Noise processing method and system for voiceprint recognition

Abstract

The invention relates to the technical field of noise processing, and discloses a noise processing method and system for voiceprint recognition, and the method comprises the steps: collecting environment sound information when a pump unit does not work and fused sound information when the pump unit works; performing time-frequency analysis on the environment sound information to obtain an environment noise time-frequency graph, and performing time-frequency analysis on the fused sound information to obtain a fused time-frequency graph; calculating a feature vector of each environmental noise according to the environmental noise time-frequency graph, calculating a feature vector of each fused sound according to the fused time-frequency graph, and calculating similarity information between the feature vector of each environmental noise and the feature vector of each fused sound; removing each environment noise in the fused time-frequency graph according to the similarity information to obtain a denoised time-frequency graph; preliminarily judging whether the pump unit breaks down or not according to the de-noised time-frequency diagram, and judging the state of the pump unit again in combination with sensor data. The method can effectively distinguish the environmental noise from the working sound of the pump unit, and provides an accurate sound feature extraction means for fault detection.

Description

Technical Field

[0001] The present invention relates to the field of noise processing technology, and in particular to a noise processing method and system for voiceprint recognition. Background Art

[0002] With the continuous improvement of industrial automation and intelligence, pump units are important equipment widely used in chemical, electric power, petroleum and other fields. The monitoring of their reliability and operating status has become the key to ensuring production safety and efficiency. During long-term operation, pump units will be affected by external environmental noise, which not only increases the complexity of monitoring, but also may affect the early detection and diagnosis of faults. Therefore, how to accurately extract effective information from the soundprint signal of the pump unit, remove environmental noise, and combine other sensor data for comprehensive analysis is an important issue in the current pump unit fault monitoring technology.

[0003] Traditional pump unit fault detection methods mostly rely on sensor data such as vibration and temperature, and make abnormal judgments by setting thresholds. However, these methods often fail to capture subtle changes in equipment in real time and accurately, especially in high-noise environments, where sensor signals are easily disturbed, affecting the accuracy of fault diagnosis.

[0004] Therefore, it is necessary to provide a noise processing method and system for voiceprint recognition to solve the problem that the prior art has poor effect in distinguishing between environmental noise and the working sound of the pump unit. Summary of the invention

[0005] In view of this, the present invention proposes a noise processing method and system for voiceprint recognition, aiming to solve the problem that the prior art has poor effect in distinguishing between environmental noise and the working sound of a pump unit.

[0006] On the one hand, the present invention proposes a noise processing method for voiceprint recognition, comprising: Collect the environmental sound information when the pump unit is not working and the fused sound information when the pump unit is working; Performing time-frequency analysis on the ambient sound information to obtain an ambient noise time-frequency graph, and performing time-frequency analysis on the fused sound information to obtain a fused time-frequency graph; Calculating the feature vector of each environmental noise according to the environmental noise time-frequency diagram, calculating the feature vector of each fused sound according to the fused time-frequency diagram, and calculating the similarity information between the feature vector of each environmental noise and the feature vector of each fused sound; According to the similarity information, various environmental noises in the fused time-frequency graph are removed to obtain a denoised time-frequency graph; A preliminary judgment is made based on the denoised time-frequency diagram as to whether the pump unit is faulty. If the preliminary judgment result is that a fault has occurred, the sensor data of the pump unit is obtained, and the state of the pump unit is judged again in combination with the sensor data; wherein the sensor data includes temperature data and vibration data.

[0007] Furthermore, the collection of the environmental sound information when the pump unit is not working and the fused sound information when the pump unit is working includes: A sound collection device is provided on the pump unit, and a sound collection duration is preset; Collecting environmental sound information with the sound collection time as the length; storing the environmental sound information; After the environmental sound information is collected, the fused sound information of the pump unit when it is working is collected, and the fused sound information is divided and stored in units of the sound collection time.

[0008] Furthermore, the performing of time-frequency analysis on the ambient sound information to obtain an ambient noise time-frequency graph, and the performing of time-frequency analysis on the fused sound information to obtain a fused time-frequency graph, include: Decomposing the ambient sound signal into several time-frequency frames by short-time Fourier transform, each frame corresponds to a frequency component in a time window, obtaining the frequency distribution in each time-frequency frame by Fourier transform, generating an ambient noise time-frequency graph from the frequency distribution of several time-frequency frames, and classifying and numbering the noises of different frequencies in the ambient noise time-frequency graph by K-means clustering algorithm; The fused sound signal of the pump unit at work is also decomposed into several time-frequency frames through short-time Fourier transform processing to generate a fused time-frequency graph. The sounds of different frequencies in the fused time-frequency graph are classified and numbered through the K-means clustering algorithm.

[0009] Furthermore, the step of calculating the characteristic vectors of each environmental noise according to the environmental noise time-frequency diagram and calculating the characteristic vectors of each fused sound according to the fused time-frequency diagram includes: The feature vectors of each environmental noise and each fused sound feature vector include Mel-frequency cepstral coefficients and spectrum centroids; Apply the Mel filter bank to the spectrum of each frame, convert the spectrum into a spectrum representation on the Mel frequency band, perform logarithmic transformation on the Mel spectrum, and use discrete cosine transform to extract the first 13 Mel frequency cepstral coefficients; Calculate the spectrum centroid of each frame of spectrum; wherein the spectrum centroid is calculated by the following formula: ; Where f represents the frequency of each frame spectrum, |S(f)| represents the amplitude at frequency f; Construct the feature vector sequence Ai=(ai1, ai2, ai3, ..., ai13, Ci) of each environmental noise; wherein Ai represents the feature vector sequence of the i-th environmental noise, Ci represents the spectrum centroid of the i-th environmental noise, aik represents the k-th Mel-frequency cepstrum coefficient of the i-th environmental noise, and k=1, 2, 3, ..., 13; Construct the feature vector sequence Bj=(bj1,bj2,bj3,…,bj13,Cj) of each fused sound; wherein Bj represents the feature vector sequence of the j-th fused sound in the fused sound, Cj represents the spectral centroid of the j-th fused sound, bjk represents the k-th Mel-frequency cepstral coefficient of the j-th fused sound, and k=1,2,3,…,13.

[0010] Furthermore, the calculation of the similarity information between the feature vectors of each environmental noise and the feature vectors of each fused sound includes: The similarity between the feature vector of each environmental noise and the feature vector of each fused sound is calculated by the following formula: ; In the above formula, S represents the similarity between the feature vector of each ambient noise and the feature vector of each fused sound, aik represents the kth Mel-frequency cepstral coefficient of the ith ambient noise, k=1, 2, 3, ..., 13; bjk represents the kth Mel-frequency cepstral coefficient of the jth fused sound, k=1, 2, 3, ..., 13; Ci represents the spectral centroid of the ith ambient noise, and Cj represents the spectral centroid of the jth fused sound.

[0011] Furthermore, the removing of various environmental noises in the fused time-frequency graph according to the similarity information to obtain a denoised time-frequency graph includes: A similarity threshold is set, and if the similarity between the feature vector of the environmental noise and the feature vector of the fused sound is greater than or equal to the similarity threshold, it is determined to remove the fused sound corresponding to the environmental noise in the fused time-frequency graph.

[0012] Furthermore, the preliminary determination of whether the pump unit fails according to the denoised time-frequency graph includes: Collecting sound amplitude data of the denoised time-frequency graph, and calculating the standard deviation of the sound amplitude data; Setting a maximum standard deviation value, if the standard deviation is greater than or equal to the maximum standard deviation value, preliminarily determining that the pump unit fails; If the standard deviation is less than the maximum standard deviation, it is preliminarily determined that the pump unit has not failed.

[0013] Furthermore, if the preliminary determination result is that a fault has occurred, obtaining the sensor data of the pump unit, and determining the state of the pump unit again in combination with the sensor data, includes: Setting a maximum vibration value and a temperature safety value, if the vibration data of the pump unit is greater than or equal to the maximum vibration value, and / or the temperature data is greater than or equal to the temperature safety value, then determining that the pump unit is faulty again; Otherwise, it is determined again that the state of the pump unit is a safe state.

[0014] Furthermore, if the preliminary determination result is that a fault has occurred, obtaining the sensor data of the pump unit, and determining the state of the pump unit again in combination with the sensor data, further includes: If the initial judgment result and the re-judgment result are inconsistent, a safety warning will be issued; If both the initial judgment result and the re-judgment result indicate a fault, a fault alarm is issued; If both the initial judgment result and the re-judgment result are that no fault has occurred, no safety warning or fault alarm will be issued.

[0015] Compared with the prior art, the beneficial effect of the present invention is that the present invention can effectively distinguish environmental noise from the working sound of the pump unit by performing time-frequency analysis on the soundprint signal of the pump unit, and provide an accurate sound feature extraction method for fault detection. When the pump unit is not working, the collected environmental sound information can be used as a noise model for comparison and analysis with the fused sound information when the pump unit is working. By calculating the feature vector of the time-frequency graph of the environmental noise and the fused sound, important sound features such as Mel frequency cepstral coefficients (MFCC) and spectrum centroid can be extracted, which can effectively distinguish noise from normal working sounds. Then, by calculating the similarity between the environmental noise feature and the fused sound feature, the environmental noise in the time-frequency graph is removed to obtain a purer denoised time-frequency graph, which can accurately identify and filter noise without relying on a large amount of sensor data, thereby improving the accuracy and response speed of fault detection. If the denoised sound signal indicates that the pump unit may fail, further judgment is made in combination with sensor data such as temperature and vibration, so as to achieve more comprehensive and accurate pump unit fault monitoring. This fault detection method based on comprehensive analysis of voiceprint features and sensor data helps to improve equipment maintenance efficiency, reduce downtime, and ensure the reliability and safety of pump units during operation.

[0016] On the other hand, the present application also provides a noise processing system for voiceprint recognition, comprising: A collection module is configured to collect environmental sound information when the pump unit is not working and fused sound information when the pump unit is working; A time-frequency graph output module is configured to perform time-frequency analysis on the ambient sound information to obtain an ambient noise time-frequency graph, and perform time-frequency analysis on the fused sound information to obtain a fused time-frequency graph; A comparison module is configured to calculate the feature vector of each environmental noise according to the environmental noise time-frequency diagram, calculate the feature vector of each fused sound according to the fused time-frequency diagram, and calculate the similarity information between the feature vector of each environmental noise and the feature vector of each fused sound; The comparison module is further configured to remove various environmental noises in the fused time-frequency graph according to the similarity information to obtain a denoised time-frequency graph; The judgment module is configured to preliminarily judge whether the pump unit has a fault based on the denoised time-frequency diagram. If the preliminary judgment result is a fault, the pump unit sensor data is obtained, and the state of the pump unit is judged again in combination with the sensor data; wherein the sensor data includes temperature data and vibration data.

[0017] It can be understood that the noise processing method and system for voiceprint recognition provided in the present application have the same beneficial effects, which will not be described in detail here. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings: Figure 1 A flowchart of a noise processing method for voiceprint recognition provided by an embodiment of the present invention; Figure 2 This is a functional block diagram of a noise processing system for voiceprint recognition provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0020] In some embodiments of the present application, see Figure 1 As shown, this embodiment provides a noise processing method for voiceprint recognition, comprising the following steps: S100, collecting environmental sound information when the pump unit is not working and fused sound information when the pump unit is working; S200, performing time-frequency analysis on the ambient sound information to obtain an ambient noise time-frequency graph, and performing time-frequency analysis on the fused sound information to obtain a fused time-frequency graph; S300, calculating the feature vector of each environmental noise according to the environmental noise time-frequency diagram, calculating the feature vector of each fused sound according to the fused time-frequency diagram, and calculating the similarity information between the feature vector of each environmental noise and the feature vector of each fused sound; S400, removing various environmental noises in the fused time-frequency graph according to the similarity information to obtain a denoised time-frequency graph; S500. Preliminarily determine whether the pump unit is faulty based on the denoised time-frequency graph. If the preliminary determination result is that a fault has occurred, obtain sensor data of the pump unit, and determine the state of the pump unit again based on the sensor data; wherein the sensor data includes temperature data and vibration data.

[0021] It can be understood that the present invention can effectively distinguish between environmental noise and the working sound of the pump unit by performing time-frequency analysis on the soundprint signal of the pump unit, and provide an accurate sound feature extraction method for fault detection. When the pump unit is not working, the collected environmental sound information can be used as a noise model for comparison and analysis with the fused sound information when the pump unit is working. By calculating the feature vector of the time-frequency graph of the environmental noise and the fused sound, important sound features such as Mel frequency cepstral coefficients (MFCC) and spectrum centroid can be extracted, which can effectively distinguish between noise and normal working sounds. Then, by calculating the similarity between the environmental noise feature and the fused sound feature, the environmental noise in the time-frequency graph is removed to obtain a purer denoised time-frequency graph, which can accurately identify and filter noise without relying on a large amount of sensor data, thereby improving the accuracy and response speed of fault detection. If the denoised sound signal indicates that the pump unit may fail, further judgment is made in combination with sensor data such as temperature and vibration, so as to achieve more comprehensive and accurate pump unit fault monitoring. This fault detection method based on comprehensive analysis of voiceprint features and sensor data helps to improve equipment maintenance efficiency, reduce downtime, and ensure the reliability and safety of pump units during operation.

[0022] In some embodiments of the present application, when collecting the environmental sound information when the pump unit is not working and the fused sound information when the pump unit is working, it includes: A sound collection device is installed on the pump unit, and the sound collection duration is pre-set; Collecting environmental sound information with the sound collection time as the length; storing the environmental sound information; When the environmental sound information is collected, the fused sound information of the pump unit when it is working is collected, and the fused sound information is divided and stored in units of sound collection time.

[0023] It is understandable that the present invention sets a sound collection device and pre-sets the collection time. By collecting and storing the ambient sound when the pump unit is not working and the fused sound when it is working, it helps to accurately obtain the difference between the ambient noise and the working soundprint of the pump unit. First, the collection of ambient sound information can provide reliable benchmark data for subsequent noise removal and feature extraction, and help distinguish the interference between the working soundprint of the pump unit and the ambient noise. Secondly, by segmenting and storing the fused sound when the pump unit is working, the collected data can be managed in a refined manner to facilitate subsequent time-frequency analysis and feature extraction. Setting a consistent collection time can ensure data consistency, reduce accidental errors in the collection process, and help improve the accuracy of noise processing and fault judgment. In addition, the segmented storage method facilitates the rapid processing and analysis of subsequent data, and also provides support for real-time monitoring and fault warning. Overall, this data collection method improves the accuracy and efficiency of pump unit status monitoring, and helps to achieve more intelligent and accurate fault diagnosis.

[0024] In some embodiments of the present application, performing time-frequency analysis on ambient sound information to obtain an ambient noise time-frequency graph, and performing time-frequency analysis on fused sound information to obtain a fused time-frequency graph, includes: The environmental sound signal is decomposed into several time-frequency frames through short-time Fourier transform, each frame corresponds to a frequency component in a time window, and the frequency distribution in each time-frequency frame is obtained through Fourier transform. The frequency distribution of several time-frequency frames is used to generate an environmental noise time-frequency graph, and the noise of different frequencies in the environmental noise time-frequency graph is classified and numbered through the K-means clustering algorithm; Through short-time Fourier transform processing, the fused sound signal of the pump unit when working is also decomposed into several time-frequency frames to generate a fused time-frequency graph. The sounds of different frequencies in the fused time-frequency graph are classified and numbered through the K-means clustering algorithm.

[0025] It is understandable that the present invention uses short-time Fourier transform (STFT) to perform time-frequency analysis on environmental sound information and fused sound information, which can effectively decompose the sound signal into time-frequency frames and extract the frequency components in each frame. By performing time-frequency analysis on the environmental sound and the fused sound when the pump unit is working, the frequency changes in different time windows can be accurately captured, thereby obtaining a more detailed time-frequency diagram. By using the K-means clustering algorithm to classify and number the noise and sound of different frequencies in the time-frequency diagram, the noise from different sources and the working sound of the pump unit can be distinguished, reducing the interference of noise on the sound analysis results. This classification and numbering method provides a clearer basis for subsequent noise removal, sound recognition and fault diagnosis. Through this method, not only the analysis accuracy of the time-frequency diagram is improved, but also the ability to accurately identify the working status of the pump unit in a complex environment is enhanced, ensuring the efficiency and reliability of fault diagnosis.

[0026] In some embodiments of the present application, when calculating the feature vector of each ambient noise according to the ambient noise time-frequency diagram and calculating the feature vector of each fused sound according to the fused time-frequency diagram, the method includes: The feature vector of each environmental noise and the feature vector of each fused sound include Mel frequency cepstral coefficients and spectrum centroid; Apply the Mel filter bank to the spectrum of each frame, convert the spectrum into a spectrum representation on the Mel frequency band, perform logarithmic transformation on the Mel spectrum, and use discrete cosine transform to extract the first 13 Mel frequency cepstral coefficients; Calculate the spectrum centroid of each frame of spectrum; the spectrum centroid is calculated by the following formula: ; Where f represents the frequency of each frame spectrum, |S(f)| represents the amplitude at frequency f; Construct the feature vector sequence Ai=(ai1, ai2, ai3, ..., ai13, Ci) of each environmental noise; wherein Ai represents the feature vector sequence of the i-th environmental noise, Ci represents the spectrum centroid of the i-th environmental noise, aik represents the k-th Mel-frequency cepstrum coefficient of the i-th environmental noise, and k=1, 2, 3, ..., 13; Construct the feature vector sequence Bj=(bj1,bj2,bj3,…,bj13,Cj) of each fused sound; wherein Bj represents the feature vector sequence of the j-th fused sound in the fused sound, Cj represents the spectral centroid of the j-th fused sound, bjk represents the k-th Mel-frequency cepstral coefficient of the j-th fused sound, and k=1,2,3,…,13.

[0027] It can be understood that the present invention can effectively capture the time-frequency characteristics in the sound signal by calculating the feature vectors of the ambient noise and the fused sound and extracting the Mel frequency cepstral coefficients (MFCC) and the spectral centroid. First, the Mel frequency cepstral coefficients can effectively describe the spectral characteristics of the sound signal, reflecting the perception characteristics of the human ear to different frequencies, and are suitable for the analysis of ambient noise and pump unit sound. By applying the Mel filter bank to the spectrum and extracting the first 13 MFCCs using discrete cosine transform, the low-frequency and high-frequency characteristics of the sound can be finely extracted to help distinguish different sound sources. The spectral centroid, as the center of gravity of the frequency distribution, can provide the overall frequency offset information of the sound signal, further improving the distinguishability of the features. Combining these features into a feature vector sequence and establishing models for the ambient noise and fused sound respectively can effectively extract key information from the noise and useful signals. In practical applications, this method helps to improve the ability to identify the sound and noise of the pump unit in a complex environment, reduce the impact of external interference, and thus provide accurate and reliable data support for fault diagnosis and status monitoring.

[0028] In some embodiments of the present application, calculating the similarity information between the feature vectors of each environmental noise and the feature vectors of each fused sound includes: The similarity between the feature vector of each environmental noise and the feature vector of each fused sound is calculated by the following formula: ; In the above formula, S represents the similarity between the feature vector of each ambient noise and the feature vector of each fused sound, aik represents the kth Mel-frequency cepstral coefficient of the ith ambient noise, k=1, 2, 3, ..., 13; bjk represents the kth Mel-frequency cepstral coefficient of the jth fused sound, k=1, 2, 3, ..., 13; Ci represents the spectral centroid of the ith ambient noise, and Cj represents the spectral centroid of the jth fused sound.

[0029] It can be understood that the present invention can accurately evaluate the similarity between the environmental noise and the working sound of the pump unit by calculating the similarity between the feature vector of the environmental noise and the fused sound feature vector, thereby effectively distinguishing noise from useful signals. By using the Mel-frequency cepstral coefficient (MFCC) and the spectral centroid as features, the frequency and time domain characteristics of the sound signal can be fully captured, respectively representing the frequency distribution and overall frequency offset of different sound components. By calculating the similarity between each environmental noise feature vector and the fused sound feature vector, the degree of difference between the noise and the signal can be quantified, which helps to accurately identify and remove noise, significantly improve the effect of noise removal, reduce environmental interference, and ensure that the recognition of the pump unit sound is more accurate and reliable. In practical applications, this similarity calculation provides an effective basis for further fault diagnosis and status monitoring, and can improve the accuracy and real-time performance of judging the operating status of the pump unit.

[0030] In some embodiments of the present application, when removing various environmental noises in the fused time-frequency graph according to the similarity information to obtain the denoised time-frequency graph, the method includes: A similarity threshold is set. If the similarity between the feature vector of the environmental noise and the feature vector of the fused sound is greater than or equal to the similarity threshold, it is determined to remove the fused sound corresponding to the environmental noise in the fused time-frequency graph.

[0031] It can be understood that the present invention sets a similarity threshold to determine whether to remove the environmental noise in the fused time-frequency diagram, which can effectively improve the accuracy and intelligence of noise removal. By calculating the similarity between the environmental noise and the fused sound feature vector of the pump unit and setting a threshold, when the similarity reaches or exceeds the threshold, it indicates that the sound component has a strong noise feature, so it can be determined as environmental noise and removed from the time-frequency diagram. It can automatically identify sound components that are highly similar to environmental noise, avoid the complexity of manual intervention, and improve the level of automation. At the same time, the setting of the threshold can be dynamically adjusted according to the actual situation of different noise environments, ensuring that the denoising process neither misses important information nor effectively eliminates unnecessary noise. This method ensures that the working status signal of the pump unit is not interfered with while denoising, thereby improving the accuracy of voiceprint recognition and helping to achieve more accurate fault diagnosis and status monitoring.

[0032] In some embodiments of the present application, when preliminarily judging whether a pump unit fails according to the denoised time-frequency graph, the method includes: Collect the sound amplitude data of the denoised time-frequency graph and calculate the standard deviation of the sound amplitude data; Set the maximum value of the standard deviation. If the standard deviation is greater than or equal to the maximum value of the standard deviation, it is preliminarily determined that the pump unit has a fault. If the standard deviation is less than the maximum standard deviation, it can be preliminarily determined that the pump unit has not failed.

[0033] It is understandable that the standard deviation of the sound amplitude reflects the degree of fluctuation of the sound signal. A large standard deviation usually indicates instability or abnormal changes in the signal, while a small standard deviation indicates a relatively stable signal, which may indicate that the pump unit is operating normally. By setting the maximum value of the standard deviation, it is possible to automatically identify whether there are signs of faults. When the standard deviation exceeds the set maximum value, it means that the working state of the pump unit has fluctuated greatly, which may indicate that a fault has occurred; otherwise, it indicates that the pump unit is operating smoothly. This method is simple and efficient, avoids excessive reliance on complex algorithms, and can respond quickly in real-time monitoring, providing timely preliminary judgments for fault warnings of the pump unit. Through this standard deviation method, the error rate of judgment can be effectively reduced, ensuring that the status assessment of the pump unit is more accurate and reliable.

[0034] In some embodiments of the present application, if the preliminary determination result is that a fault has occurred, obtaining the sensor data of the pump unit, and determining the state of the pump unit again in combination with the sensor data includes: Set the maximum vibration value and the temperature safety value. If the vibration data of the pump unit is greater than or equal to the maximum vibration value, and / or the temperature data is greater than or equal to the temperature safety value, the pump unit is judged to be faulty again. Otherwise, the state of the pump unit is determined to be a safe state again.

[0035] It is understandable that vibration and temperature are key indicators of pump unit failure. Excessive vibration and temperature are often precursors to mechanical failure, friction or overload. Therefore, when the vibration data exceeds the set maximum vibration value, or the temperature data exceeds the temperature safety value, it can be detected and judged in real time that the pump unit may fail. In this way, the preliminary judgment results can be combined with the sensor data to enhance the accuracy of fault diagnosis and avoid misjudgment caused by a single sound amplitude standard deviation judgment. At the same time, this method has high security because by monitoring key indicators (vibration and temperature), timely warnings can be issued and effective measures can be taken to prevent equipment damage or production accidents. In addition, if the vibration and temperature data do not reach the fault threshold, it can be determined that the pump unit is in a safe state, thereby saving unnecessary maintenance costs. This multi-dimensional state assessment method improves the reliability and operation and maintenance efficiency of the pump unit.

[0036] In some embodiments of the present application, if the preliminary judgment result is that a fault occurs, the sensor data of the pump unit is obtained, and the state of the pump unit is judged again in combination with the sensor data, it also includes: If the initial judgment result and the re-judgment result are inconsistent, a safety warning will be issued; If both the initial judgment result and the re-judgment result indicate a fault, a fault alarm is issued; If both the initial judgment result and the re-judgment result are that no fault has occurred, no safety warning or fault alarm will be issued.

[0037] It is understandable that by comparing the initial judgment result and the re-judgment result after combining the sensor data, the accuracy and reliability of the pump unit fault diagnosis can be effectively improved. When the initial judgment result and the judgment result after combining the sensor data are inconsistent, a safety warning is triggered, which provides an additional warning for the operator to avoid potential risks caused by misjudgment or data abnormalities. At the same time, if both are judged to have a fault, the system will immediately issue a fault alarm to ensure a quick response and take measures such as maintenance or shutdown, thereby reducing the losses caused by the fault. If the initial and re-judgment results are both that no fault has occurred, the system will not make any warnings or alarms, avoiding unnecessary interference and additional maintenance costs. This multi-level judgment mechanism can effectively reduce the probability of false alarms and missed alarms by combining information from different sources, and improve the safety, stability and operation and maintenance efficiency of the pump unit.

[0038] On the other hand, see Figure 2 As shown, the present application also provides a noise processing system for voiceprint recognition, which is used to apply the above-mentioned noise processing method for voiceprint recognition, including: A collection module is configured to collect environmental sound information when the pump unit is not working and fused sound information when the pump unit is working; The time-frequency graph output module is configured to perform time-frequency analysis on the ambient sound information to obtain an ambient noise time-frequency graph, and perform time-frequency analysis on the fused sound information to obtain a fused time-frequency graph; A comparison module is configured to calculate the feature vector of each environmental noise according to the environmental noise time-frequency diagram, calculate the feature vector of each fused sound according to the fused time-frequency diagram, and calculate the similarity information between the feature vector of each environmental noise and the feature vector of each fused sound; The comparison module is further configured to remove various environmental noises in the fused time-frequency graph according to the similarity information to obtain a denoised time-frequency graph; The judgment module is configured to preliminarily judge whether the pump unit has a fault based on the denoised time-frequency diagram. If the preliminary judgment result is a fault, the pump unit sensor data is obtained, and the state of the pump unit is judged again in combination with the sensor data; wherein the sensor data includes temperature data and vibration data.

[0039] It is understandable that the present invention can efficiently and accurately process the noise of the pump unit and realize fault detection by integrating multiple functional modules. First, the acquisition module can respectively obtain the environmental sound information when the pump unit is not working and the fused sound information when it is working, which provides a rich data source for subsequent time-frequency analysis. The time-frequency graph output module generates an environmental noise time-frequency graph and a fused time-frequency graph by performing time-frequency analysis on these sound information, so that the system can effectively distinguish between environmental noise and the working sound of the pump unit in the frequency domain. The comparison module further calculates the characteristic vectors of the environmental noise and the fused sound, and removes the environmental noise in the fused time-frequency graph by similarity calculation to obtain a denoised time-frequency graph, thereby ensuring the clarity and accuracy of the sound signal. The judgment module uses the denoised data to determine whether the pump unit has a fault, and makes another judgment by combining sensor data such as temperature and vibration to ensure the multi-level and reliability of fault detection. In addition, the design of the system allows the accuracy of fault diagnosis to be further improved by supplementing sensor data, thereby reducing the false alarm rate. Overall, the system can not only efficiently process and eliminate noise interference, but also accurately monitor and warn of pump unit faults without affecting performance, greatly improving the operating safety and maintenance efficiency of the pump unit.

[0040] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0041] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0042] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0043] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0044] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A noise processing method for voiceprint recognition, characterized in that: include: Collect the environmental sound information when the pump unit is not working and the fused sound information when the pump unit is working; Performing time-frequency analysis on the ambient sound information to obtain an ambient noise time-frequency graph, and performing time-frequency analysis on the fused sound information to obtain a fused time-frequency graph; Calculating the feature vector of each environmental noise according to the environmental noise time-frequency diagram, calculating the feature vector of each fused sound according to the fused time-frequency diagram, and calculating the similarity information between the feature vector of each environmental noise and the feature vector of each fused sound; According to the similarity information, various environmental noises in the fused time-frequency graph are removed to obtain a denoised time-frequency graph; A preliminary judgment is made based on the denoised time-frequency diagram as to whether the pump unit is faulty. If the preliminary judgment result is that a fault has occurred, the sensor data of the pump unit is obtained, and the state of the pump unit is judged again in combination with the sensor data; wherein the sensor data includes temperature data and vibration data.

2. The noise processing method for voiceprint recognition according to claim 1, characterized in that: The collecting of the environmental sound information when the pump unit is not working and the fused sound information when the pump unit is working includes: A sound collection device is provided on the pump unit, and a sound collection duration is preset; Collecting environmental sound information with the sound collection time as the length; storing the environmental sound information; After the environmental sound information is collected, the fused sound information of the pump unit when it is working is collected, and the fused sound information is divided and stored in units of the sound collection time.

3. The noise processing method for voiceprint recognition according to claim 2, characterized in that: The step of performing time-frequency analysis on the ambient sound information to obtain an ambient noise time-frequency graph and performing time-frequency analysis on the fused sound information to obtain a fused time-frequency graph comprises: Decomposing the ambient sound signal into several time-frequency frames by short-time Fourier transform, each frame corresponds to a frequency component in a time window, obtaining the frequency distribution in each time-frequency frame by Fourier transform, generating an ambient noise time-frequency graph from the frequency distribution of several time-frequency frames, and classifying and numbering the noises of different frequencies in the ambient noise time-frequency graph by K-means clustering algorithm; The fused sound signal of the pump unit at work is also decomposed into several time-frequency frames through short-time Fourier transform processing to generate a fused time-frequency graph. The sounds of different frequencies in the fused time-frequency graph are classified and numbered through the K-means clustering algorithm.

4. The noise processing method for voiceprint recognition according to claim 3, characterized in that: The step of calculating the characteristic vector of each environmental noise according to the environmental noise time-frequency diagram and calculating the characteristic vector of each fused sound according to the fused time-frequency diagram comprises: The feature vectors of each environmental noise and each fused sound feature vector include Mel-frequency cepstral coefficients and spectrum centroids; Apply the Mel filter bank to the spectrum of each frame, convert the spectrum into a spectrum representation on the Mel frequency band, perform logarithmic transformation on the Mel spectrum, and use discrete cosine transform to extract the first 13 Mel frequency cepstral coefficients; Calculate the spectrum centroid of each frame of spectrum; wherein the spectrum centroid is calculated by the following formula: ； Where f represents the frequency of each frame spectrum, |S(f)| represents the amplitude at frequency f; Construct the feature vector sequence Ai=(ai1, ai2, ai3, ..., ai13, Ci) of each environmental noise; wherein Ai represents the feature vector sequence of the i-th environmental noise, Ci represents the spectrum centroid of the i-th environmental noise, aik represents the k-th Mel-frequency cepstrum coefficient of the i-th environmental noise, and k=1, 2, 3, ..., 13; Construct the feature vector sequence Bj=(bj1,bj2,bj3,…,bj13,Cj) of each fused sound; wherein Bj represents the feature vector sequence of the j-th fused sound in the fused sound, Cj represents the spectral centroid of the j-th fused sound, bjk represents the k-th Mel-frequency cepstral coefficient of the j-th fused sound, and k=1,2,3,…,13.

5. The noise processing method for voiceprint recognition according to claim 4, characterized in that: The calculating of the similarity information between the feature vectors of each environmental noise and the feature vectors of each fused sound includes: The similarity between the feature vector of each environmental noise and the feature vector of each fused sound is calculated by the following formula: ； In the above formula, S represents the similarity between the feature vector of each ambient noise and the feature vector of each fused sound, aik represents the kth Mel-frequency cepstral coefficient of the ith ambient noise, k=1, 2, 3, ..., 13; bjk represents the kth Mel-frequency cepstral coefficient of the jth fused sound, k=1, 2, 3, ..., 13; Ci represents the spectral centroid of the ith ambient noise, and Cj represents the spectral centroid of the jth fused sound.

6. The noise processing method for voiceprint recognition according to claim 5, characterized in that: The method of removing the environmental noises in the fused time-frequency graph according to the similarity information to obtain a denoised time-frequency graph includes: A similarity threshold is set, and if the similarity between the feature vector of the environmental noise and the feature vector of the fused sound is greater than or equal to the similarity threshold, it is determined to remove the fused sound corresponding to the environmental noise in the fused time-frequency graph.

7. The noise processing method for voiceprint recognition according to claim 6, characterized in that: The preliminarily judging whether the pump unit fails according to the denoised time-frequency graph includes: Collecting sound amplitude data of the denoised time-frequency graph, and calculating the standard deviation of the sound amplitude data; Setting a maximum standard deviation value, if the standard deviation is greater than or equal to the maximum standard deviation value, preliminarily determining that the pump unit fails; If the standard deviation is less than the maximum standard deviation, it is preliminarily determined that the pump unit has not failed.

8. The noise processing method for voiceprint recognition according to claim 7, characterized in that: If the preliminary determination result is that a fault has occurred, obtaining the sensor data of the pump unit, and determining the state of the pump unit again in combination with the sensor data, includes: Setting a maximum vibration value and a temperature safety value, if the vibration data of the pump unit is greater than or equal to the maximum vibration value, and / or the temperature data is greater than or equal to the temperature safety value, then determining that the pump unit is faulty again; Otherwise, it is determined again that the state of the pump unit is a safe state.

9. The noise processing method for voiceprint recognition according to claim 8, characterized in that: If the preliminary judgment result is that a fault occurs, the sensor data of the pump unit is obtained, and the state of the pump unit is judged again in combination with the sensor data, the method further includes: If the initial judgment result and the re-judgment result are inconsistent, a safety warning will be issued; If both the initial judgment result and the re-judgment result indicate a fault, a fault alarm is issued; If both the initial judgment result and the re-judgment result are that no fault has occurred, no safety warning or fault alarm will be issued.

10. A noise processing system for voiceprint recognition, used for applying the noise processing method for voiceprint recognition according to any one of claims 1 to 9, characterized in that: include: A collection module is configured to collect environmental sound information when the pump unit is not working and fused sound information when the pump unit is working; A time-frequency graph output module is configured to perform time-frequency analysis on the ambient sound information to obtain an ambient noise time-frequency graph, and perform time-frequency analysis on the fused sound information to obtain a fused time-frequency graph; A comparison module is configured to calculate the feature vector of each environmental noise according to the environmental noise time-frequency diagram, calculate the feature vector of each fused sound according to the fused time-frequency diagram, and calculate the similarity information between the feature vector of each environmental noise and the feature vector of each fused sound; The comparison module is further configured to remove various environmental noises in the fused time-frequency graph according to the similarity information to obtain a denoised time-frequency graph; The judgment module is configured to preliminarily judge whether the pump unit has a fault based on the denoised time-frequency diagram. If the preliminary judgment result is a fault, the pump unit sensor data is obtained, and the state of the pump unit is judged again in combination with the sensor data; wherein the sensor data includes temperature data and vibration data.

Images