Industrial noise multi-target orientation identification method and system based on sound source separation identification

Abstract

The invention relates to the technical field of sensors, in particular to an industrial noise multi-target orientation recognition method and system based on sound source separation recognition, and the method comprises the steps: setting a sensor array, collecting a multi-target industrial noise audio, recording a sampling frequency, executing data standardization, obtaining a calibration audio, and determining a target sampling frequency; the method comprises the following steps: performing sampling frequency conversion based on a target sampling frequency to obtain a conversion signal, calculating audio energy of the conversion signal, performing normalization to obtain a normalized audio, calculating an amplitude spectrum and a phase spectrum, generating a preliminary mask, calculating time domain loss and frequency domain loss, calculating an optimized mask, calculating a reconstructed signal by using the optimized mask, and calculating a sensing time difference. And calculating a horizontal angle and a vertical angle by using the sensing time difference, calculating a sound source distance, and calculating a spatial position based on the horizontal angle, the vertical angle and the sound source distance to complete orientation recognition. According to the invention, the accuracy of multi-target orientation identification can be improved.

Description

Technical Field

[0001] The present invention relates to the field of sensor technology, and in particular to a method and system for industrial noise multi-target orientation recognition based on sound source separation and recognition. Background Art

[0002] With the rapid development of sensor technology and artificial intelligence algorithms, multi-target audio processing and sound source localization technology are widely used in autonomous driving, industrial noise, conference systems and other fields. How to efficiently process multi-target audio, accurately extract key sound signals, and achieve direction recognition has become an important research direction for technological development.

[0003] Traditional sound source localization technology mainly relies on simple acoustic models and low-complexity algorithms. It is difficult to achieve stable audio processing and high-precision spatial positioning in complex scenarios such as multi-source interference or echo environments. In addition, existing methods lack system optimization in the processing of audio amplitude spectrum and phase spectrum, which easily leads to inaccurate sound source location recognition or excessive delay. Therefore, how to improve the accuracy of multi-target location recognition in a complex sound field environment is a difficult problem that current technology urgently needs to break through. Summary of the invention

[0004] The present invention provides a method and system for industrial noise multi-target position recognition based on sound source separation and recognition, the main purpose of which is to improve the accuracy of multi-target position recognition.

[0005] To achieve the above object, the present invention provides a method for multi-target azimuth identification of industrial noise based on sound source separation and identification, comprising:

[0006] Setting a sensor array, collecting multi-target industrial noise audio based on the sensor array and recording the sampling frequency, performing data standardization based on the multi-target industrial noise audio, and obtaining a calibrated audio;

[0007] Determine a target sampling frequency, perform sampling frequency matching based on the sampling frequency to obtain matching data, and perform sampling frequency conversion based on the matching data, the calibration audio and the target sampling frequency to obtain a conversion signal;

[0008] Calculate the audio energy of the converted signal and normalize the audio energy to obtain normalized audio;

[0009] Calculating an amplitude spectrum and a phase spectrum based on the normalized audio, and generating a preliminary mask based on the amplitude spectrum, the phase spectrum and a preset deep neural network, wherein the deep neural network is a neural network that combines a convolutional neural network and a fully connected network;

[0010] Calculate the time domain loss and the frequency domain loss, calculate the optimized mask based on the time domain loss, the frequency domain loss and the preliminary mask, use the optimized mask to perform sound source separation calculation, and obtain the reconstructed signal;

[0011] The sensing time difference is calculated based on the reconstructed signal, the horizontal angle and the vertical angle are calculated using the sensing time difference, the sound source distance is calculated, and the spatial position is calculated based on the horizontal angle, the vertical angle and the sound source distance to complete the azimuth recognition.

[0012] Optionally, the determining a target sampling frequency, performing sampling frequency matching based on the sampling frequency, and obtaining matching data includes:

[0013] Based on the sampling frequency integration, a sampling frequency set is obtained, a maximum frequency in the sampling frequency set is screened, and the maximum frequency is determined as a target sampling frequency;

[0014] Setting a matching function, extracting sampling frequencies in sequence based on the sampling frequency set, and inputting the extracted sampling frequencies and the target sampling frequencies into the matching function to obtain a function output;

[0015] Matching is performed based on the function output to obtain matching data.

[0016] Optionally, the setting of the matching function includes:

[0017] Set a confirmation flag and an error flag, and set a matching function based on the confirmation flag, the error flag, the sampling frequency, and the target sampling frequency:

[0018]

[0019] Among them, E refers to the matching function, T refers to the confirmation flag, F refers to the error flag, q1 refers to the sampling frequency, and q2 refers to the target sampling frequency.

[0020] Optionally, performing sampling frequency conversion based on the matching data, the calibration audio and the target sampling frequency to obtain a conversion signal includes:

[0021] A preliminary sampling time point is obtained by uniform sampling based on the sampling frequency, and a preliminary index and a target index are set. The target sampling time point is calculated according to the preliminary sampling time point, the preliminary index and the target index:

[0022]

[0023] Among them, u z Refers to the target sampling time point when the target index is z, where z refers to the target index and t i Refers to the initial sampling time point when the initial index is i, where i refers to the initial index;

[0024] Define the time window and obtain the frequency components, calculate the frequency-weighted average of the frequency components based on the time window, set the data weight and weight coefficient, and build a dynamic weighting function based on the frequency-weighted average, data weight and weight coefficient:

[0025] R = a + s × sin (2πft i )

[0026] Among them, R refers to the dynamic weighting function, a refers to the data weight, s refers to the weight coefficient, sin refers to the sine function, π refers to pi, and f refers to the frequency-weighted average;

[0027] A conversion formula is constructed based on the dynamic weighting function, the preliminary sampling time point and the target sampling time point, and conversion is performed based on the conversion formula, matching data and calibration audio to obtain a conversion signal.

[0028] Optionally, constructing a conversion formula based on the dynamic weighting function, the preliminary sampling time point and the target sampling time point includes:

[0029] Based on the preliminary sampling time point and the calibration audio, numerical values ​​are collected to obtain the calibration signal value, and a conversion formula is constructed according to the calibration signal value, the dynamic weighting function, the preliminary sampling time point and the target sampling time point:

[0030]

[0031] Among them, T(u z ) refers to the target sampling time point u z The conversion formula for sampling frequency conversion, T(t i ) refers to the initial sampling time point t i The calibration signal value at time T(t i+1 ) refers to the initial sampling time point t i+1 The calibration signal value at .

[0032] Optionally, the calculating audio energy of the conversion signal and normalizing the audio energy to obtain normalized audio includes:

[0033] Get the audio amplitude of the converted signal and the total number of samples at the target sampling time point, and calculate the audio energy based on the audio amplitude and the total number of samples:

[0034]

[0035] Among them, S refers to the audio energy, N refers to the total number of samples, and y z Refers to the audio amplitude of the converted signal at the target sampling time point corresponding to the zth target index;

[0036] The normalized audio amplitude is calculated using the following formula:

[0037]

[0038] Among them, D z Refers to the normalized audio amplitude at the zth target sampling time point;

[0039] The normalized audio amplitude is subjected to data processing to obtain a normalized audio.

[0040] Optionally, the calculating an amplitude spectrum and a phase spectrum based on the normalized audio, and generating a preliminary mask based on the amplitude spectrum, the phase spectrum and a preset deep neural network, comprises:

[0041] Set the window function, time window, window length, window index and complex unit, and calculate the signal complex representation data based on the target sampling frequency, window function, time window, window length, window index and complex unit:

[0042]

[0043] Wherein, F(q2, d) refers to the complex number representation data of the normalized audio signal at the target sampling frequency q2 and time window d, H refers to the window length, refers to the window index, θ refers to the window function, e refers to the Euler number, and j refers to the complex unit;

[0044] The amplitude spectrum is calculated using the complex representation data of the signal:

[0045]

[0046] Among them, |G| refers to the amplitude spectrum, Re refers to the real part, and Im refers to the imaginary part;

[0047] The phase spectrum is calculated using the complex representation data of the signal:

[0048]

[0049] Where L refers to the phase spectrum, arctan refers to the inverse tangent function;

[0050] An activation function, a first weight matrix, a second weight matrix, and an adjustment item are set, and a preliminary mask is generated based on the amplitude spectrum, the phase spectrum, the activation function, the first weight matrix, the second weight matrix, and the adjustment item:

[0051] M=τ(ε1×|G|)+ε2×L+p

[0052] Among them, M refers to the preliminary mask, τ refers to the activation function, ε1 refers to the first weight matrix, ε2 refers to the second weight matrix, and p refers to the adjustment term.

[0053] Optionally, the calculating time domain loss and frequency domain loss, calculating an optimized mask based on the time domain loss, the frequency domain loss and the preliminary mask, performing sound source separation calculation using the optimized mask, and obtaining a reconstructed signal includes:

[0054] Obtain a transformation function, and calculate a preliminary signal based on the preliminary mask, the amplitude spectrum, the phase spectrum and the transformation function:

[0055] δ1=γ×(M×|G|×e j×L )

[0056] Among them, δ1 refers to the preliminary signal, and γ refers to the transformation function;

[0057] Obtain a target signal, and calculate a time domain loss based on the preliminary signal and the target signal:

[0058]

[0059] Among them, V1 refers to the time domain loss, δ 2z refers to the target signal at the zth target sampling time point, δ 1z Refers to the preliminary signal at the zth target sampling time point;

[0060] The frequency domain loss is calculated based on the amplitude spectrum and the phase spectrum, and the learning parameters are set. The optimization mask is calculated using the preliminary mask, the time domain loss, the frequency domain loss and the learning parameters:

[0061]

[0062] Among them, M2 refers to the optimization mask, ∈ refers to the learning parameter, λ refers to the partial derivative, and V2 refers to the frequency domain loss;

[0063] The reconstructed signal is calculated based on the transformation function, the optimized mask, the amplitude spectrum and the phase spectrum:

[0064] δ2=γ×(M2×|G|×e j×L )

[0065] Here, δ2 refers to the reconstructed signal.

[0066] Optionally, the calculating the spatial position based on the horizontal angle, the vertical angle and the sound source distance includes:

[0067] Obtain the center position, determine the center position as the origin, and establish an orientation coordinate system based on the origin;

[0068] Calculate the spatial position based on the azimuth coordinate system, horizontal angle, vertical angle and sound source distance:

[0069] x′=r′×cosφ×sinζ

[0070] y′=r′×cosφ×cosζ

[0071] z′=r′×sinφ

[0072] s′=(x′,y′,z′)

[0073] Among them, x′ refers to the x-axis coordinate, r′ refers to the distance of the sound source, cos refers to the cosine function, φ refers to the vertical angle, ζ refers to the horizontal angle, y′ refers to the y-axis coordinate, z′ refers to the z-axis coordinate, and S′ refers to the spatial position.

[0074] To achieve the above object, the present invention also provides an industrial noise multi-target orientation recognition system based on sound source separation and recognition, comprising:

[0075] A conversion signal calculation module is used to set a sensor array, collect multi-target industrial noise audio based on the sensor array and record the sampling frequency, perform data standardization based on the multi-target industrial noise audio to obtain calibration audio; determine the target sampling frequency, perform sampling frequency matching based on the sampling frequency to obtain matching data, perform sampling frequency conversion based on the matching data, the calibration audio and the target sampling frequency to obtain a conversion signal;

[0076] A preliminary mask calculation module is used to calculate the audio energy of the conversion signal and normalize the audio energy to obtain normalized audio; calculate the amplitude spectrum and phase spectrum based on the normalized audio, and generate a preliminary mask based on the amplitude spectrum, phase spectrum and a preset deep neural network, wherein the deep neural network is a neural network that combines a convolutional neural network and a fully connected network;

[0077] A reconstruction signal calculation module is used to calculate the time domain loss and the frequency domain loss, calculate the optimized mask based on the time domain loss, the frequency domain loss and the preliminary mask, and calculate the reconstructed signal using the optimized mask;

[0078] The spatial position calculation module is used to calculate the sensing time difference based on the reconstructed signal, calculate the horizontal angle and the vertical angle using the sensing time difference, calculate the sound source distance, calculate the spatial position based on the horizontal angle, the vertical angle and the sound source distance, and complete the orientation recognition.

[0079] In order to solve the above problem, the present invention further provides an electronic device, the electronic device comprising:

[0080] A memory storing at least one instruction;

[0081] The processor executes the instructions stored in the memory to implement the above-mentioned industrial noise multi-target orientation recognition method based on sound source separation and recognition.

[0082] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored. The at least one instruction is executed by a processor in an electronic device to implement the above-mentioned industrial noise multi-target orientation identification method based on sound source separation and identification.

[0083] In order to solve the problems described in the background technology, the present invention realizes multi-target azimuth recognition of industrial noise by collecting multi-target industrial noise audio, recording sampling frequency, constructing a comprehensive process of multiple steps such as sampling frequency conversion, audio energy normalization, generating preliminary mask, calculating reconstruction signal and calculating spatial position; firstly, a sensor array is set up and multi-target industrial noise audio is collected. Through the sensor array, industrial noise information from multiple directions can be captured, laying the foundation for subsequent sound source positioning. The design of the array makes the acquired signal contain time and space information, providing necessary data for azimuth and distance calculation, collecting multi-target mixed audio, being able to process multiple noise sources at one time, improving The practicality and efficiency of the system; secondly, record the sampling frequency and perform standardization on the data to obtain the calibrated audio. The standardization process can eliminate the amplitude differences and interference of different signal sources, making subsequent processing more reliable. The calibrated audio unifies the data benchmark, which helps to improve the accuracy of the algorithm and reduce deviations; then, determine the target sampling frequency, match and convert the sampling frequency, obtain the converted signal, calculate the audio energy and normalize it, and obtain the normalized audio. By unifying the sampling frequency, the consistency of data from different channels is ensured, avoiding information loss or processing difficulties caused by sampling rate differences, and the signal quality is retained after the sampling frequency conversion, reducing distortion problems in processing and enhancing the signal. Time-frequency resolution, audio energy normalization balances the difference in signal strength and improves the ability to detect distant sound sources. After energy normalization, the robustness of the processing steps is enhanced and it is more adaptable to different noise environments. Further, the amplitude spectrum and phase spectrum are calculated, and a preliminary mask is generated using a deep neural network. The time domain and frequency domain losses are calculated, the mask is optimized and the sound source is separated to obtain the reconstructed signal. The amplitude spectrum and phase spectrum extract the time-frequency characteristics of the sound signal, making the noise separation more accurate. The generation of the preliminary mask can separate different sound sources in the frequency domain, effectively reducing the interference of reverberation and noise on positioning. The combination of time domain loss and frequency domain loss improves the effect of mask optimization, making the sound source separation more accurate. Subsequent positioning provides high-quality input; finally, the sensing time difference is calculated based on the reconstructed signal, the horizontal angle and the vertical angle are calculated using the sensing time difference, the sound source distance is calculated, and the spatial position is calculated based on the horizontal angle, the vertical angle and the sound source distance. The calculation of the horizontal angle and the vertical angle provides two-dimensional azimuth information for the spatial position of the sound source. The direction of the sound source is determined by the angle, which can help quickly lock the position of the noise source in the industrial environment. The acquisition of the sound source distance can help evaluate the intensity attenuation of the noise source and provide data support for noise control. The distance information can be combined with the angle information to provide a complete three-dimensional spatial position. The accurate determination of the three-dimensional spatial position can accurately describe the specific position of the sound source in the industrial environment. Therefore, the present invention can improve the accuracy of multi-target azimuth recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1A schematic flow chart of a method for multi-target orientation recognition of industrial noise based on sound source separation and recognition provided by an embodiment of the present invention;

[0085] Figure 2 A functional module diagram of an industrial noise multi-target orientation recognition system based on sound source separation and recognition provided by an embodiment of the present invention;

[0086] Figure 3 A schematic diagram of the structure of an electronic device for implementing the method for multi-target orientation recognition of industrial noise based on sound source separation and recognition provided by an embodiment of the present invention.

[0087] Description of reference numerals:

[0088] 1. Electronic device; 10. Processor; 11. Memory; 12. Bus.

[0089] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0090] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0091] The embodiment of the present application provides a method for multi-target position identification of industrial noise based on sound source separation and identification. The execution subject of the method for multi-target position identification of industrial noise based on sound source separation and identification includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the method for multi-target position identification of industrial noise based on sound source separation and identification can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0092] Reference Figure 1 FIG. 1 is a flow chart of a method for industrial noise multi-target position identification based on sound source separation and identification provided by an embodiment of the present invention. In this embodiment, the method for industrial noise multi-target position identification based on sound source separation and identification includes:

[0093] S1. Set up a sensor array, collect multi-target industrial noise audio based on the sensor array and record the sampling frequency, perform data standardization based on the multi-target industrial noise audio, and obtain a calibrated audio.

[0094] It can be explained that a sensor array refers to a device composed of multiple sensors and used to collect sound signals from different directions in space. A sensor refers to a microphone. The sensor array in the present invention is configured as a three-dimensional array. The sensors are arranged in a three-dimensional form and are suitable for locating sound sources in three-dimensional space. Multi-target industrial noise audio refers to sound signals emitted simultaneously by multiple industrial noise sources. These signals are received by the sensor array at the same time. Due to the different geographical locations and corresponding environments of the sound sources, these signals will overlap and interfere with each other in space to form a mixed signal. For example, suppose that in an industrial production environment, there are three devices A, B, and C working simultaneously. When the noise generated by each device is received by the sensor array, the sensor will record the superposition of the sound signals of the three devices. For example, device A emits low-frequency noise, device B emits high-frequency noise, and device C emits medium-frequency noise. The low-frequency noise, high-frequency noise, and medium-frequency noise converge to the sensor array at the same time point to form a multi-target industrial noise audio signal. The sampling frequency refers to the number of times the sensor array samples the sound signal per second. The higher the sampling frequency, the shorter the time interval between the sound signal sampling, and the more accurately the changes in the sound signal can be captured. Calibration audio refers to the audio signal obtained after data standardization of the multi-target industrial noise audio signal.

[0095] It can be understood that performing data standardization on multi-target industrial noise audio to obtain calibration audio means setting the sampling time, calculating the signal mean and signal standard deviation based on the sampling time, and calculating the calibration audio using the signal mean and signal standard deviation:

[0096]

[0097] Among them, Q refers to the calibration audio, W refers to the multi-target industrial noise audio, e refers to the signal mean, and r refers to the signal standard deviation.

[0098] It can be explained that the sampling time refers to the time for sampling the multi-target industrial noise audio, the signal mean refers to the average value of the multi-target industrial noise audio within the sampling time, and the signal standard deviation refers to the standard deviation of the multi-target industrial noise audio within the sampling time.

[0099] S2. Determine a target sampling frequency, perform sampling frequency matching based on the sampling frequency to obtain matching data, and perform sampling frequency conversion based on the matching data, the calibration audio and the target sampling frequency to obtain a conversion signal.

[0100] It can be explained that the target sampling frequency refers to a sampling frequency set in order to unify the sampling frequency. Since different sensors may use different sampling frequencies, the sampling frequencies need to be unified. The target sampling frequency is specifically the largest sampling frequency in the sampling frequency set. The matching data refers to the result obtained after matching the sampling frequency with the calibration audio. The sampling frequency set refers to the set formed by the sampling frequencies. Selecting the highest sampling frequency in the sampling frequency set as the target sampling frequency can avoid information loss. The conversion signal refers to the signal obtained after converting the sampling frequency into the target sampling frequency.

[0101] Exemplarily, assuming that the sampling frequency of sensor A is 44.1 kHz and the sampling frequency of sensor B is 48 kHz, sampling frequency matching is performed based on the sampling frequencies, the target sampling frequency is confirmed to be 48 kHz, and the sampling frequencies of all sensors are adjusted to 48 kHz.

[0102] In detail, the determining of the target sampling frequency, performing sampling frequency matching based on the sampling frequency, and obtaining matching data includes:

[0103] Based on the sampling frequency integration, a sampling frequency set is obtained, a maximum frequency in the sampling frequency set is screened, and the maximum frequency is determined as a target sampling frequency;

[0104] Setting a matching function, extracting sampling frequencies in sequence based on the sampling frequency set, and inputting the extracted sampling frequencies and the target sampling frequencies into the matching function to obtain a function output;

[0105] Matching is performed based on the function output to obtain matching data.

[0106] It can be explained that the maximum frequency refers to the largest sampling frequency in the sampling frequency set, the matching function refers to a function that matches the target sampling frequency with the sampling frequency to determine whether the sampling frequency is the target sampling frequency, and the function output refers to the output obtained after inputting the sampling frequency and the target sampling frequency into the matching function and comparing them based on the matching function. The function output can confirm whether the sampling frequency is the target sampling frequency.

[0107] In detail, the setting matching function includes:

[0108] Set a confirmation flag and an error flag, and set a matching function based on the confirmation flag, the error flag, the sampling frequency, and the target sampling frequency:

[0109]

[0110] Among them, E refers to the matching function, T refers to the confirmation flag, F refers to the error flag, q1 refers to the sampling frequency, and q2 refers to the target sampling frequency.

[0111] It can be explained that the confirmation flag refers to the output when the sampling frequency is the same as the target sampling frequency, and the error flag refers to the output when the sampling frequency is different from the target sampling frequency.

[0112] In detail, the performing sampling frequency conversion based on the matching data, the calibration audio and the target sampling frequency to obtain a conversion signal includes:

[0113] A preliminary sampling time point is obtained by uniform sampling based on the sampling frequency, and a preliminary index and a target index are set. The target sampling time point is calculated according to the preliminary sampling time point, the preliminary index and the target index:

[0114]

[0115] Among them, u z Refers to the target sampling time point when the target index is z, where z refers to the target index and t i Refers to the initial sampling time point when the initial index is i, where i refers to the initial index;

[0116] Define the time window and obtain the frequency components, calculate the frequency-weighted average of the frequency components based on the time window, set the data weight and weight coefficient, and build a dynamic weighting function based on the frequency-weighted average, data weight and weight coefficient:

[0117] R = a + s × sin (2πft i )

[0118] Among them, R refers to the dynamic weighting function, a refers to the data weight, s refers to the weight coefficient, sin refers to the sine function, π refers to pi, and f refers to the frequency-weighted average;

[0119] A conversion formula is constructed based on the dynamic weighting function, the preliminary sampling time point and the target sampling time point, and conversion is performed based on the conversion formula, matching data and calibration audio to obtain a conversion signal.

[0120] Explainable, the initial sampling time point refers to the time point at which sampling is performed at the sampling frequency, and the time points are uniformly distributed. The initial index refers to the parameters of the initial sampling time point, the target index refers to the parameters of the target sampling time point, the target sampling time point refers to the time point at which sampling is performed at the target sampling frequency, and is also uniformly distributed. The time window refers to a time period of the calibration audio signal, and the calibration audio signal refers to the audio signal corresponding to the calibration audio. A calibration audio signal can be regarded as composed of multiple sine waves, each sine wave corresponds to a specific frequency, and the frequency component refers to the frequency of these sine waves and their corresponding amplitudes, which are obtained by Fourier transform. The frequency-weighted average refers to the weighted average of all frequency components, and the data weight refers to a basic value, which is equivalent to the bottom line of the dynamic weighting function. Regardless of the dynamic part s×sin(2πft i) changes, this value will not be weakened or disappear. The weight coefficient refers to a coefficient used for adjustment, which controls the dynamic part s×sin(2πft i ) fluctuation amplitude, the dynamic weighting function refers to the function that dynamically adjusts the conversion result, the conversion result refers to the result output by the conversion formula, and the conversion formula refers to the formula that converts the sampling frequency into the target sampling frequency.

[0121] In detail, the conversion formula is constructed based on the dynamic weighting function, the preliminary sampling time point and the target sampling time point, including:

[0122] Based on the preliminary sampling time point and the calibration audio, numerical values ​​are collected to obtain the calibration signal value, and a conversion formula is constructed according to the calibration signal value, the dynamic weighting function, the preliminary sampling time point and the target sampling time point:

[0123]

[0124] Among them, T(u z ) refers to the target sampling time point u z The conversion formula for sampling frequency conversion, T(t i ) refers to the initial sampling time point t i The calibration signal value at time T(t i+1 ) refers to the initial sampling time point t i+1 The calibration signal value at .

[0125] It can be explained that the calibration signal value refers to the signal value corresponding to the calibration audio signal.

[0126] S3. Calculate the audio energy of the converted signal and normalize the audio energy to obtain normalized audio.

[0127] To explain, audio energy refers to the energy of the converted signal, and normalized audio refers to the signal obtained by adjusting the audio energy of the converted signal to a uniform scale.

[0128] In detail, the step of calculating the audio energy of the conversion signal and normalizing the audio energy to obtain normalized audio includes:

[0129] Get the audio amplitude of the converted signal and the total number of samples at the target sampling time point, and calculate the audio energy based on the audio amplitude and the total number of samples:

[0130]

[0131] Among them, S refers to the audio energy, N refers to the total number of samples, and y z Refers to the audio amplitude of the converted signal at the target sampling time point corresponding to the zth target index;

[0132] The normalized audio amplitude is calculated using the following formula:

[0133]

[0134] Among them, D z Refers to the normalized audio amplitude at the zth target sampling time point;

[0135] The normalized audio amplitude is subjected to data processing to obtain a normalized audio.

[0136] It can be explained that audio amplitude refers to the value of the converted signal at the target sampling time point, indicating the intensity of the vibration. The total number of samples refers to the number of target sampling time points. Normalized audio amplitude refers to the data obtained after normalizing the audio amplitude.

[0137] For example, assuming that the conversion signal is: [0.5, -0.3, 0.8, -0.2] Pa, S = 0.25 + 0.09 + 0.64 + 0.04 = 1.02 Pa, the audio energy is 1.02 Pa, and the audio energy is substituted into the calculation of the normalized audio amplitude: And so on.

[0138] S4. Calculate the amplitude spectrum and phase spectrum based on the normalized audio, and generate a preliminary mask based on the amplitude spectrum, phase spectrum and a preset deep neural network, wherein the deep neural network is a neural network that combines a convolutional neural network and a fully connected network.

[0139] Explainable, the deep neural network uses a combination of standard convolutional neural networks and fully connected networks. The deep neural network includes convolutional layers, batch normalization layers, activation functions, fully connected layers and output layers. The preliminary mask refers to a matrix obtained after processing by the deep neural network, which is used to adjust the energy distribution of the normalized audio signal at different target sampling frequencies and sampling times. It assigns a weight to each frequency component to highlight the focus signal. The normalized audio signal refers to the signal corresponding to the normalized audio, and the focus signal refers to the signal you want to identify. For example, the collected audio signal contains the sounds of multiple targets. The preliminary mask can help highlight the target you want to identify. Suppose there is an audio signal, and there are the sounds of machine A and machine B in the audio signal. Through the preliminary mask, we can highlight the sound of machine A or the sound of machine B.

[0140] For example, assuming that the target sampling frequency is 16kHz, the sampling time is 2s, the size of the amplitude spectrum and the phase spectrum is 257×63, the calculated amplitude spectrum and phase spectrum are input into the deep neural network, and a preliminary mask is output. The size of the preliminary mask is also 257×63. The larger the weight, the more important the corresponding frequency component of the surface, and it will have a higher priority in subsequent signal reconstruction.

[0141] In detail, the calculating the amplitude spectrum and the phase spectrum based on the normalized audio, and generating a preliminary mask based on the amplitude spectrum, the phase spectrum and a preset deep neural network, include:

[0142] Set the window function, time window, window length, window index and complex unit, and calculate the signal complex representation data based on the target sampling frequency, window function, time window, window length, window index and complex unit:

[0143]

[0144] Wherein, F(q2, d) refers to the complex number representation data of the normalized audio signal at the target sampling frequency q2 and time window d, H refers to the window length, refers to the window index, θ refers to the window function, e refers to the Euler number, and j refers to the complex unit;

[0145] The amplitude spectrum is calculated using the complex representation data of the signal:

[0146]

[0147] Among them, |G| refers to the amplitude spectrum, Re refers to the real part, and Im refers to the imaginary part;

[0148] The phase spectrum is calculated using the complex representation data of the signal:

[0149]

[0150] Where L refers to the phase spectrum, arctan refers to the inverse tangent function;

[0151] An activation function, a first weight matrix, a second weight matrix, and an adjustment item are set, and a preliminary mask is generated based on the amplitude spectrum, the phase spectrum, the activation function, the first weight matrix, the second weight matrix, and the adjustment item:

[0152] M=τ(ε1×|G|)+ε2×L+p

[0153] Among them, M refers to the preliminary mask, τ refers to the activation function, ε1 refers to the first weight matrix, ε2 refers to the second weight matrix, and p refers to the adjustment term.

[0154] Explainable, window function refers to the Hamming window, time window refers to a time period for segmenting the normalized audio signal, window length refers to the length of the time window, window index refers to the parameter traversing the window length, and complex unit refers to The signal complex representation data refers to the complex form of the normalized audio at the frequency components and the target sampling time points. The first weight matrix refers to the matrix for adjusting the influence of the amplitude spectrum on the preliminary mask. The second weight matrix refers to the matrix for adjusting the influence of the phase spectrum on the preliminary mask. The adjustment term refers to the input value used to translate the activation function, which is a fixed value.

[0155] S5. Calculate the time domain loss and the frequency domain loss, calculate the optimized mask based on the time domain loss, the frequency domain loss and the preliminary mask, perform sound source separation calculation using the optimized mask, and obtain a reconstructed signal.

[0156] Explainable, time domain loss refers to the difference between the target signal and the preliminary signal in the time domain, frequency domain loss refers to the difference between the target signal and the preliminary signal in the frequency domain, optimized mask refers to the matrix obtained after optimizing and adjusting the preliminary mask, sound source separation calculation refers to sound source separation, separating the sound source whose position you want to confirm, reconstructed signal refers to the time domain signal reconstructed by optimizing mask, the reconstructed signal separates the main sound information from the mixed signal, for example, there are human voice, machine sound, air conditioning sound, the mixed signal is a sound signal of human voice, machine sound, and air conditioning sound mixed together, the main sound information you want to retain is the machine sound, then the reconstructed signal is the signal obtained after separating the machine sound, the target signal refers to an ideal reference signal, this ideal reference signal is a clean signal without noise actually measured, through which the source of the corresponding signal can be accurately identified, for example, a noise-free human voice of Xiao Zhang, through which Xiao Zhang can be accurately identified, the preliminary signal refers to the signal obtained by the preliminary mask, amplitude spectrum and phase spectrum, which can roughly determine the source of the signal.

[0157] In detail, the calculating of the time domain loss and the frequency domain loss, calculating the optimized mask based on the time domain loss, the frequency domain loss and the preliminary mask, and calculating the reconstructed signal using the optimized mask include:

[0158] Obtain a transformation function, and calculate a preliminary signal based on the preliminary mask, the amplitude spectrum, the phase spectrum and the transformation function:

[0159] δ1=γ×(M×|G|×e j×L )

[0160] Among them, δ1 refers to the preliminary signal, and γ refers to the transformation function;

[0161] Obtain a target signal, and calculate a time domain loss based on the preliminary signal and the target signal:

[0162]

[0163] Among them, V1 refers to the time domain loss, δ 2z refers to the target signal at the zth target sampling time point, δ 1z Refers to the preliminary signal at the zth target sampling time point;

[0164] The frequency domain loss is calculated based on the amplitude spectrum and the phase spectrum, and the learning parameters are set. The optimization mask is calculated using the preliminary mask, the time domain loss, the frequency domain loss and the learning parameters:

[0165]

[0166] Among them, M2 refers to the optimization mask, ∈ refers to the learning parameter, λ refers to the partial derivative, and V2 refers to the frequency domain loss;

[0167] The reconstructed signal is calculated based on the transformation function, the optimized mask, the amplitude spectrum and the phase spectrum:

[0168] δ2=γ×(M2×|G|×e j×L )

[0169] Here, δ2 refers to the reconstructed signal.

[0170] Interpretable, the transformation function refers to the inverse short-time Fourier transform function, which is used to convert the frequency domain signal into the time domain, and the learning parameter refers to the learning rate, which is a hyperparameter in the deep neural network.

[0171] S6. Calculate the sensing time difference based on the reconstructed signal, use the sensing time difference to calculate the horizontal angle and the vertical angle, calculate the sound source distance, calculate the spatial position based on the horizontal angle, the vertical angle and the sound source distance, and complete the azimuth recognition.

[0172] It can be explained that the sensing time difference refers to the time difference between the target sound source and the sensor array, the target sound source refers to the sound source corresponding to the reconstructed signal, the sound source distance refers to the distance between the target sound source and the sensor array, and the spatial position refers to the position of the object emitting the target sound source.

[0173] In detail, the calculating the spatial position based on the horizontal angle, the vertical angle and the sound source distance includes:

[0174] Obtain the center position, determine the center position as the origin, and establish an orientation coordinate system based on the origin;

[0175] Calculate the spatial position based on the azimuth coordinate system, horizontal angle, vertical angle and sound source distance:

[0176] x′=r′×cosφ×sinζ

[0177] y′=r′×cosφ×cosζ

[0178] z′=r′×sinφ

[0179] S′=(x′,y′,z′)

[0180] Among them, x′ refers to the x-axis coordinate, r′ refers to the distance of the sound source, cos refers to the cosine function, φ refers to the vertical angle, ζ refers to the horizontal angle, y′ refers to the y-axis coordinate, z′ refers to the z-axis coordinate, and S′ refers to the spatial position.

[0181] It can be explained that the center position refers to the center position of the sensor array, the origin refers to the origin of the azimuth coordinate system, and the azimuth coordinate system refers to a coordinate system for confirming the spatial position, which is a three-dimensional coordinate system.

[0182] In order to solve the problems described in the background technology, the present invention realizes multi-target azimuth recognition of industrial noise by collecting multi-target industrial noise audio, recording sampling frequency, constructing a comprehensive process of multiple steps such as sampling frequency conversion, audio energy normalization, generating preliminary mask, calculating reconstruction signal and calculating spatial position; firstly, a sensor array is set up and multi-target industrial noise audio is collected. Through the sensor array, industrial noise information from multiple directions can be captured, laying the foundation for subsequent sound source positioning. The design of the array makes the acquired signal contain time and space information, providing necessary data for azimuth and distance calculation, collecting multi-target mixed audio, being able to process multiple noise sources at one time, improving The practicality and efficiency of the system; secondly, record the sampling frequency and perform standardization on the data to obtain the calibrated audio. The standardization process can eliminate the amplitude differences and interference of different signal sources, making subsequent processing more reliable. The calibrated audio unifies the data benchmark, which helps to improve the accuracy of the algorithm and reduce deviations; then, determine the target sampling frequency, match and convert the sampling frequency, obtain the converted signal, calculate the audio energy and normalize it, and obtain the normalized audio. By unifying the sampling frequency, the consistency of data from different channels is ensured, avoiding information loss or processing difficulties caused by sampling rate differences, and the signal quality is retained after the sampling frequency conversion, reducing distortion problems in processing and enhancing the signal. Time-frequency resolution, audio energy normalization balances the difference in signal strength and improves the ability to detect distant sound sources. After energy normalization, the robustness of the processing steps is enhanced and it is more adaptable to different noise environments. Further, the amplitude spectrum and phase spectrum are calculated, and a preliminary mask is generated using a deep neural network. The time domain and frequency domain losses are calculated, the mask is optimized and the sound source is separated to obtain the reconstructed signal. The amplitude spectrum and phase spectrum extract the time-frequency characteristics of the sound signal, making the noise separation more accurate. The generation of the preliminary mask can separate different sound sources in the frequency domain, effectively reducing the interference of reverberation and noise on positioning. The combination of time domain loss and frequency domain loss improves the effect of mask optimization, making the sound source separation more accurate. Subsequent positioning provides high-quality input; finally, the sensing time difference is calculated based on the reconstructed signal, the horizontal angle and the vertical angle are calculated using the sensing time difference, the sound source distance is calculated, and the spatial position is calculated based on the horizontal angle, the vertical angle and the sound source distance. The calculation of the horizontal angle and the vertical angle provides two-dimensional azimuth information for the spatial position of the sound source. The direction of the sound source is determined by the angle, which can help quickly lock the position of the noise source in the industrial environment. The acquisition of the sound source distance can help evaluate the intensity attenuation of the noise source and provide data support for noise control. The distance information can be combined with the angle information to provide a complete three-dimensional spatial position. The accurate determination of the three-dimensional spatial position can accurately describe the specific position of the sound source in the industrial environment. Therefore, the present invention can improve the accuracy of multi-target azimuth recognition.

[0183] like Figure 2, which is a functional module diagram of an industrial noise multi-target orientation recognition system based on sound source separation and recognition provided by an embodiment of the present invention.

[0184] The industrial noise multi-target position recognition system 100 based on sound source separation and recognition of the present invention can be installed in an electronic device. According to the functions to be implemented, the industrial noise multi-target position recognition system 100 based on sound source separation and recognition can include a conversion signal calculation module 101, a preliminary mask calculation module 102, a reconstruction signal calculation module 103 and a spatial position calculation module 104. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0185] The conversion signal calculation module 101 is used to set a sensor array, collect multi-target industrial noise audio based on the sensor array and record the sampling frequency, perform data standardization based on the multi-target industrial noise audio to obtain calibration audio; determine the target sampling frequency, perform sampling frequency matching based on the sampling frequency to obtain matching data, perform sampling frequency conversion based on the matching data, the calibration audio and the target sampling frequency to obtain a conversion signal;

[0186] The preliminary mask calculation module 102 is used to calculate the audio energy of the conversion signal and normalize the audio energy to obtain normalized audio; calculate the amplitude spectrum and phase spectrum based on the normalized audio, and generate a preliminary mask based on the amplitude spectrum, phase spectrum and a preset deep neural network, wherein the deep neural network is a neural network that combines a convolutional neural network and a fully connected network;

[0187] The reconstructed signal calculation module 103 is used to calculate the time domain loss and the frequency domain loss, calculate the optimized mask based on the time domain loss, the frequency domain loss and the preliminary mask, and calculate the reconstructed signal using the optimized mask;

[0188] The spatial position calculation module 104 is used to calculate the sensing time difference based on the reconstructed signal, calculate the horizontal angle and the vertical angle using the sensing time difference, calculate the sound source distance, calculate the spatial position based on the horizontal angle, the vertical angle and the sound source distance, and complete the orientation recognition.

[0189] In detail, the modules in the industrial noise multi-target position recognition system 100 based on sound source separation and recognition in the embodiment of the present invention are used in the same manner as described above. Figure 1 The same technical means are used as the method for multi-target orientation recognition of industrial noise based on sound source separation and recognition described in , and can produce the same technical effects, so they will not be repeated here.

[0190] like Figure 3, which is a schematic diagram of the structure of an electronic device for implementing a method for multi-target orientation recognition of industrial noise based on sound source separation and recognition provided by an embodiment of the present invention.

[0191] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a method program for multi-target orientation recognition of industrial noise based on sound source separation and recognition.

[0192] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. The memory 11 may be an internal storage unit of the electronic device 1 in some embodiments, such as a mobile hard disk of the electronic device 1. The memory 11 may also be an external storage device of the electronic device 1 in other embodiments, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Further, the memory 11 also includes an internal storage unit of the electronic device 1 and an external storage device. The memory 11 can not only be used to store application software and various types of data installed in the electronic device 1, such as the code of the industrial noise multi-target orientation recognition method program based on sound source separation and recognition, but also can be used to temporarily store data that has been output or is to be output.

[0193] The processor 10 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes or executes programs or modules stored in the memory 11 (such as a method program for multi-target azimuth identification of industrial noise based on sound source separation and identification, etc.), and calls data stored in the memory 11 to execute various functions of the electronic device 1 and process data.

[0194] The bus 12 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus 12 may be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0195] Figure 3 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 3 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0196] For example, although not shown, the electronic device 1 may also include a power source (such as a battery) for supplying power to each component. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include any components such as one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, etc. The electronic device 1 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0197] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices.

[0198] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), or a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode) touch device. The display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual user interface.

[0199] The program of the method for industrial noise multi-target orientation recognition based on sound source separation and recognition stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When running in the processor 10, the following can be achieved:

[0200] Setting a sensor array, collecting multi-target industrial noise audio based on the sensor array and recording the sampling frequency, performing data standardization based on the multi-target industrial noise audio, and obtaining a calibrated audio;

[0201] Determine a target sampling frequency, perform sampling frequency matching based on the sampling frequency to obtain matching data, and perform sampling frequency conversion based on the matching data, the calibration audio and the target sampling frequency to obtain a conversion signal;

[0202] Calculate the audio energy of the converted signal and normalize the audio energy to obtain normalized audio;

[0203] Calculating an amplitude spectrum and a phase spectrum based on the normalized audio, and generating a preliminary mask based on the amplitude spectrum, the phase spectrum and a preset deep neural network, wherein the deep neural network is a neural network that combines a convolutional neural network and a fully connected network;

[0204] Calculate the time domain loss and the frequency domain loss, calculate the optimized mask based on the time domain loss, the frequency domain loss and the preliminary mask, use the optimized mask to perform sound source separation calculation, and obtain the reconstructed signal;

[0205] The sensing time difference is calculated based on the reconstructed signal, the horizontal angle and the vertical angle are calculated using the sensing time difference, the sound source distance is calculated, and the spatial position is calculated based on the horizontal angle, the vertical angle and the sound source distance to complete the azimuth recognition.

[0206] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figures 1 to 3 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0207] Furthermore, if the module / unit integrated in the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0208] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, the computer program can implement:

[0209] Setting a sensor array, collecting multi-target industrial noise audio based on the sensor array and recording the sampling frequency, performing data standardization based on the multi-target industrial noise audio, and obtaining a calibrated audio;

[0210] Determine a target sampling frequency, perform sampling frequency matching based on the sampling frequency to obtain matching data, and perform sampling frequency conversion based on the matching data, the calibration audio and the target sampling frequency to obtain a conversion signal;

[0211] Calculate the audio energy of the converted signal and normalize the audio energy to obtain normalized audio;

[0212] Calculating an amplitude spectrum and a phase spectrum based on the normalized audio, and generating a preliminary mask based on the amplitude spectrum, the phase spectrum and a preset deep neural network, wherein the deep neural network is a neural network that combines a convolutional neural network and a fully connected network;

[0213] Calculate the time domain loss and the frequency domain loss, calculate the optimized mask based on the time domain loss, the frequency domain loss and the preliminary mask, use the optimized mask to perform sound source separation calculation, and obtain the reconstructed signal;

[0214] The sensing time difference is calculated based on the reconstructed signal, the horizontal angle and the vertical angle are calculated using the sensing time difference, the sound source distance is calculated, and the spatial position is calculated based on the horizontal angle, the vertical angle and the sound source distance to complete the azimuth recognition.

[0215] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, systems and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative, and actual implementation may have other division methods.

[0216] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0217] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0218] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0219] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for multi-target azimuth identification of industrial noise based on sound source separation and identification, characterized in that: The method comprises: Setting a sensor array, collecting multi-target industrial noise audio based on the sensor array and recording the sampling frequency, performing data standardization based on the multi-target industrial noise audio, and obtaining a calibrated audio; Determine a target sampling frequency, perform sampling frequency matching based on the sampling frequency to obtain matching data, and perform sampling frequency conversion based on the matching data, the calibration audio and the target sampling frequency to obtain a conversion signal; Calculate the audio energy of the converted signal and normalize the audio energy to obtain normalized audio; Calculating an amplitude spectrum and a phase spectrum based on the normalized audio, and generating a preliminary mask based on the amplitude spectrum, the phase spectrum and a preset deep neural network, wherein the deep neural network is a neural network that combines a convolutional neural network and a fully connected network; Calculate the time domain loss and the frequency domain loss, calculate the optimized mask based on the time domain loss, the frequency domain loss and the preliminary mask, use the optimized mask to perform sound source separation calculation, and obtain the reconstructed signal; The sensing time difference is calculated based on the reconstructed signal, the horizontal angle and the vertical angle are calculated using the sensing time difference, the sound source distance is calculated, and the spatial position is calculated based on the horizontal angle, the vertical angle and the sound source distance to complete the azimuth recognition.

2. The method for multi-target orientation recognition of industrial noise based on sound source separation and recognition according to claim 1 is characterized in that: The determining of the target sampling frequency, performing sampling frequency matching based on the sampling frequency, and obtaining matching data includes: Based on the sampling frequency integration, a sampling frequency set is obtained, a maximum frequency in the sampling frequency set is screened, and the maximum frequency is determined as a target sampling frequency; Setting a matching function, extracting sampling frequencies in sequence based on the sampling frequency set, and inputting the extracted sampling frequencies and the target sampling frequencies into the matching function to obtain a function output; Matching is performed based on the function output to obtain matching data.

3. The method for multi-target orientation recognition of industrial noise based on sound source separation and recognition as claimed in claim 2, characterized in that: The setting matching function includes: Set a confirmation flag and an error flag, and set a matching function based on the confirmation flag, the error flag, the sampling frequency, and the target sampling frequency: Among them, E refers to the matching function, T refers to the confirmation flag, F refers to the error flag, q1 refers to the sampling frequency, and q2 refers to the target sampling frequency.

4. The method for multi-target orientation recognition of industrial noise based on sound source separation and recognition as claimed in claim 3 is characterized in that: The performing sampling frequency conversion based on the matching data, the calibration audio and the target sampling frequency to obtain a conversion signal includes: A preliminary sampling time point is obtained by uniform sampling based on the sampling frequency, and a preliminary index and a target index are set. The target sampling time point is calculated according to the preliminary sampling time point, the preliminary index and the target index: Among them, u z Refers to the target sampling time point when the target index is z, where z refers to the target index and t i Refers to the initial sampling time point when the initial index is i, where i refers to the initial index; Define the time window and obtain the frequency components, calculate the frequency-weighted average of the frequency components based on the time window, set the data weight and weight coefficient, and build a dynamic weighting function based on the frequency-weighted average, data weight and weight coefficient: R=a+s×sin(2πft i ) Among them, R refers to the dynamic weighting function, a refers to the data weight, s refers to the weight coefficient, sin refers to the sine function, π refers to pi, and f refers to the frequency-weighted average; A conversion formula is constructed based on the dynamic weighting function, the preliminary sampling time point and the target sampling time point, and conversion is performed based on the conversion formula, matching data and calibration audio to obtain a conversion signal.

5. The method for multi-target orientation recognition of industrial noise based on sound source separation and recognition as claimed in claim 4 is characterized in that: The constructing of a conversion formula based on the dynamic weighting function, the preliminary sampling time point and the target sampling time point includes: Based on the preliminary sampling time point and the calibration audio, numerical values ​​are collected to obtain the calibration signal value, and a conversion formula is constructed according to the calibration signal value, the dynamic weighting function, the preliminary sampling time point and the target sampling time point: Among them, T(u z ) refers to the target sampling time point u z The conversion formula for sampling frequency conversion, T(t i ) refers to the initial sampling time point t i The calibration signal value at time T(t i+1 ) refers to the initial sampling time point t i+1 The calibration signal value at .

6. The method for multi-target orientation recognition of industrial noise based on sound source separation and recognition as claimed in claim 1, characterized in that: The step of calculating the audio energy of the converted signal and normalizing the audio energy to obtain normalized audio includes: Get the audio amplitude of the converted signal and the total number of samples at the target sampling time point, and calculate the audio energy based on the audio amplitude and the total number of samples: Among them, S refers to the audio energy, N refers to the total number of samples, z refers to the target index, and y z Refers to the audio amplitude of the converted signal at the target sampling time point corresponding to the zth target index; The normalized audio amplitude is calculated using the following formula: Among them, D z Refers to the normalized audio amplitude at the zth target sampling time point; The normalized audio amplitude is subjected to data processing to obtain a normalized audio.

7. The method for multi-target orientation recognition of industrial noise based on sound source separation and recognition as claimed in claim 6, characterized in that: The calculating the amplitude spectrum and the phase spectrum based on the normalized audio, and generating a preliminary mask based on the amplitude spectrum, the phase spectrum and a preset deep neural network, includes: Set the window function, time window, window length, window index and complex unit, and calculate the signal complex representation data based on the target sampling frequency, window function, time window, window length, window index and complex unit: Wherein, F(q2, d) refers to the complex number representation data of the normalized audio signal at the target sampling frequency q2 and time window d, H refers to the window length, refers to the window index, θ refers to the window function, e refers to the Euler number, and j refers to the complex unit; The amplitude spectrum is calculated using the complex representation data of the signal: Among them, |G| refers to the amplitude spectrum, Re refers to the real part, and Im refers to the imaginary part; The phase spectrum is calculated using the complex representation data of the signal: Where L refers to the phase spectrum, arctan refers to the inverse tangent function; An activation function, a first weight matrix, a second weight matrix, and an adjustment item are set, and a preliminary mask is generated based on the amplitude spectrum, the phase spectrum, the activation function, the first weight matrix, the second weight matrix, and the adjustment item: M=τ(ε1×|G|)+ε2×L+p Among them, M refers to the preliminary mask, τ refers to the activation function, ε1 refers to the first weight matrix, ε2 refers to the second weight matrix, and p refers to the adjustment term.

8. The method for multi-target orientation recognition of industrial noise based on sound source separation and recognition as claimed in claim 7, characterized in that: The calculating of the time domain loss and the frequency domain loss, calculating the optimized mask based on the time domain loss, the frequency domain loss and the preliminary mask, performing the sound source separation calculation using the optimized mask, and obtaining the reconstructed signal includes: Obtain a transformation function, and calculate a preliminary signal based on the preliminary mask, the amplitude spectrum, the phase spectrum and the transformation function: δ1=γ×(M×|G|×e j×L ) Among them, δ1 refers to the preliminary signal, and γ refers to the transformation function; Obtain a target signal, and calculate a time domain loss based on the preliminary signal and the target signal: Among them, V1 refers to the time domain loss, δ 2z refers to the target signal at the zth target sampling time point, δ 1z Refers to the preliminary signal at the zth target sampling time point; The frequency domain loss is calculated based on the amplitude spectrum and the phase spectrum, and the learning parameters are set. The optimization mask is calculated using the preliminary mask, the time domain loss, the frequency domain loss and the learning parameters: Among them, M2 refers to the optimization mask, ∈ refers to the learning parameter, λ refers to the partial derivative, and V2 refers to the frequency domain loss; The reconstructed signal is calculated based on the transformation function, the optimized mask, the amplitude spectrum and the phase spectrum: δ2=γ×(M2×|G|×e j×L ) Here, δ2 refers to the reconstructed signal.

9. The method for multi-target orientation recognition of industrial noise based on sound source separation and recognition as claimed in claim 8, characterized in that: The calculating the spatial position based on the horizontal angle, the vertical angle and the sound source distance comprises: Obtain the center position, determine the center position as the origin, and establish an orientation coordinate system based on the origin; Calculate the spatial position based on the azimuth coordinate system, horizontal angle, vertical angle and sound source distance: x′=r′×cosφ×sinζ y′=r′×cosφ×cosζ z′=r′×sinφ S′=(x′,y′,z′) Among them, x′ refers to the x-axis coordinate, r′ refers to the distance of the sound source, cos refers to the cosine function, φ refers to the vertical angle, ζ refers to the horizontal angle, y′ refers to the y-axis coordinate, z′ refers to the z-axis coordinate, and S′ refers to the spatial position.

10. An industrial noise multi-target position recognition system based on sound source separation and recognition, characterized in that: The system comprises: A conversion signal calculation module is used to set a sensor array, collect multi-target industrial noise audio based on the sensor array and record the sampling frequency, perform data standardization based on the multi-target industrial noise audio to obtain calibration audio; determine the target sampling frequency, perform sampling frequency matching based on the sampling frequency to obtain matching data, perform sampling frequency conversion based on the matching data, the calibration audio and the target sampling frequency to obtain a conversion signal; A preliminary mask calculation module is used to calculate the audio energy of the conversion signal and normalize the audio energy to obtain normalized audio; calculate the amplitude spectrum and phase spectrum based on the normalized audio, and generate a preliminary mask based on the amplitude spectrum, phase spectrum and a preset deep neural network, wherein the deep neural network is a neural network that combines a convolutional neural network and a fully connected network; A reconstruction signal calculation module is used to calculate the time domain loss and the frequency domain loss, calculate the optimized mask based on the time domain loss, the frequency domain loss and the preliminary mask, and calculate the reconstructed signal using the optimized mask; The spatial position calculation module is used to calculate the sensing time difference based on the reconstructed signal, calculate the horizontal angle and the vertical angle using the sensing time difference, calculate the sound source distance, calculate the spatial position based on the horizontal angle, the vertical angle and the sound source distance, and complete the orientation recognition.

Images