received signal, improve the signal-to-noise ratio of the received signal, and realize weak sound feature extraction from the moving sound source.

Description

Technical Field

[0001] The present invention relates to signal processing technology, and in particular to a method for extracting weak sound features of a moving sound source. Background Art

[0002] When measuring the acoustic energy of a moving sound source in an environment containing interference and background noise, the signal received by the receiver contains the moving sound source signal, background noise, and impulse interference. When the background noise level is high, the moving sound source signal is often drowned out by the background noise, resulting in a low signal-to-noise ratio (SNR) of the received signal, making it impossible to accurately extract the weak sound characteristics of the moving sound source and accurately assess its level. Summary of the Invention

[0003] Purpose of the invention: In order to overcome the deficiencies in the prior art, the present invention provides a method for extracting weak sound features of a moving sound source. Starting from the characteristic that the energy of the moving sound source signal at the receiver end changes with time due to the change in relative position between the moving sound source and the receiver from far to near and from near to far during the measurement process, a method for removing fixed components of background noise and random interference is established through Fourier series decomposition and inverse Fourier transform methods, as well as based on the criterion for removing fixed components of background noise from the received signal, thereby improving the signal-to-noise ratio of the received signal and realizing the extraction of weak sound features of the moving sound source.

[0004] To achieve the above object, the technical solution adopted by the present invention is:

[0005] A method for extracting weak sound features of a moving sound source comprises the following steps:

[0006] Step 1: For the time domain signal of the motion sound source received by the receiver, data of a certain length is selected with the positive and negative moments as the center, and the time domain signal of the motion sound source received by the receiver is segmented according to equal time intervals, equal time lengths and fixed aliasing rates. The power spectrum of each frequency / frequency band of each time domain signal is calculated within a certain frequency range, and the power spectrum calculation results of the same frequency / frequency band are sorted in chronological order to obtain the energy time history data of each frequency / frequency band.

[0007] Step 2: For the energy time history data of each frequency / frequency band described in step 1, select one frequency / frequency band, use a filtering algorithm to remove the pulse interference of the energy time history data of the frequency / frequency band, and perform sliding average processing to obtain the processed energy time history data of the frequency / frequency band;

[0008] Step 3: Perform the frequency / band energy time history data processed in step 2. Fourier series decomposition;

[0009] Step 4: Step 3 The amplitude of the Fourier series obtained by decomposing the Fourier series retains the The remaining order amplitudes are set to zero and then the N-order inverse Fourier transform is performed to obtain the array ;

[0010] Step 5: Based on the criterion for removing the fixed component of background noise from the received signal, the array described in step 4 is Remove the fixed component of background noise and obtain the maximum signal energy at the positive and negative moments of the frequency / band The criterion for removing the fixed component of background noise from the received signal is as follows:

[0011] Set the threshold to , the unit is , substitute into the array The maximum value and minimum value , judge whether formula (1) is true:

[0012] (1)

[0013] If (1) holds true, The maximum value of the signal energy at the positive and negative moments assigned to the frequency / band :

[0014] ;

[0015] Otherwise, from the array Subtract a fixed minimum value , substitute into the array The maximum value and minimum value , judge whether formula (2) is true:

[0016] (2)

[0017] If (2) holds true, The maximum value of the signal energy at the positive and negative moments assigned to the frequency / band :

[0018] ;

[0019] Otherwise, continue from the array Subtract a fixed minimum value , and so on, until we get the equation (3) that holds true value:

[0020] (3)

[0021] Will The maximum value of the signal energy at the positive and negative moments assigned to the frequency / band :

[0022] ;

[0023] In order to prevent the situation where the number of calculations is too large or the convergence does not occur, it is necessary to set a reasonable fixed minimum value based on the average energy of the received signal during the processing period. And the maximum number of calculations If the maximum number of calculations is exceeded When (3) still does not hold, the array described in step 4 is The maximum value of the signal energy at the frequency / band is assigned to the maximum value of the signal energy at the positive moment ,Right now ;

[0024] Repeat steps 2 to 5 for each frequency / frequency band until the maximum signal energy of all frequencies / frequency bands at the positive and negative moments is obtained. The total energy of the sound energy of the moving sound source received within the certain frequency range can be further calculated by energy summation.

[0025] Preferably, the time domain signal of the moving sound source received by the receiver in step 1 is composed of the superposition of the sound source signal, background noise and pulse interference; wherein, the sound source signal is a smooth, slowly changing function with a maximum value near the center of the data, tends to the edge of the data, and the signal attenuates from the maximum value by no less than a certain order of magnitude; the main component of the background noise is a Gaussian random process containing a fixed component; the pulse interference can be a random pulse interference of a single data point, or a random pulse interference composed of multiple continuous data points; the sound source signal can be a certain order of magnitude lower than the background noise; the positive horizontal moment in step 1 is the moment when the moving sound source is closest to the receiver.

[0026] Preferably, the certain frequency range in step 1 is determined based on the moving sound source and the receiver, the frequency is a line spectrum frequency, and the frequency band is a 1 / 3 octave band or a 1 octave band.

[0027] Preferably, in step 2, one of the frequencies / frequency bands is selected, a filtering algorithm is used to remove the pulse interference of the energy time history data of the frequency / frequency band, and a sliding average process is performed. The specific steps are as follows:

[0028] Step S2-1: removing single-point random pulse interference from the energy time history data of the frequency / frequency band;

[0029] Step S2-2: removing random pulse interference consisting of two consecutive data points in the energy time history data of the frequency / frequency band;

[0030] Step S2-3: Perform sliding average processing on the energy time history data of the frequency / frequency band after removing the pulse interference.

[0031] Preferably, the step 4 The order refers to the order that contains the main energy of the frequency / frequency band.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] The present invention starts from the characteristic that the energy of the received signal of a moving sound source changes with time, adopts a filtering algorithm, Fourier series decomposition and inverse Fourier transform method, and based on the criterion of removing the fixed component of background noise from the received signal, removes the fixed component of background noise and pulse interference from the received signal with a low signal-to-noise ratio, improves the signal-to-noise ratio of the received signal, and realizes the extraction of weak sound features of the moving sound source. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flow chart of a method for extracting weak sound features of a moving sound source according to the present invention;

[0035] Figure 2 A comparison chart of signals before and after removing pulse interference using a filtering algorithm and sliding average according to an embodiment of the present invention;

[0036] Figure 3 The energy time history curve of the motion sound source signal provided by the embodiment of the present invention;

[0037] Figure 4 Background noise curve provided by the embodiment of the present invention;

[0038] Figure 5 The energy time history curve of the received signal mixed with the moving sound source signal, background noise and impulse interference provided by the embodiment of the present invention;

[0039] Figure 6 The embodiment of the present invention provides an energy time history curve of a moving sound source signal obtained by removing background noise and pulse interference from a received signal. DETAILED DESCRIPTION

[0040] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the embodiments described are only a part of the embodiments of the present invention, rather than all the embodiments. It should be understood that these examples are only used to illustrate the present invention and are not used to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0041] like Figure 1 As shown, the embodiment of the present invention discloses a method for extracting weak sound features of a moving sound source, comprising the following steps:

[0042] Step 1: Process the time domain signal of the motion sound source received by the receiver to obtain the energy time history data of each frequency / frequency band;

[0043] Step 2: Select one of the frequencies / frequency bands, use a filtering algorithm to remove the pulse interference of the energy time history data of the frequency / frequency band, and perform sliding average processing;

[0044] Step 3: Perform N-order Fourier series decomposition on the frequency / band energy time history data;

[0045] Step 4: retain the first M orders of the Fourier series amplitude, set the amplitudes of the remaining orders to zero, and then perform an N-order inverse Fourier transform;

[0046] Step 5: Based on the criterion for removing the fixed component of background noise from the received signal, the fixed component of background noise is removed to obtain the maximum value of the signal energy at the positive horizontal moment of the frequency / frequency band.

[0047] The above steps are further described below in conjunction with specific embodiments.

[0048] In one embodiment, in step 1, the time domain signal of the moving sound source received by the receiver is centered at the positive and horizontal moments, and data of 263s in length is selected. The selected data is segmented according to a time interval of 1s, a data segment time length of 8s, and an aliasing rate of 87.5% to obtain 256 data segments. The power spectrum of each 1 / 3 octave frequency band of each signal segment is calculated within the frequency range of interest, and the power spectrum calculation results of the same 1 / 3 octave frequency band are sorted in chronological order to obtain energy time history data of 256 points in each 1 / 3 octave frequency band.

[0049] In one embodiment, in step 1, a time domain signal of a moving sound source received by a receiver is selected, with the positive and negative moments as the center, for a period of time, and the selected data is segmented according to equal time intervals, equal time lengths, and a fixed aliasing rate. The power spectrum of each line spectrum frequency of each segment of the signal is calculated within the frequency range of interest, and the power spectrum calculation results of the same line spectrum frequency are sorted in chronological order to obtain the energy time history data of each line spectrum frequency.

[0050] In one embodiment, one of the frequencies / frequency bands is selected in step 2, and the energy time history data of the frequency / frequency band is expressed as , a filtering algorithm is used to remove the pulse interference of the frequency / frequency band energy time history data and perform sliding average processing, as described in steps S2-1, S2-2 and S2-3.

[0051] Step S2-1: Remove single-point random pulse interference from the energy time history data of the frequency / frequency band. The specific method is as follows:

[0052] Analyze every 3 consecutive data If the value of If formula (4) is satisfied, no changes will be made to the data. Representation data A collection of .

[0053] (4)

[0054] if , then the data Replace with .

[0055] if , then the data Replace with .

[0056] In the compared sequences In, when When , only the numbers in the original data are used.

[0057] Step S2-2: remove the random pulse interference consisting of two consecutive data points in the energy time history data of the frequency / frequency band. The specific method is as follows:

[0058] Analyze every 6 consecutive data Value:

[0059] First, use inequalities (5) to (8) to test the random pulse interference composed of two consecutive data points. If all four inequalities (5) to (8) are true, then Assigned to ,Right now ; If any of the inequalities (5) to (8) does not hold, no substitution is performed on the data. Representation data A collection of Representation data A collection of .

[0060] (5)

[0061] (6)

[0062] (7)

[0063] (8)

[0064] Then, use inequalities (9) to (12) to continue testing the random pulse interference composed of two consecutive data points. If all four inequalities (9) to (12) are true, then Assigned to ,Right now ; If any of the inequalities (9) to (12) does not hold, no substitution is performed in the data.

[0065] (9)

[0066] (10)

[0067] (11)

[0068] (12)

[0069] In the compared sequences In, when When , only the numbers in the original data are used.

[0070] Step S2-3: performing sliding average processing on the energy time history data of the frequency / frequency band after removing the pulse interference, specifically performing sliding average processing on three consecutive points.

[0071] In this embodiment, the energy time history data of the 1 / 3 octave frequency band with a center frequency of 200 Hz is generated by simulation calculation. The filtering algorithm is first used to remove the single-point pulse interference of the energy time history data of the received signal in this frequency band, and then the pulse interference composed of two consecutive points is removed. The three-point sliding average processing is performed. The processing results are shown as follows: Figure 2 As shown. Figure 2 It can be seen that the use of filtering algorithm and sliding average processing can obtain the energy time history curve of the received signal in the 1 / 3 octave frequency band with smaller fluctuations.

[0072] In one embodiment, simulation calculation is used to obtain the weak sound feature extraction result of the moving sound source. Figure 3 As shown in the figure, the energy time history data of the received motion sound source signal with a 1 / 3 octave frequency band with a center frequency of 1kHz is a smooth, slowly changing function with a maximum value of 80dB near the center of the data, and the signal attenuates 6dB from the maximum value toward the edge of the data. Figure 4 As shown, the background noise is set to be a Gaussian random process with a mean of 86dB and a standard deviation of 0.45dB, and is distributed with random pulse interference. Figure 5 As shown in the figure, the energy time history of the 1 / 3 octave band of the received signal is composed of the superposition of the moving sound source signal, background noise and pulse interference. Figure 5 It can be seen that the moving sound source signal is submerged in the background noise and the signal-to-noise ratio of the received signal is low. In step 3, the 1 / 3 octave frequency band energy time history data is decomposed into 256-order Fourier series. In step 4, the first 8 orders of the Fourier series amplitude are retained, and the remaining order amplitudes are set to zero and then a 256-order inverse Fourier transform is performed. In step 5, the threshold is set. , set the maximum number of calculations , set a fixed minimum value ,in Represents an array The average value of . Figure 6 As shown in the figure, after removing the fixed component of background noise and pulse interference, the maximum value of the signal energy at the time of the positive horizontal moment of the frequency band is 79.8dB, which is close to the maximum value of the signal energy at the time of the positive horizontal moment of the frequency band set at 80dB. Figure 6 It can be seen from the simulation calculation results that when the energy of the motion sound source signal is 6 dB lower than the background noise, the motion sound source signal can be effectively extracted from the received signal using the method of the present invention.

[0073] It should be understood that the above-described embodiments of the present invention are intended only to illustrate or explain the principles of the present invention and do not constitute limitations of the present invention. Therefore, any modifications, equivalent substitutions, improvements, etc. made without departing from the spirit and scope of the present invention should be included within the scope of protection of the present invention. In addition, the appended claims are intended to cover all variations and modifications that fall within the scope and metes and bounds of the appended claims, or equivalents of such scope and metes and bounds.

Claims

1. A method for extracting weak sound features of a moving sound source, characterized in that: The following steps are involved: Step 1: For the time domain signal of the motion sound source received by the receiver, data of a certain length is selected with the positive and negative moments as the center, and the received time domain signal of the motion sound source is segmented according to equal time intervals, equal time lengths and fixed aliasing rates. The power spectrum of each frequency / frequency band of each time domain signal is calculated within a certain frequency range, and the power spectrum calculation results of the same frequency / frequency band are sorted in chronological order to obtain the energy time history data of each frequency / frequency band. Step 2: For the energy time history data of each frequency / frequency band described in step 1, select one frequency / frequency band, use a filtering algorithm to remove the pulse interference of the energy time history data of the frequency / frequency band, and perform sliding average processing to obtain the processed energy time history data of the frequency / frequency band; Step 3, performing N-order Fourier series decomposition on the frequency / frequency band energy time history data after the processing in step 2; Step 4: For the Fourier series amplitudes obtained by the N-order Fourier series decomposition in step 3, retain the first M orders, set the amplitudes of the remaining orders to zero, and then perform an N-order inverse Fourier transform to obtain the array {x n ]}n=1,2,…,N; Step 5: Based on the criterion for removing the fixed component of background noise from the received signal, the array {x n }Remove the fixed component of background noise and obtain the maximum value S of the signal energy at the positive and negative moments of the frequency / band max The criterion for removing the fixed component of background noise from the received signal is as follows: Set the threshold to T, in dB, and substitute it into the array {x n The maximum value max{x} n } and the minimum value min{x n }, judge whether formula (1) is true: If (1) holds, max{x n The maximum value S of the signal energy at the moment of the horizontal stroke assigned to the frequency / band max : S max =max{x n } Otherwise, from the array {x n }, subtract a fixed small value Δx and substitute it into the array {x n The maximum value of max{x -Δx} n -Δx} and the minimum value min{x n -Δx}, judge whether formula (2) holds true: If (2) holds, max{x n -Δx} assigns the maximum value S of the signal energy at the positive moment of the frequency / band max : S max =max{x n -Δx} Otherwise, continue from the array {x n -Δx}, subtract a fixed small value Δx, and so on, until the K value that makes equation (3) valid is obtained: Max{x n -KΔx} assigns the maximum value S of the signal energy at the positive moment of the frequency / band max : S max =max{x n -KΔx} In order to prevent the situation where the number of calculations is too large or the convergence does not occur, it is necessary to set a reasonable fixed minimum value Δx and the maximum number of calculations L according to the average energy of the received signal during the processing period. If the maximum number of calculations L is exceeded and the formula (3) is still not satisfied, the array {x n }, n=1,2,..., the maximum value of N is assigned to the maximum value S of the signal energy at the positive horizontal moment of the frequency / band max , that is, S max =max{x n }; Repeat steps 2 to 5 for each frequency / frequency band until the maximum signal energy of all frequencies / frequency bands at the positive and negative moments is obtained. The total energy of the sound radiated by the moving sound source received within the certain frequency range can be further calculated by energy summation.

2. The method for extracting weak sound features of a moving sound source according to claim 1, wherein: The time domain signal of the moving sound source received by the receiver in step 1 is composed of the superposition of the sound source signal, background noise and pulse interference; wherein, the sound source signal is a smooth, slowly changing function with a maximum value near the center of the data, tends to the edge of the data, and the signal attenuates from the maximum value by no less than a certain order of magnitude; the main component of the background noise is a Gaussian random process with a fixed component; the pulse interference is a random pulse interference of a single data point, or a random pulse interference composed of multiple continuous data points; the sound source signal is a certain order of magnitude lower than the background noise; the positive horizontal moment in step 1 is the moment when the moving sound source is closest to the receiver.

3. The method for extracting weak sound features of a moving sound source according to claim 1, wherein: The certain frequency range in step 1 is determined based on the moving sound source and the receiver, the frequency is a line spectrum frequency, and the frequency band is a 1 / 3 octave band or a 1 octave band.

4. The method for extracting weak sound features of a moving sound source according to claim 1, wherein: In step 2, one of the frequencies / frequency bands is selected, a filtering algorithm is used to remove the pulse interference of the energy time history data of the frequency / frequency band, and a sliding average process is performed. The specific steps are as follows: Step S2-1: removing single-point random pulse interference from the energy time history data of the frequency / frequency band; Step S2-2: removing random pulse interference consisting of two consecutive data points in the energy time history data of the frequency / frequency band; Step S2-3: Perform sliding average processing on the energy time history data of the frequency / frequency band after removing the pulse interference.

5. The method for extracting weak sound features of a moving sound source according to claim 1, wherein: The first M orders in step 4 refer to the orders containing the main energy of the frequency / frequency band.

Images