Abnormal sound detection method, device, equipment, storage medium and program product

Abstract

The present application discloses an abnormal sound detection method, device, equipment, storage medium and program product, which relates to the field of artificial intelligence technology, including: collecting audio signals from various monitoring points of industrial equipment; corresponding to each monitoring point, reconstructing the target data of the monitoring point based on the audio signal of the monitoring point, the target data represents a normal audio signal; calculating the similarity between the audio signal of the monitoring point and the target data; if the similarity is less than the target threshold, determining that the audio signal of the monitoring point is an abnormal audio signal. The abnormal sound detection scheme provided by the present application is suitable for performing abnormal sound detection on multiple different monitoring points of industrial equipment, realizing a relatively comprehensive abnormality detection of industrial equipment, and ensuring the overall safety and reliability of industrial equipment.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, equipment, storage medium and program product for detecting abnormal sounds. Background Art

[0002] At present, in order to ensure the normal operation of industrial equipment, it is necessary to detect abnormal conditions of industrial equipment. Among them, abnormal sound detection (abnormal sound for short) is an important monitoring method for abnormal monitoring of industrial equipment.

[0003] Currently, abnormal noise monitoring of industrial equipment can only perform abnormal noise detection on a single monitoring point of the industrial equipment, and it is impossible to realize abnormal noise detection at multiple monitoring points. Summary of the invention

[0004] In view of the above problems, the present application provides an abnormal sound detection method, device, equipment, storage medium and program product to realize abnormal sound detection at multiple monitoring points of industrial equipment. The specific scheme is as follows:

[0005] The first aspect of the present application provides a method for detecting abnormal sound, comprising:

[0006] Collect audio signals from various monitoring points of industrial equipment;

[0007] Corresponding to each monitoring point, reconstructing target data of the monitoring point based on the audio signal of the monitoring point, wherein the target data represents a normal audio signal;

[0008] Calculate the similarity between the audio signal of the monitoring point and the target data;

[0009] If the similarity is less than the target threshold, it is determined that the audio signal at the monitoring point is an abnormal audio signal.

[0010] In a possible implementation, the process of reconstructing target data of any monitoring point based on the audio signal of any monitoring point includes:

[0011] Performing noise reduction processing on the audio signal of any monitoring point to obtain an initial noise-reduced signal of any monitoring point;

[0012] Performing spectrum enhancement on the initial noise-reduced signal to obtain a noise-reduced signal at any monitoring point;

[0013] The target data of any monitoring point is reconstructed based on the noise-reduced signal of any monitoring point.

[0014] In a possible implementation, performing noise reduction processing on the audio signal of any monitoring point includes:

[0015] Performing short-time Fourier transform on the audio signal of any monitoring point to obtain a time-frequency signal of any monitoring point;

[0016] Performing noise estimation on the time-frequency signal of any monitoring point to obtain a noise spectrum of any monitoring point;

[0017] Based on the noise spectrum of any monitoring point, the time-frequency signal of any monitoring point is subjected to noise reduction processing to obtain an initial noise-reduced signal of any monitoring point.

[0018] In a possible implementation, the performing noise estimation on the time-frequency signal of any monitoring point includes:

[0019] Performing noise estimation on the time-frequency signal of any monitoring point by using a minimum mean square error algorithm or a minimum variance distortion-free response algorithm to obtain a noise spectrum of any monitoring point;

[0020] Alternatively, the noise of the time-frequency signal of any monitoring point is estimated by using a pre-trained deep convolutional neural network to obtain a noise spectrum of any monitoring point.

[0021] In a possible implementation, the process of reconstructing target data of any monitoring point based on the audio signal of any monitoring point includes:

[0022] Extracting acoustic features of the audio signal of any monitoring point;

[0023] The extracted acoustic features are input into a pre-trained reconstruction model, and the reconstruction model is obtained to randomly mask the received acoustic features, and the masked acoustic features are reconstructed based on the masked acoustic features; the reconstructed masked acoustic features and the unmasked acoustic features constitute the target data.

[0024] In a possible implementation, the reconstruction model is trained in the following manner:

[0025] Performing self-supervisory training on the reconstruction model through a first audio data set to obtain a first pre-trained reconstruction model; the audio data in the first audio data set is normal audio data in a general scenario;

[0026] Performing weakly supervised learning on the first pre-trained reconstruction model through a second audio data set to obtain a second pre-trained reconstruction model; the audio data in the second audio data set includes abnormal audio data and normal audio data in a general scenario;

[0027] The second pre-trained reconstruction model is fine-tuned through a third audio data set to obtain a trained reconstruction model; the audio data in the third audio data set includes abnormal audio data and normal audio data of the industrial equipment in the scenario to which the industrial equipment belongs.

[0028] In a possible implementation, performing weakly supervised learning on the first pre-trained reconstruction model using the second audio data set includes:

[0029] Extracting a batch of audio data from the second audio data set, inputting the acoustic features of the batch of audio data into the first pre-trained reconstruction model, and obtaining the acoustic features of masked speech frames of each audio data in the batch of audio data output by the first pre-trained reconstruction model; the reconstructed masked acoustic features and unmasked acoustic features of each audio data constitute the target data corresponding to the audio data;

[0030] The parameters of the first pre-trained reconstruction model are updated with the goal of at least increasing the difference between the target data corresponding to each abnormal audio data in the batch of audio data and the each abnormal audio data in the batch of audio data, and decreasing the difference between the target data corresponding to each normal audio data in the batch of audio data and the each normal audio data in the batch of audio data.

[0031] In a possible implementation, the method further includes: if it is determined that the audio signal at any monitoring point is an abnormal audio signal, outputting first prompt information, wherein the first prompt information indicates that an abnormality occurs at any monitoring point;

[0032] If feedback information is received for the first prompt information input, the feedback information indicates that the first prompt information is wrong, and the number of wrong prompts within a preset time period is counted;

[0033] If the number of false prompts within the preset time period is greater than the number threshold, the target threshold is increased to reduce the number of false prompts.

[0034] A second aspect of the present application provides an abnormal sound monitoring device, comprising:

[0035] Acquisition module, used to collect audio signals from various monitoring points of industrial equipment;

[0036] A reconstruction module, for corresponding to each monitoring point, reconstructing target data of the monitoring point based on the audio signal of the monitoring point, wherein the target data represents a normal audio signal;

[0037] A calculation module, used to calculate the similarity between the audio signal of the monitoring point and the target data;

[0038] The determination module is used to determine that the audio signal at the monitoring point is an abnormal audio signal if the similarity is less than a target threshold.

[0039] A third aspect of the present application provides a computer program product, including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements the abnormal sound detection method of the first aspect or any implementation of the first aspect.

[0040] A fourth aspect of the present application provides an electronic device, comprising at least one processor and a memory connected to the processor, wherein:

[0041] The memory is used to store computer programs;

[0042] The processor is used to execute the computer program so that the electronic device can implement the abnormal sound detection method of the first aspect or any implementation manner of the first aspect.

[0043] A fifth aspect of the present application provides a computer storage medium, which carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement the abnormal sound detection method of the first aspect or any implementation method of the first aspect.

[0044] By means of the above technical scheme, the abnormal sound detection method, device, equipment, storage medium and program product provided by the present application collect audio signals from various monitoring points of industrial equipment; corresponding to each monitoring point, reconstruct the target data of the monitoring point based on the audio signal of the monitoring point, and the target data represents the normal audio signal; calculate the similarity between the audio signal of the monitoring point and the target data; if the similarity is less than the target threshold, determine that the audio signal of the monitoring point is an abnormal audio signal. The abnormal sound detection method provided by the present application is suitable for abnormal sound detection at multiple different monitoring points of industrial equipment, realizing a relatively comprehensive abnormality detection of industrial equipment, and ensuring the overall safety and reliability of industrial equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale.

[0046] Figure 1 A flowchart of an implementation of the abnormal sound detection method provided in this application;

[0047] Figure 2 A flowchart for implementing the reconstruction of target data of any monitoring point based on the audio signal of any monitoring point provided by the present application;

[0048] Figure 3 A flowchart for implementing noise reduction processing of an audio signal at any monitoring point provided by the present application;

[0049] Figure 4 A flowchart for implementing the reconstruction of target data of any monitoring point based on the audio signal of any monitoring point provided by the present application;

[0050] Figure 5 A schematic diagram of the structure of the abnormal sound detection device provided in this application;

[0051] Figure 6 A schematic diagram of the structure of an electronic device provided in this application. DETAILED DESCRIPTION

[0052] The following describes the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. The terms used in the implementation method section of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0053] The embodiments of the present application are described below in conjunction with the accompanying drawings. Those skilled in the art will appreciate that, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0054] The terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, which is only to describe the distinction mode adopted by the objects of the same attributes when describing in the embodiments of the present application. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that the process, method, system, product or equipment comprising a series of units need not be limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0055] The present application is proposed in order to at least realize abnormal sound detection at multiple different monitoring points of industrial equipment.

[0056] Industrial equipment may include but is not limited to: power equipment, processing and production equipment, transportation equipment, etc.

[0057] For the convenience of description, the following description will be given by taking a water turbine as an example of industrial equipment.

[0058] With the growth of electricity demand and the adjustment of energy structure, hydropower plants, as an important part of clean energy, have put forward higher requirements for the safe operation of equipment. As the core equipment of hydropower plants, the safe and stable operation of turbine generator sets is crucial to the continuity of power output. As a reliable monitoring method, sound signals are of great significance for the detection of abnormal conditions of turbines.

[0059] Traditional abnormal noise detection methods mainly focus on the detection of single components (such as bearings, gear boxes, etc.). Not only is the accuracy low, but it is also unable to detect abnormal noise of other key components of the turbine (such as the nose slip ring chamber, waterwheel chamber, lower wind tunnel, etc.). That is, the scheme used to detect abnormal noise of bearings and gear boxes cannot be used for abnormal noise detection of nose slip ring chamber, waterwheel chamber, lower wind tunnel, etc.

[0060] like Figure 1 As shown, a flowchart of an implementation of the abnormal sound detection method provided in an embodiment of the present application may include:

[0061] Step S101: collecting audio signals from various monitoring points of industrial equipment.

[0062] Each monitoring point of industrial equipment can be the core part (or key part) of the industrial equipment, which is determined according to the actual operation of the industrial equipment.

[0063] Taking a turbine as an example of industrial equipment, the monitoring points of the turbine may include but are not limited to the following monitoring points: nose slip ring room, water wheel room, lower wind tunnel, pressure reducing valve, bearing, gear box, etc.

[0064] The methods of collecting audio signals at different monitoring points may be the same or different. For example, a contact sound sensor (such as a vibration sensor) may be used to collect audio signals, or a non-contact sound pickup (such as a microphone) may be used to collect audio signals.

[0065] Generally, if the monitoring point needs to collect ambient sound, a non-contact microphone can be used to collect audio signals; if the monitoring point does not need to collect ambient sound, a contact microphone can be used to collect audio signals.

[0066] For example, in the turbine's nose slip ring room, waterwheel room, lower wind tunnel and other places, non-contact pickups can be used to collect audio signals; while in the turbine's pressure reducing valve, bearings, gearboxes and other places, contact pickups can be used to collect audio signals.

[0067] Each monitoring point can be provided with one microphone, or multiple microphones. When multiple microphones are provided at one monitoring point, different microphones at the same monitoring point are distributed at different positions of the same monitoring point.

[0068] When multiple non-contact microphones are arranged at the same monitoring point, the audio signals collected by the multiple non-contact microphones can be combined by using beamforming technology to obtain the audio signal of the monitoring point.

[0069] When multiple contact microphones are set at the same monitoring point, the audio signal collected by each contact microphone at the same monitoring point is an audio signal of the monitoring point. Based on this, if the audio signal collected by any contact microphone at the monitoring point is abnormal, it indicates that the monitoring point is abnormal.

[0070] Step S102: corresponding to each monitoring point, reconstructing target data of the monitoring point based on the audio signal of the monitoring point, where the target data represents a normal audio signal.

[0071] Optionally, the target data may be an acoustic feature that characterizes a normal audio signal, that is, the audio signal corresponding to the acoustic feature is a normal audio signal. The target data may also be other data that can characterize a normal audio signal.

[0072] The acoustic features of the audio signal of the monitoring point can be processed by a pre-trained reconstruction model to obtain the target data of the monitoring point. The reconstruction model can be a deep neural network using an encoder-decoder structure, for example, a Transformer deep neural network, or other forms of deep neural networks. This application does not make specific restrictions, as long as the encoder-decoder structure is used.

[0073] In the present application, the acoustic features may include but are not limited to any of the following: Perceptual Linear Predictive (PLP), Mel-Frequency Cepstral Coefficients (MFCCF), and FBank (FilterBank).

[0074] Step S103: Calculate the similarity between the audio signal of the monitoring point and the target data.

[0075] If the target data is an acoustic feature, the distance between the acoustic feature of the audio signal of the monitoring point (ie, the collected audio signal) and the target data may be calculated, and the similarity between the audio signal of the monitoring point and the target data may be determined based on the distance.

[0076] As an example, the distance between the acoustic characteristics of the audio signal of the monitoring point and the target data is the Euclidean distance, and the similarity between the audio signal of the monitoring point and the target data is negatively correlated with the above distance, that is, the greater the distance, the smaller the similarity between the audio signal representing the monitoring point and the target data; the smaller the distance, the greater the similarity between the audio signal representing the monitoring point and the target data.

[0077] As an example, the distance between the acoustic feature of the audio signal of the monitoring point and the target data is the cosine distance (i.e., the cosine value of the angle), and the similarity between the audio signal of the monitoring point and the target data is positively correlated with the above distance, that is, the greater the distance, the greater the similarity between the audio signal of the monitoring point and the target data; the smaller the distance, the smaller the similarity between the audio signal of the monitoring point and the target data. In this example, the cosine distance can be used as the similarity between the audio signal of the monitoring point and the target data.

[0078] Step S104: if the similarity between the audio signal at the monitoring point and the target data is less than the target threshold, the audio signal at the monitoring point is determined to be an abnormal audio signal.

[0079] Because the target data represents a normal audio signal, if the similarity between the audio signal at the monitoring point and the target data is less than the target threshold, it means that the audio signal at the monitoring point is an abnormal audio signal, indicating that the industrial equipment is abnormal at the monitoring point.

[0080] Further, if it is determined that the audio signal at the monitoring point is an abnormal audio signal, a prompt message (for the convenience of description, recorded as the first prompt message) may be output, and the first prompt message indicates that an abnormality exists at the monitoring point.

[0081] Optionally, each microphone has a unique ID, and the association between the microphone ID and the position of the microphone on the industrial equipment can be recorded in advance. Each time audio data is obtained from the microphone, the obtained audio data can be associated with the microphone ID. In this way, when a certain audio signal is determined to be an abnormal audio signal, the microphone ID associated with the abnormal audio signal can be obtained, and then the position of the microphone that collects the abnormal audio signal can be determined based on the association between the microphone ID and the position of the microphone, that is, the position where the abnormal audio signal is located. Based on this, the first prompt information can also include the position of the microphone that collects the abnormal audio signal, so that relevant personnel can locate the abnormal sound.

[0082] The abnormal sound detection method provided in the embodiment of the present application collects audio signals from various monitoring points of industrial equipment (e.g., a turbine); for each monitoring point, reconstructs the target data of the monitoring point based on the audio signal of the monitoring point, and the target data represents a normal audio signal; calculates the similarity between the audio signal of the monitoring point and the target data; if the similarity is less than the target threshold, determines that the audio signal of the monitoring point is an abnormal audio signal. The abnormal sound detection method provided in the present application is suitable for performing abnormal sound detection on multiple different monitoring points of industrial equipment (e.g., a turbine), realizes a relatively comprehensive abnormality detection of industrial equipment (e.g., a turbine), and ensures the overall safety and reliability of the turbine.

[0083] The present application performs abnormal sound detection in the same manner at each monitoring point. In an optional embodiment, a flowchart for implementing target data of any monitoring point based on the audio signal of the monitoring point is shown in FIG. Figure 2 As shown, this may include:

[0084] Step S201: performing noise reduction processing on the audio signal of any monitoring point to obtain an initial noise-reduced signal of the any monitoring point.

[0085] Step S202: Spectrum enhancement is performed on the initial noise-reduced signal to obtain a noise-reduced signal of any monitoring point.

[0086] By performing spectrum enhancement on the signal after initial noise reduction, the signal quality can be improved and the noise can be further suppressed.

[0087] Step S203: reconstructing the target data of any monitoring point based on the noise-reduced signal of any monitoring point.

[0088] By reducing the noise of the audio signal and enhancing the spectrum, the noise in the audio signal can be filtered out, the quality of the audio signal can be improved, the interference of noise on data reconstruction can be avoided, the accuracy of the reconstructed target data can be improved, and the accuracy of abnormal sound detection can be improved.

[0089] In an optional embodiment, a flowchart of implementing the above-mentioned noise reduction processing of the audio signal of any monitoring point is as follows: Figure 3 As shown, this may include:

[0090] Step S301: performing short-time Fourier transform on the audio signal of any monitoring point to obtain the time-frequency signal of the any monitoring point.

[0091] Short-time Fourier transform can analyze the changes of audio signals in both time and frequency dimensions simultaneously, thereby capturing the time-varying characteristics of the signal's frequency domain information (e.g., frequency), and can more comprehensively capture the characteristics of the audio signal.

[0092] Assuming that the audio signal of any monitoring point is y(t), the corresponding time-frequency signal Y(k,m) can be written as:

[0093] Y(k,m)=STFT(y(t)) (1)

[0094] Where k represents the frequency index, m represents the time frame index, and STFT represents short-time Fourier transform.

[0095] Step S302: performing noise estimation on the time-frequency signal of any monitoring point to obtain a noise spectrum of the any monitoring point.

[0096] The noise of the time-frequency signal of any monitoring point can be estimated by using any of the following two noise estimation methods:

[0097] Method 1: An algorithm can be used to estimate the noise of the time-frequency signal of any monitoring point.

[0098] As an example, the noise of the time-frequency signal of any monitoring point can be estimated by using a minimum mean square error (MMSE) algorithm (also called a mean smoothing algorithm) to obtain a noise spectrum of the any monitoring point.

[0099] As an example, the noise of the time-frequency signal of any monitoring point can be estimated by using a minimum variance distortionless response (MVDR) algorithm to obtain a noise spectrum of the any monitoring point.

[0100] Method 2: The noise of the time-frequency signal of any monitoring point can be estimated through a pre-trained deep convolutional neural network to obtain the noise spectrum of any monitoring point output by the deep convolutional neural network. It can be expressed as:

[0101] (2)

[0102] in, Represents the noise spectrogram estimated by a deep convolutional neural network, and CNN represents a deep convolutional neural network.

[0103] Optionally, the deep convolutional neural network is trained using the time-frequency signal with noisy frequency as a training sample and the spectrum of noise in the noisy frequency as a label.

[0104] Step S303: performing noise reduction processing on the time-frequency signal of any monitoring point based on the noise spectrum of any monitoring point to obtain an initial noise-reduced signal of any monitoring point.

[0105] Wiener filtering can be used to reduce the noise of the time-frequency signal of any monitoring point. Alternatively, spectral subtraction can be used to reduce the noise of the time-frequency signal of any monitoring point.

[0106] As an example, one implementation method of using Wiener filtering to perform noise reduction processing on the time-frequency signal of any monitoring point may be:

[0107] (3)

[0108] in, is the initial denoised signal. Whether it is a time-frequency signal or a noise spectrogram, it is composed of frequency information of multiple time frames (i.e., audio frames). Therefore, Formula 3 is a formula for calculating the frequency information of each time frame.

[0109] Through deep learning-based adaptive noise estimation (i.e. noise estimation based on deep convolutional neural networks) and noise reduction technology, the frequency domain resolution and time domain accuracy of audio signals can be significantly improved.

[0110] In an optional embodiment, the present application provides an implementation flowchart of reconstructing the target data of any monitoring point based on the audio signal of the monitoring point as shown in FIG. Figure 4 As shown, this may include:

[0111] Step S401: extracting the acoustic features of the audio signal of any monitoring point.

[0112] Optionally, the acoustic feature extraction can be performed directly on the audio signal of any monitoring point. Usually, the audio signal is firstly processed into frames to obtain multiple audio frames, and then the acoustic features of each audio frame are extracted.

[0113] Optionally, the audio signal of any monitoring point can be denoised based on the aforementioned method to obtain an initial denoised signal of any monitoring point, the initial denoised signal can be spectrally enhanced to obtain a denoised signal of any monitoring point, and acoustic features can be extracted from each audio frame of the denoised signal of any monitoring point.

[0114] Step S402: Input the extracted acoustic features into a pre-trained reconstruction model, obtain the reconstruction model to randomly mask the received acoustic features, and reconstruct the masked acoustic features based on the masked acoustic features; the reconstructed masked acoustic features and the unmasked acoustic features constitute the target data.

[0115] After the extracted acoustic features are input into the pre-trained reconstruction model, the reconstruction model randomly masks the received acoustic features to randomly cover the acoustic features of part of the audio frame, obtains the masked acoustic features (including the covered acoustic features and the uncovered acoustic features), encodes the masked acoustic features to obtain the encoded features, and decodes the encoded features to obtain the reconstructed acoustic features of the covered part of the speech frame. The reconstructed acoustic features and the acoustic features of the uncovered speech frame constitute the target data.

[0116] In order to capture the complex patterns in the audio signal, the reconstruction model can use the Transformer deep neural network, which includes an encoding layer based on the attention mechanism and a decoding layer based on the attention mechanism. Among them, the encoding layer based on the attention mechanism can effectively capture the timing information and frequency domain information in the audio signal.

[0117] In an optional embodiment, the reconstruction model can be obtained by training in three training stages: self-supervised training, weakly supervised training, and fine-tuning.

[0118] By adopting a combination of self-supervision and weak supervision pre-training, the reconstruction model can effectively learn the complex patterns and potential structures of audio signals, which can improve the generalization ability and detection accuracy of the reconstruction model.

[0119] Optionally, the reconstruction model may be self-supervised trained using a first audio data set to obtain a first pre-trained reconstruction model; the audio data in the first audio data set is normal audio data in a general scenario.

[0120] The general scenario includes multiple scenarios, which may include scenarios to which industrial equipment belongs, or may not include scenarios to which industrial equipment belongs. Normal audio data in any scenario refers to audio data of a monitoring point of a device collected when the device is operating normally in any scenario.

[0121] As an example, industrial equipment in common scenarios may include multiple types of industrial equipment (e.g., power equipment, processing and production equipment, transportation equipment, etc.), and each type of industrial equipment can be further divided into different subtypes (e.g., power equipment can be further divided into: turbines, steam turbines, diesel engines, generators, transformers, wind turbines, etc.).

[0122] By using normal audio data in common scenarios to perform self-supervisory training on the reconstruction model, the reconstruction model can learn richer feature representations of normal audio data in common scenarios.

[0123] Optionally, a method for implementing self-supervised training of the reconstruction model may be:

[0124] For each audio data in the first audio data set, the acoustic features of the audio data are input into the reconstruction model to obtain the acoustic features of the covered time frame corresponding to the audio data output by the reconstruction model; the acoustic features output by the reconstruction model and the uncovered acoustic features corresponding to the audio data constitute the target data.

[0125] The reconstruction model randomly masks the acoustic features of the audio data to cover the acoustic features of part of the time frame to obtain the masked acoustic features, encodes the masked acoustic features to obtain the encoded features, and decodes the encoded features to obtain the acoustic features of the covered time frame.

[0126] The parameters of the reconstruction model are updated with the goal of reducing the difference between the target data corresponding to the audio data and the audio data (that is, the difference between the acoustic features of the covered time frame output by the reconstruction model and the covered acoustic features).

[0127] Optionally, the parameters of the reconstruction model may be updated with the goal of making the acoustic features of the covered time frame output by the reconstruction model approach the original acoustic features of the covered time frame.

[0128] The first pre-trained reconstruction model is subjected to weakly supervised learning through a second audio data set to obtain a second pre-trained reconstruction model; the audio data in the second audio data set includes abnormal audio data and normal audio data in a common scenario.

[0129] Optionally, in the second audio data set, a proportion of abnormal audio data is smaller than a proportion of normal audio data.

[0130] The goal of performing weakly supervised learning on the first pre-trained reconstruction model through the second audio data set is to enable the reconstruction model to better learn the difference between abnormal audio data and normal audio data.

[0131] Optionally, the first pre-trained reconstruction model can be weakly supervised trained in the following way:

[0132] Each time a batch of audio data is extracted from the second audio data set, the acoustic features of the batch of audio data are input into the first pre-trained reconstruction model to obtain the acoustic features of the masked audio frames of each audio data in the batch of audio data output by the first pre-trained reconstruction model; the reconstructed masked acoustic features and unmasked acoustic features of each audio data constitute the target data corresponding to the audio data.

[0133] For each audio data in a batch of audio data, the first pre-trained reconstruction model randomly masks the acoustic features of the audio data, that is, randomly selects some audio frames, masks the acoustic features of these audio frames to obtain the masked acoustic features, encodes the masked acoustic features to obtain encoded features, and decodes the encoded features to obtain the acoustic features of the masked audio frames of the audio data.

[0134] Optionally, the parameters of the first pre-trained reconstruction model can be updated with the goal of increasing the difference between target data corresponding to each abnormal audio data in the batch of audio data and each abnormal audio data in the batch of audio data, and decreasing the difference between target data corresponding to each normal audio data in the batch of audio data and each normal audio data in the batch of audio data.

[0135] Optionally, the parameters of the first pre-trained reconstruction model may be updated with the goal of increasing the difference between target data corresponding to each abnormal audio data in the batch of audio data and each abnormal audio data in the batch of audio data, decreasing the difference between target data corresponding to each abnormal audio data in the batch of audio data and each normal audio data in the batch of audio data, and decreasing the difference between target data corresponding to each normal audio data in the batch of audio data and each normal audio data in the batch of audio data.

[0136] Optionally, the parameters of the first pre-trained reconstruction model may be updated with the goal of increasing the difference between target data corresponding to each abnormal audio data in the batch of audio data and each abnormal audio data in the batch of audio data, decreasing the difference between target data corresponding to each normal audio data in the batch of audio data and each normal audio data in the batch of audio data, and increasing the difference between target data corresponding to each normal audio data in the batch of audio data and each abnormal audio data in the batch of audio data.

[0137] Optionally, the parameters of the first pre-trained reconstruction model may be updated with the goal of increasing the difference between target data corresponding to each abnormal audio data in the batch of audio data and each abnormal audio data in the batch of audio data, decreasing the difference between target data corresponding to each abnormal audio data in the batch of audio data and each normal audio data in the batch of audio data, decreasing the difference between target data corresponding to each normal audio data in the batch of audio data and each normal audio data in the batch of audio data, and increasing the difference between target data corresponding to each normal audio data in the batch of audio data and each abnormal audio data in the batch of audio data.

[0138] That is to say, each batch of audio data contains both normal audio data and abnormal audio data.

[0139] The second pre-trained reconstruction model is fine-tuned by the third audio data set to obtain a trained reconstruction model. The audio data in the third audio data set includes abnormal audio data and normal audio data of the industrial equipment in the scene to which the industrial equipment belongs.

[0140] For example, if the industrial equipment is a water turbine, the audio data in the third audio data set includes abnormal audio data and normal audio data of each water turbine in the water turbine scene.

[0141] Optionally, in the third audio data set, the proportion of abnormal audio data is less than the proportion of normal audio data. The audio data in the third audio data set includes audio data of various monitoring points of industrial equipment, each monitoring point collects multiple audio data, and different audio data of the same monitoring point correspond to different environmental factors. Different environmental factors include at least one of the following differences:

[0142] The turbines are different. Including but not limited to at least one of the following: different manufacturers, different models, etc.

[0143] The operating parameters of the turbine are different, including but not limited to at least one of the following: different operating conditions, different environmental conditions (temperature, humidity, etc.), different load conditions (for example, power, frequency, etc.), etc.

[0144] The goal of fine-tuning the second pre-trained reconstruction model through the third audio data set is to make the reconstruction model better adapt to the abnormal sound detection in the scene of industrial equipment and achieve high-precision abnormal sound detection. By collecting audio data under different environmental factors, the reconstruction model can adapt to different environmental factors in the scene of industrial equipment.

[0145] Optionally, the process of fine-tuning the second pre-trained reconstruction model by using the third audio data set may refer to the aforementioned process of performing weakly supervised training on the first pre-trained reconstruction model by using the second audio data set, and may specifically include:

[0146] Each time a batch of audio data is extracted from the third audio data set, the acoustic features of the batch of audio data are input into the second pre-trained reconstruction model to obtain the acoustic features of the masked speech frames of each audio data in the batch of audio data output by the second pre-trained reconstruction model; the reconstructed masked acoustic features and unmasked acoustic features of each audio data constitute the target data corresponding to the audio data.

[0147] Optionally, the parameters of the second pre-trained reconstruction model can be updated with the goal of increasing the difference between target data corresponding to each abnormal audio data in the batch of audio data and each abnormal audio data in the batch of audio data, and decreasing the difference between target data corresponding to each normal audio data in the batch of audio data and each normal audio data in the batch of audio data.

[0148] Optionally, the parameters of the first pre-trained reconstruction model may be updated with the goal of increasing the difference between target data corresponding to each abnormal audio data in the batch of audio data and each abnormal audio data in the batch of audio data, decreasing the difference between target data corresponding to each abnormal audio data in the batch of audio data and each normal audio data in the batch of audio data, and decreasing the difference between target data corresponding to each normal audio data in the batch of audio data and each normal audio data in the batch of audio data.

[0149] Optionally, the parameters of the first pre-trained reconstruction model may be updated with the goal of increasing the difference between target data corresponding to each abnormal audio data in the batch of audio data and each abnormal audio data in the batch of audio data, decreasing the difference between target data corresponding to each normal audio data in the batch of audio data and each normal audio data in the batch of audio data, and increasing the difference between target data corresponding to each normal audio data in the batch of audio data and each abnormal audio data in the batch of audio data.

[0150] Optionally, the parameters of the first pre-trained reconstruction model may be updated with the goal of increasing the difference between target data corresponding to each abnormal audio data in the batch of audio data and each abnormal audio data in the batch of audio data, decreasing the difference between target data corresponding to each abnormal audio data in the batch of audio data and each normal audio data in the batch of audio data, decreasing the difference between target data corresponding to each normal audio data in the batch of audio data and each normal audio data in the batch of audio data, and increasing the difference between target data corresponding to each normal audio data in the batch of audio data and each abnormal audio data in the batch of audio data.

[0151] That is to say, each batch of audio data contains both normal audio data and abnormal audio data.

[0152] In an optional embodiment, if the audio signal of any monitoring point is determined to be an abnormal audio signal, a first prompt information is output, and the first prompt information indicates that the any monitoring point is abnormal. The first prompt information can be provided in at least one of the following prompting methods, but is not limited to: alarm sound prompt, light prompt, text prompt, voice prompt, etc.

[0153] If feedback information is received for the first prompt information input, the feedback information indicates that the first prompt information is wrong, and the number of wrong prompts within a preset time period is counted.

[0154] In an embodiment of the present application, after outputting the first prompt information, the user can check the actual situation of the monitoring point prompted by the first prompt information (i.e., whether the prompted monitoring point is actually abnormal). If the monitoring point prompted by the first prompt information does not have an abnormality, the user can provide feedback on the first prompt information, i.e., input feedback information, so as to inform the abnormal sound monitoring device that this prompt is wrong.

[0155] Each time a user receives feedback indicating a false alarm, the number of false alarms within a preset time period is counted. Here, the statistics are made for the entire industrial equipment, that is, no matter which monitoring point the false alarm is made at, it is counted as one false alarm, regardless of the monitoring point.

[0156] If the number of false prompts within the preset time period is greater than the number threshold, the target threshold is increased to reduce the number of false prompts.

[0157] If the number of false prompts within the preset time period is greater than the number threshold, it means that the probability of false prompts is high and the target threshold is set too low. The target threshold needs to be increased.

[0158] When raising the target threshold, the preset step length may be raised first, and then the number of false prompts within the preset time length may be continuously monitored, and then whether to continue adjusting the target threshold may be determined according to the number of false prompts within the preset time length.

[0159] This application implements adaptive adjustment of the threshold based on user feedback, which can effectively reduce false prompts caused by environmental changes and improve the stability and reliability of abnormal sound detection.

[0160] It should be noted that the above embodiments are described only by taking a water turbine as an example. In actual application scenarios, the abnormal sound detection method of the present application can also be used for abnormal sound detection of other industrial equipment.

[0161] Corresponding to the method embodiment, the present application also provides an abnormal sound detection device, such as Figure 5 FIG. 1 is a schematic diagram of a structure of an abnormal sound detection device provided in an embodiment of the present application, comprising:

[0162] Acquisition module 501, reconstruction module 502, calculation module 503 and determination module 504;

[0163] Wherein, the acquisition module 501 is used to collect audio signals from various monitoring points of industrial equipment;

[0164] The reconstruction module 502 is used to reconstruct the target data of each monitoring point based on the audio signal of the monitoring point, wherein the target data represents a normal audio signal;

[0165] The calculation module 503 is used to calculate the similarity between the audio signal of the monitoring point and the target data;

[0166] The determination module 504 is configured to determine that the audio signal at the monitoring point is an abnormal audio signal if the similarity is less than a target threshold.

[0167] The abnormal sound detection device provided in the embodiment of the present application collects audio signals from various monitoring points of industrial equipment; for each monitoring point, the target data of the monitoring point is reconstructed based on the audio signal of the monitoring point, and the target data represents a normal audio signal; the similarity between the audio signal of the monitoring point and the target data is calculated; if the similarity is less than the target threshold, the audio signal of the monitoring point is determined to be an abnormal audio signal. The abnormal sound detection method provided in the present application is suitable for performing abnormal sound detection on multiple different monitoring points of industrial equipment, realizing a relatively comprehensive abnormality detection of industrial equipment, and ensuring the overall safety and reliability of industrial equipment.

[0168] In an optional implementation, when the reconstruction module 502 reconstructs the target data of any monitoring point based on the audio signal of any monitoring point, it is used to:

[0169] Performing noise reduction processing on the audio signal of any monitoring point to obtain an initial noise-reduced signal of any monitoring point;

[0170] Performing spectrum enhancement on the initial noise-reduced signal to obtain a noise-reduced signal at any monitoring point;

[0171] The target data of any monitoring point is reconstructed based on the noise-reduced signal of any monitoring point.

[0172] In an optional implementation, when the reconstruction module 502 performs noise reduction processing on the audio signal of any monitoring point, it is used to:

[0173] Performing short-time Fourier transform on the audio signal of any monitoring point to obtain a time-frequency signal of any monitoring point;

[0174] Performing noise estimation on the time-frequency signal of any monitoring point to obtain a noise spectrum of any monitoring point;

[0175] Based on the noise spectrum of any monitoring point, the time-frequency signal of any monitoring point is subjected to noise reduction processing to obtain an initial noise-reduced signal of any monitoring point.

[0176] In an optional implementation, when the reconstruction module 502 performs noise estimation on the time-frequency signal of any monitoring point, it is used to:

[0177] Performing noise estimation on the time-frequency signal of any monitoring point by using a minimum mean square error algorithm or a minimum variance distortion-free response algorithm to obtain a noise spectrum of any monitoring point;

[0178] Alternatively, the noise of the time-frequency signal of any monitoring point is estimated by using a pre-trained deep convolutional neural network to obtain a noise spectrum of any monitoring point.

[0179] In an optional implementation, when the reconstruction module 502 reconstructs the target data of any monitoring point based on the audio signal of any monitoring point, it is used to:

[0180] Extracting acoustic features of the audio signal of any monitoring point;

[0181] The extracted acoustic features are input into a pre-trained reconstruction model, and the reconstruction model is obtained to randomly mask the received acoustic features, and the masked acoustic features are reconstructed based on the masked acoustic features; the reconstructed masked acoustic features and the unmasked acoustic features constitute the target data.

[0182] In an optional implementation, the abnormal sound detection device further includes a training module, which is used to train the reconstruction model in the following manner:

[0183] Performing self-supervisory training on the reconstruction model through a first audio data set to obtain a first pre-trained reconstruction model; the audio data in the first audio data set is normal audio data in a general scenario;

[0184] Performing weakly supervised learning on the first pre-trained reconstruction model through a second audio data set to obtain a second pre-trained reconstruction model; the audio data in the second audio data set includes abnormal audio data and normal audio data in a general scenario;

[0185] The second pre-trained reconstruction model is fine-tuned through a third audio data set to obtain a trained reconstruction model; the audio data in the third audio data set includes abnormal audio data and normal audio data of the industrial equipment in the scenario to which the industrial equipment belongs.

[0186] In an optional implementation, when the training module performs weakly supervised learning on the first pre-trained reconstruction model through the second audio data set, it is used to:

[0187] Extracting a batch of audio data from the second audio data set, inputting the acoustic features of the batch of audio data into the first pre-trained reconstruction model, and obtaining the acoustic features of masked speech frames of each audio data in the batch of audio data output by the first pre-trained reconstruction model; the reconstructed masked acoustic features and unmasked acoustic features of each audio data constitute the target data corresponding to the audio data;

[0188] The parameters of the first pre-trained reconstruction model are updated with the goal of at least increasing the difference between the target data corresponding to each abnormal audio data in the batch of audio data and the each abnormal audio data in the batch of audio data, and decreasing the difference between the target data corresponding to each normal audio data in the batch of audio data and the each normal audio data in the batch of audio data.

[0189] In an optional implementation, the abnormal sound detection device further includes:

[0190] A prompt module, configured to output first prompt information if it is determined that the audio signal of any monitoring point is an abnormal audio signal, wherein the first prompt information indicates that an abnormality occurs at any monitoring point;

[0191] A statistics module, configured to, if feedback information is received for the first prompt information input, the feedback information indicates that the first prompt information is wrong, and to count the number of wrong prompts within a preset time period;

[0192] The threshold adjustment module is used to increase the target threshold to reduce the number of false prompts if the number of false prompts within the preset time period is greater than the number threshold.

[0193] The present application also provides an electronic device in an embodiment. Figure 6 As shown, it shows a structural schematic diagram of an electronic device suitable for implementing the abnormal sound detection method in the embodiment of the present application. The electronic device in the embodiment of the present application can be a terminal device (for example, a mobile phone, a tablet computer, a laptop computer, a desktop computer, etc.), or a server (can be a single server, can be a server cluster, or can be a cloud server, etc.). Figure 6 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0194] like Figure 6 As shown, the electronic device may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 to a random access memory (RAM) 603. When the electronic device is powered on, various programs and data required for the operation of the electronic device are also stored in the RAM 603. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0195] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a memory card, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although Figure 6 An electronic device having various devices is shown, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0196] Also provided in an embodiment of the present application is a computer program product including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements any one of the question-answering methods provided in the embodiments of the present application.

[0197] A computer-readable storage medium is also provided in an embodiment of the present application. The storage medium carries one or more computer programs. When one or more computer programs are executed by an electronic device, the electronic device can implement any one of the question-answering methods provided in the embodiment of the present application.

[0198] It should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the device embodiments provided by the present application, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines.

[0199] Through the description of the above implementation mode, the technicians in the field can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a ROM, a RAM, a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a training device, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0200] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware, or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product. Professionals and technicians may use different methods to implement the described functions for each specific solution, but such implementation should not be considered beyond the scope of this application.

[0201] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, a computer, a training device, or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, training device, or data center. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device, a data center, etc. that includes one or more available media integrations. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

[0202] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0203] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting abnormal sound, characterized in that: include: Collect audio signals from various monitoring points of industrial equipment; Corresponding to each monitoring point, reconstructing target data of the monitoring point based on the audio signal of the monitoring point, wherein the target data represents a normal audio signal; Calculate the similarity between the audio signal of the monitoring point and the target data; If the similarity is less than the target threshold, it is determined that the audio signal at the monitoring point is an abnormal audio signal.

2. The method according to claim 1, characterized in that The process of reconstructing target data of any monitoring point based on the audio signal of any monitoring point includes: Performing noise reduction processing on the audio signal of any monitoring point to obtain an initial noise-reduced signal of any monitoring point; Performing spectrum enhancement on the initial noise-reduced signal to obtain a noise-reduced signal at any monitoring point; The target data of any monitoring point is reconstructed based on the noise-reduced signal of any monitoring point.

3. The method according to claim 2, characterized in that The audio signal of any monitoring point is subjected to noise reduction processing, comprising: Performing short-time Fourier transform on the audio signal of any monitoring point to obtain a time-frequency signal of any monitoring point; Performing noise estimation on the time-frequency signal of any monitoring point to obtain a noise spectrum of any monitoring point; Based on the noise spectrum of any monitoring point, the time-frequency signal of any monitoring point is subjected to noise reduction processing to obtain an initial noise-reduced signal of any monitoring point.

4. The method according to claim 3, characterized in that The performing noise estimation on the time-frequency signal of any monitoring point comprises: Performing noise estimation on the time-frequency signal of any monitoring point by using a minimum mean square error algorithm or a minimum variance distortion-free response algorithm to obtain a noise spectrum of any monitoring point; Alternatively, the noise of the time-frequency signal of any monitoring point is estimated by using a pre-trained deep convolutional neural network to obtain a noise spectrum of any monitoring point.

5. The method according to any one of claims 1 to 4, characterized in that: The process of reconstructing target data of any monitoring point based on the audio signal of any monitoring point includes: Extracting acoustic features of the audio signal of any monitoring point; The extracted acoustic features are input into a pre-trained reconstruction model, and the reconstruction model is obtained to randomly mask the received acoustic features, and the masked acoustic features are reconstructed based on the masked acoustic features; the reconstructed masked acoustic features and the unmasked acoustic features constitute the target data.

6. The method according to claim 5, characterized in that The reconstruction model is trained in the following way: Performing self-supervisory training on the reconstruction model through a first audio data set to obtain a first pre-trained reconstruction model; the audio data in the first audio data set is normal audio data in a general scenario; Performing weakly supervised learning on the first pre-trained reconstruction model through a second audio data set to obtain a second pre-trained reconstruction model; the audio data in the second audio data set includes abnormal audio data and normal audio data in a general scenario; The second pre-trained reconstruction model is fine-tuned through a third audio data set to obtain a trained reconstruction model; the audio data in the third audio data set includes abnormal audio data and normal audio data of the industrial equipment in the scenario to which the industrial equipment belongs.

7. The method according to claim 6, characterized in that The performing weakly supervised learning on the first pre-trained reconstruction model by using the second audio data set includes: Extracting a batch of audio data from the second audio data set, inputting the acoustic features of the batch of audio data into the first pre-trained reconstruction model, and obtaining the acoustic features of masked speech frames of each audio data in the batch of audio data output by the first pre-trained reconstruction model; the reconstructed masked acoustic features and unmasked acoustic features of each audio data constitute the target data corresponding to the audio data; The parameters of the first pre-trained reconstruction model are updated with the goal of at least increasing the difference between the target data corresponding to each abnormal audio data in the batch of audio data and the each abnormal audio data in the batch of audio data, and decreasing the difference between the target data corresponding to each normal audio data in the batch of audio data and the each normal audio data in the batch of audio data.

8. The method according to claim 1, characterized in that Also includes: If it is determined that the audio signal of any monitoring point is an abnormal audio signal, outputting first prompt information, wherein the first prompt information indicates that an abnormality occurs at any monitoring point; If feedback information is received for the first prompt information input, the feedback information indicates that the first prompt information is wrong, and the number of wrong prompts within a preset time period is counted; If the number of false prompts within the preset time period is greater than the number threshold, the target threshold is increased to reduce the number of false prompts.

9. An abnormal sound monitoring device, characterized in that: include: Acquisition module, used to collect audio signals from various monitoring points of industrial equipment; A reconstruction module, for corresponding to each monitoring point, reconstructing target data of the monitoring point based on the audio signal of the monitoring point, wherein the target data represents a normal audio signal; A calculation module, used to calculate the similarity between the audio signal of the monitoring point and the target data; The determination module is used to determine that the audio signal at the monitoring point is an abnormal audio signal if the similarity is less than a target threshold.

10. A computer program product, characterized in that The method comprises computer-readable instructions, and when the computer-readable instructions are executed on an electronic device, the electronic device implements the abnormal sound detection method according to any one of claims 1 to 8.

11. An electronic device, characterized in that: The electronic device comprises at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program so that the electronic device can implement the abnormal sound detection method as described in any one of claims 1 to 8.

12. A computer storage medium, characterized in that: The storage medium carries one or more computer programs, and when the one or more computer programs are executed by an electronic device, the electronic device can implement the abnormal sound detection method as described in any one of claims 1 to 8.

Images