Voiceprint recognition method and system based on machine learning

Abstract

The invention relates to the technical field of voiceprint recognition, and discloses a voiceprint recognition method and system based on machine learning, and the method comprises the steps: dividing a to-be-detected region into a plurality of sub-detection regions; voiceprint data are collected, de-noising processing is carried out, a voiceprint spectrum is obtained, and a preliminary recognition identifier is generated; when the identification is a spectrum suspected abnormal identification, extracting a sub-detection area meeting a preset condition, analyzing voiceprint spectrums in other sub-detection areas, and calculating a spectrum data deviation factor; classifying all the frequency spectrum data deviation factors to obtain a frequency spectrum deviation factor set of a plurality of sub-detection areas, and obtaining a suspected abnormal deviation factor and a final suspected abnormal deviation factor; and inputting the final suspected abnormal deviation factor into a pre-trained neural network model, outputting a voiceprint recognition result, and performing early warning reminding. The oil and gas pipeline anomaly detection efficiency and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of voiceprint recognition, and in particular to a voiceprint recognition method and system based on machine learning. Background Art

[0002] With the continuous advancement of industrialization, oil and gas pipelines, as important energy transmission infrastructure, have a direct relationship with the reliability of energy supply and public safety in terms of their operating stability and safety. However, due to aging, corrosion, external damage or changes in the operating environment, pipelines are prone to leakage, cracks and abnormal vibration in long-term operation. Once these problems occur, they may cause environmental pollution, economic losses, and even endanger personal safety. Therefore, it is particularly important to effectively monitor pipelines and identify early faults.

[0003] Acoustic detection technology has become a new detection technology due to its non-contact, real-time and applicable characteristics in complex environments. As a key representation of the operating status of a pipeline, the spectrum characteristics of the soundprint can reflect the flow, vibration status and leakage status of the medium inside the pipeline. However, traditional detection usually conducts a unified analysis of the entire monitoring area, lacks the ability to monitor the pipeline in sections or regions, and is difficult to accurately determine the source of abnormal soundprints. In addition, it relies on manually set thresholds or simple signal processing algorithms, which makes it difficult to deal with the diverse abnormal patterns in pipeline operation.

[0004] Therefore, it is necessary to design a voiceprint recognition method and system based on machine learning to solve the problems existing in current technology. Summary of the invention

[0005] In view of this, the present invention proposes a voiceprint recognition method and system based on machine learning, aiming to solve the problems of low voiceprint recognition accuracy and lack of intelligent analysis capabilities in current pipeline detection technology.

[0006] In one aspect, the present invention proposes a voiceprint recognition method based on machine learning, comprising:

[0007] Determine an area to be detected, and divide the area to be detected into a plurality of sub-detection areas;

[0008] Collect voiceprint data of all acoustic sensors in each of the sub-detection areas within a preset time period, perform denoising on the voiceprint data, obtain a voiceprint spectrum according to the denoised voiceprint data, and generate a preliminary identification mark for the area to be detected according to the voiceprint spectrum, wherein the preliminary identification mark includes a spectrum abnormality mark, a frequency normal mark, and a spectrum suspected abnormality mark;

[0009] When the preliminary identification mark is a suspected abnormal spectrum mark, extract the sub-detection area that meets the preset conditions, analyze the voiceprint spectrum in the remaining sub-detection areas, and calculate the spectrum data deviation factor in the remaining sub-detection areas based on the analysis results, wherein the preset condition is that the voiceprint spectrum in the sub-detection area meets the spectrum range, or the preset condition is that the voiceprint spectrum in the sub-detection area does not meet the spectrum range;

[0010] Extracting all spectrum data deviation factors, numerically sorting all spectrum data deviation factors, classifying all spectrum data deviation factors according to the sorting results, and obtaining a plurality of sub-detection area spectrum deviation factor sets;

[0011] Obtaining a suspected abnormal deviation factor according to the set of spectrum deviation factors of all the sub-detection areas, counting the number of second areas of the sub-detection areas that meet the preset condition, and optimizing the suspected abnormal deviation factor according to the number of the second areas to obtain a final suspected abnormal deviation factor of the area to be detected;

[0012] The final suspected abnormal deviation factor is input into a pre-trained neural network model to output a voiceprint recognition result, and an early warning reminder is issued according to the voiceprint recognition result.

[0013] Furthermore, when generating a preliminary identification mark for the area to be detected according to the voiceprint spectrum, it includes:

[0014] Collecting non-abnormal voiceprint data under rated flow in the area to be detected, obtaining rated voiceprint spectrum according to the non-abnormal voiceprint data, comparing each voiceprint spectrum with the rated voiceprint spectrum, and generating a preliminary identification mark for the area to be detected according to the comparison result;

[0015] When all voiceprint spectra are within the rated voiceprint spectrum range, a normal frequency mark is generated for the area to be detected;

[0016] When all voiceprint spectra are not within the rated voiceprint spectrum range, generating a spectrum anomaly mark for the area to be detected;

[0017] When one or more voiceprint spectra are within the rated voiceprint spectrum range, and one or more voiceprint spectra are not within the rated voiceprint spectrum range, a spectrum suspected abnormality mark is generated for the area to be detected.

[0018] Further, the voiceprint spectrum in the remaining sub-detection area is analyzed, and the spectrum data deviation factor in the remaining sub-detection area is calculated based on the analysis result, including:

[0019] The voiceprint spectra in the remaining sub-detection areas are used as a verification set, the voiceprint spectra in the verification set that are within the rated voiceprint spectrum range are included in a first verification set, and the voiceprint spectra in the verification set that are not within the rated voiceprint spectrum range are included in a second verification set;

[0020] Calculate the difference between each spectrum amplitude in the first verification set and the maximum spectrum amplitude in the rated voiceprint spectrum, extract the first maximum spectrum amplitude difference and the first minimum spectrum amplitude difference in the first verification set, calculate the difference between each spectrum amplitude in the second verification set and the minimum spectrum amplitude in the rated voiceprint spectrum, and extract the second maximum spectrum amplitude difference and the second minimum spectrum amplitude difference in the second verification set;

[0021] The spectrum data deviation factors in the remaining sub-detection areas are calculated according to the first maximum spectrum amplitude difference, the first minimum spectrum amplitude difference, the second maximum spectrum amplitude difference and the second minimum spectrum amplitude difference.

[0022] Further, when calculating the spectrum data deviation factor in the remaining sub-detection area according to the first maximum spectrum amplitude difference, the first minimum spectrum amplitude difference, the second maximum spectrum amplitude difference and the second minimum spectrum amplitude difference, it includes:

[0023] The spectrum data deviation factor in the sub-detection area is calculated according to the following formula:

[0024] ;

[0025] Wherein, P is the spectrum data deviation factor in the sub-detection area, m1 is the number of spectrum amplitudes in the first verification set, fi is the i-th spectrum amplitude difference in the first verification set, fmin is the first minimum spectrum amplitude difference, ymax is the first maximum spectrum amplitude difference, For all The maximum value in, m2 is the number of spectrum amplitudes in the second verification set, bj is the jth spectrum amplitude difference in the second verification set, bmin is the second minimum spectrum amplitude difference, bmax is the second maximum spectrum amplitude difference, For all The maximum value in .

[0026] Furthermore, all the spectrum data deviation factors are classified and processed according to the sorting result to obtain a plurality of sub-detection area spectrum deviation factor sets, including:

[0027] The spectrum data deviation factor in the largest sub-detection area and the spectrum data deviation factors in the adjacent sub-detection areas form an initial deviation factor set, and the spectrum data deviation factor in the third sub-detection area is used as the spectrum data deviation factor in the sub-detection area to be classified;

[0028] Calculate a first sum of the spectrum data deviation factors in two sub-detection areas in the initial deviation factor set, and calculate a first difference between the spectrum data deviation factor in the sub-detection area to be classified and the first sum, and determine whether the first difference is greater than or equal to a preset factor difference; if so, classify the spectrum data deviation factor in the sub-detection area to be classified into the initial deviation factor set; if not, use the initial deviation factor set as a sub-detection area spectrum deviation factor set, and form a second sub-detection area spectrum deviation factor set based on the spectrum data deviation factor in the sub-detection area to be classified and the spectrum data deviation factor in the fourth sub-detection area;

[0029] In this way, multiple sets of spectrum deviation factors of sub-detection areas are obtained.

[0030] Furthermore, when obtaining the suspected abnormal deviation factor according to the spectrum deviation factor set of all the sub-detection areas, it includes:

[0031] Calculating a set and value of spectrum data deviation factors within a sub-detection area in the spectrum deviation factor set of each sub-detection area, and determining a set mean according to the set and value;

[0032] All sets and values ​​smaller than the set mean are classified into a first combination, and all sets and values ​​larger than the set mean are classified into a second combination, and the suspected abnormal deviation factor is calculated based on the first combination and the second combination.

[0033] Further, when calculating the suspected abnormal deviation factor according to the first combination and the second combination, it includes:

[0034] ;

[0035] Among them, Y is the suspected abnormal deviation factor, q1 is the number of set sum values ​​in the first combination, q2 is the number of set sum values ​​in the second combination, hr is the rth set sum value in the second combination, and hmax is the maximum set sum value in the second combination.

[0036] Further, when optimizing the suspected abnormal deviation factor according to the number of the second regions, it includes:

[0037] The number of areas in which the voiceprint spectra in the sub-detection areas that meet the preset conditions all meet the spectrum range is counted and recorded as the first area number, and the difference between the total number of sub-detection areas and the first area number is recorded as the second area number, a quantity ratio is obtained according to the second area number and the first area number, and an optimization coefficient is determined according to the quantity ratio to optimize the suspected abnormal deviation factor, the optimization coefficient is proportional to the quantity ratio, and the value range of the optimization coefficient is (1, 1.2). During optimization, the product of the optimization coefficient and the suspected abnormal deviation factor is obtained to obtain the final suspected abnormal deviation factor of the area to be detected.

[0038] Furthermore, the final suspected abnormal deviation factor is input into a pre-trained neural network model to output a voiceprint recognition result, and an early warning reminder is given according to the voiceprint recognition result, including:

[0039] Collecting historical molding data, wherein the historical molding data includes historical suspected abnormal deviation factors and historical identification results;

[0040] Sampling the historical forming data according to a preset ratio to obtain a training subset and a test subset;

[0041] A pre-selected neural network model is obtained, and the neural network model is iteratively trained according to the training subset, and the iteratively trained neural network model is evaluated according to the test subset to obtain the pre-trained neural network model.

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows: by dividing the area to be detected into multiple sub-detection areas, combining the voiceprint data collected by the acoustic sensor, using denoising processing to improve the signal quality, and generating a preliminary identification mark based on the voiceprint spectrum, a fine-grained analysis of the monitoring area is achieved. For the case where the spectrum is initially identified as suspected abnormal, the abnormal area is further refined by calculating the spectrum data deviation factor, and the spectrum deviation factor set is constructed. A pre-trained neural network model is introduced to identify and analyze the final optimized abnormal deviation factor. Not only is the anti-noise performance and the recognition accuracy of complex abnormal states improved, but also through regional processing and intelligent analysis, accurate positioning and optimized judgment of abnormal areas are achieved, and intelligent early warning reminders are realized. The efficiency and reliability of oil and gas pipeline abnormality detection are improved.

[0043] On the other hand, the present application also provides a voiceprint recognition system based on machine learning, which is used to apply the above-mentioned voiceprint recognition method based on machine learning, including:

[0044] The collecting unit is configured to determine an area to be detected and divide the area to be detected into a plurality of sub-detection areas; collect voiceprint data of all acoustic sensors in each of the sub-detection areas within a preset time period, perform denoising on the voiceprint data, obtain a voiceprint spectrum according to the denoised voiceprint data, and generate a preliminary identification mark for the area to be detected according to the voiceprint spectrum, wherein the preliminary identification mark includes a spectrum abnormality mark, a frequency normal mark, and a spectrum suspected abnormality mark;

[0045] a processing unit configured to, when the preliminary identification mark is a suspected abnormal spectrum mark, extract sub-detection areas that meet preset conditions, analyze the voiceprint spectra in the remaining sub-detection areas, and calculate the spectrum data deviation factors in the remaining sub-detection areas based on the analysis results, wherein the preset condition is that the voiceprint spectra in the sub-detection areas all meet the spectrum range, or the preset condition is that the voiceprint spectra in the sub-detection areas do not meet the spectrum range;

[0046] an integration unit configured to extract all spectrum data deviation factors, numerically sort all spectrum data deviation factors, classify all spectrum data deviation factors according to the sorting result, and obtain a plurality of sub-detection area spectrum deviation factor sets;

[0047] an optimization unit, configured to obtain a suspected abnormal deviation factor according to the set of spectrum deviation factors of all the sub-detection areas, count the number of second areas of the sub-detection areas that meet the preset condition, and optimize the suspected abnormal deviation factor according to the number of the second areas to obtain a final suspected abnormal deviation factor of the area to be detected;

[0048] The early warning unit is configured to input the final suspected abnormal deviation factor into a pre-trained neural network model to output a voiceprint recognition result, and issue an early warning reminder based on the voiceprint recognition result.

[0049] It is understandable that the above-mentioned voiceprint recognition method and system based on machine learning have the same beneficial effects and will not be elaborated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] 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:

[0051] Figure 1 A flow chart of a voiceprint recognition method based on machine learning provided by an embodiment of the present invention;

[0052] Figure 2A functional block diagram of a voiceprint recognition system based on machine learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] 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.

[0054] In some embodiments of the present application, see Figure 1 As shown, a voiceprint recognition method based on machine learning includes:

[0055] S100: Determine an area to be detected, and divide the area to be detected into a plurality of sub-detection areas.

[0056] S200: Collect voiceprint data of all acoustic sensors in each sub-detection area within a preset time period, denoise the voiceprint data, obtain a voiceprint spectrum based on the denoised voiceprint data, and generate a preliminary identification mark for the detection area based on the voiceprint spectrum. The preliminary identification mark includes a spectrum abnormality mark, a frequency normal mark, and a spectrum suspected abnormality mark.

[0057] S300: When the preliminary identification mark is a suspected abnormal mark of the spectrum, the sub-detection area that meets the preset conditions is extracted, the voiceprint spectrum in the remaining sub-detection area is analyzed, and the spectrum data deviation factor in the remaining sub-detection area is calculated based on the analysis result, wherein the preset condition is that the voiceprint spectrum in the sub-detection area meets the spectrum range, or the preset condition is that the voiceprint spectrum in the sub-detection area does not meet the spectrum range.

[0058] S400: extracting all spectrum data deviation factors, numerically sorting all spectrum data deviation factors, classifying all spectrum data deviation factors according to the sorting results, and obtaining a plurality of sub-detection region spectrum deviation factor sets.

[0059] S500: Obtain suspected abnormal deviation factors according to the spectrum deviation factor set of all sub-detection areas, count the number of second areas of the sub-detection areas that meet the preset conditions, and optimize the suspected abnormal deviation factors according to the number of second areas to obtain the final suspected abnormal deviation factors of the area to be detected.

[0060] S600: Input the final suspected abnormal deviation factor into the pre-trained neural network model to output the voiceprint recognition result, and issue an early warning reminder based on the voiceprint recognition result.

[0061] Specifically, in S100 and S200, the area to be detected is divided into multiple sub-detection areas, and the acoustic sensors arranged in each sub-area are used to collect voiceprint data. After the collected data is denoised, a voiceprint spectrum is generated to reflect the energy distribution of each frequency component. Based on the voiceprint spectrum, an identification mark is initially generated and classified into a spectrum abnormality mark, a frequency normal mark, or a spectrum suspected abnormality mark. The precision of monitoring is improved by regional division, and the denoising process improves the signal quality. In S300, when the preliminary identification mark shows that the spectrum is suspected to be abnormal, the sub-detection area that meets the preset conditions (such as all spectra meet or do not meet the spectrum range) is extracted, and the spectra of other sub-areas are deeply analyzed. By calculating the spectrum data deviation factor, the abnormal degree of the remaining sub-areas is quantified, providing a reliable numerical basis for abnormal judgment. In S400, all calculated spectrum data deviation factors are numerically sorted, and multiple spectrum deviation factor sets are generated through classification processing. The classification of abnormal features is realized. In S500 and S600, suspected abnormal deviation factors are further extracted based on the spectrum deviation factor set, and the number of sub-detection areas that meet the preset conditions is counted to optimize the judgment results of abnormal factors. The optimized deviation factors are input into the pre-trained neural network model to output accurate voiceprint recognition results. By combining the learning ability of the model, complex abnormal patterns can be accurately distinguished and intelligent early warning reminders can be issued.

[0062] It is understandable that the combination of regional monitoring and machine learning analysis has improved the accuracy, reliability and intelligence of voiceprint recognition. Through regional division, refined monitoring and abnormal location of pipelines are achieved, and the adaptability to complex operating environments is improved; secondly, denoising and spectrum analysis enhance the ability to identify minor anomalies and reduce the missed detection rate; the introduction of neural network models realizes deep learning and optimized judgment of spectrum data, which can accurately respond to complex abnormal patterns and output reliable results; through multi-step layer-by-layer optimization methods, the accuracy and credibility of recognition results are ensured. It meets the high safety requirements of pipeline operation and reduces the cost of operation and maintenance.

[0063] In some embodiments of the present application, when a preliminary identification mark is generated for the area to be detected based on the voiceprint spectrum, it includes: collecting non-abnormal voiceprint data under the rated flow of the area to be detected, obtaining the rated voiceprint spectrum based on the non-abnormal voiceprint data, comparing each voiceprint spectrum with the rated voiceprint spectrum, and generating a preliminary identification mark for the area to be detected based on the comparison result.

[0064] Specifically, when all voiceprint spectra are within the rated voiceprint spectrum range, a frequency normal mark is generated for the detection area. When all voiceprint spectra are not within the rated voiceprint spectrum range, a spectrum abnormal mark is generated for the detection area. When one or more voiceprint spectra are within the rated voiceprint spectrum range, and one or more voiceprint spectra are not within the rated voiceprint spectrum range, a spectrum suspected abnormal mark is generated for the detection area.

[0065] It is understandable that the introduction of the rated voiceprint spectrum as a benchmark improves the accuracy of the initial identification mark generation. Compared with the judgment method that directly relies on real-time voiceprint data, it effectively avoids the interference of environmental noise or temporary fluctuations on the judgment results, and reduces the occurrence of false alarms and missed alarms. By comparing different voiceprint spectra with the rated spectrum, complex acoustic feature data is converted into clear anomaly categories, which facilitates subsequent intelligent analysis and model optimization. Especially in complex operating environments, this solution can quickly identify anomalies and provide more accurate screening basis for suspected abnormal areas.

[0066] In some embodiments of the present application, when analyzing the voiceprint spectrum in the remaining sub-detection area and calculating the spectrum data deviation factor in the remaining sub-detection area based on the analysis result, it includes: taking the voiceprint spectrum in the remaining sub-detection area as a verification set, including the voiceprint spectrum in the verification set within the rated voiceprint spectrum range into the first verification set, and including the voiceprint spectrum in the verification set that is not within the rated voiceprint spectrum range into the second verification set. Calculate the difference between each spectrum amplitude in the first verification set and the maximum spectrum amplitude in the rated voiceprint spectrum, extract the first maximum spectrum amplitude difference and the first minimum spectrum amplitude difference in the first verification set, calculate the difference between each spectrum amplitude in the second verification set and the minimum spectrum amplitude in the rated voiceprint spectrum, and extract the second maximum spectrum amplitude difference and the second minimum spectrum amplitude difference in the second verification set. Calculate the spectrum data deviation factor in the remaining sub-detection area according to the first maximum spectrum amplitude difference, the first minimum spectrum amplitude difference, the second maximum spectrum amplitude difference and the second minimum spectrum amplitude difference.

[0067] Specifically, when calculating the spectrum data deviation factor in the remaining sub-detection area according to the first maximum spectrum amplitude difference, the first minimum spectrum amplitude difference, the second maximum spectrum amplitude difference, and the second minimum spectrum amplitude difference, it includes:

[0068] The spectrum data deviation factor within the sub-detection area is calculated according to the following formula.

[0069] .

[0070] Wherein, P is the spectrum data deviation factor in the sub-detection area, m1 is the number of spectrum amplitudes in the first verification set, fi is the i-th spectrum amplitude difference in the first verification set, fmin is the first minimum spectrum amplitude difference, ymax is the first maximum spectrum amplitude difference, For all The maximum value in, m2 is the number of spectrum amplitudes in the second verification set, bj is the jth spectrum amplitude difference in the second verification set, bmin is the second minimum spectrum amplitude difference, bmax is the second maximum spectrum amplitude difference, For all The maximum value in .

[0071] It can be understood that by dividing the voiceprint spectrum data into two verification sets and calculating the deviation characteristics for each set, the accuracy and applicability of the spectrum data deviation factor are improved. Compared with the simple overall deviation evaluation method, the difference between normal and abnormal areas can be characterized in a more fine-grained manner, which helps to locate the potential abnormal source of the pipeline in a complex environment.

[0072] In some embodiments of the present application, when all spectrum data deviation factors are classified and processed according to the sorting result to obtain multiple sub-detection area spectrum deviation factor sets, the method includes: the spectrum data deviation factor in the largest sub-detection area and the spectrum data deviation factor in the adjacent sub-detection area form an initial deviation factor set, and the spectrum data deviation factor in the third sub-detection area is used as the spectrum data deviation factor in the sub-detection area to be classified. The first sum of the spectrum data deviation factors in the two sub-detection areas in the initial deviation factor set is calculated, and the first difference between the spectrum data deviation factor in the sub-detection area to be classified and the first sum is calculated, and it is determined whether the first difference is greater than or equal to the preset factor difference. If so, the spectrum data deviation factor in the sub-detection area to be classified is classified into the initial deviation factor set. If not, the initial deviation factor set is used as a sub-detection area spectrum deviation factor set, and the spectrum data deviation factor in the sub-detection area to be classified and the spectrum data deviation factor in the fourth sub-detection area form a second sub-detection area spectrum deviation factor set. By analogy, multiple sub-detection area spectrum deviation factor sets are obtained.

[0073] In some embodiments of the present application, obtaining the suspected abnormal deviation factor according to the spectrum deviation factor set of all sub-detection areas includes:

[0074] The set sum of the spectrum data deviation factors in the sub-detection area in each sub-detection area spectrum deviation factor set is calculated, and the set mean is determined according to the set sum.

[0075] All the set sum values ​​less than the set mean are classified into the first combination, and all the set sum values ​​greater than the set mean are classified into the second combination, and the suspected abnormal deviation factor is calculated based on the first combination and the second combination.

[0076] In some embodiments of the present application, when calculating the suspected abnormal deviation factor according to the first combination and the second combination, it includes:

[0077] .

[0078] Among them, Y is the suspected abnormal deviation factor, q1 is the number of set sum values ​​in the first combination, q2 is the number of set sum values ​​in the second combination, hr is the rth set sum value in the second combination, and hmax is the maximum set sum value in the second combination.

[0079] It can be understood that the refined classification of spectrum data deviation factors and the accurate calculation of suspected abnormal deviation factors are achieved through the method based on spectrum data deviation factor classification and set division, which has advantages in classifying and screening abnormal areas. Specifically, by dynamically calculating the difference between the initial deviation factor set and the factors to be classified, the classification strategy is adaptively adjusted, and the spectrum deviation factor set of the sub-detection area can be flexibly divided according to the actual data distribution, thereby improving the classification accuracy; at the same time, based on the first combination and second combination division of the set and value and the calculation of related characteristic values, it can effectively filter and identify suspected abnormal areas and improve the reliability of fault screening; the suspected abnormal deviation factors are calculated to enhance the reliability of the scheme. It is conducive to improving the accuracy and efficiency of abnormal area screening.

[0080] In some embodiments of the present application, when optimizing the suspected abnormal deviation factor according to the number of second areas, it includes: counting the number of areas in which the voiceprint spectra in the sub-detection areas that meet the preset conditions all meet the spectrum range, recorded as the number of first areas, and recording the difference between the total number of sub-detection areas and the number of first areas as the number of second areas, obtaining a quantity ratio according to the number of second areas and the number of first areas, determining an optimization coefficient according to the quantity ratio to optimize the suspected abnormal deviation factor, the optimization coefficient is proportional to the quantity ratio, and the value range of the optimization coefficient is (1, 1.2). During optimization, the product of the optimization coefficient and the suspected abnormal deviation factor is obtained to obtain the final suspected abnormal deviation factor of the area to be detected.

[0081] It can be understood that the accuracy and stability of the recognition results are improved by dynamically optimizing the calculation model of the suspected abnormal deviation factor. By using the ratio of the number of the first area and the number of the second area, the proportional difference of the abnormal distribution in the area to be detected is fully considered. The adaptive adjustment of the deviation factor is achieved by introducing the optimization coefficient, and the characterization ability of the deviation factor for the actual abnormal area is enhanced. At the same time, the limited range of the optimization coefficient avoids the possibility of error amplification and ensures the reliability of the calculation results. The rationality of the suspected abnormal deviation factor is improved.

[0082] In some embodiments of the present application, the final suspected abnormal deviation factor is input into a pre-trained neural network model to output a voiceprint recognition result, and an early warning reminder is given according to the voiceprint recognition result, including:

[0083] Collect historical forming data, which includes historical suspected abnormal deviation factors and historical identification results.

[0084] The historical forming data is sampled according to a preset ratio to obtain a training subset and a test subset.

[0085] A pre-selected neural network model is obtained, and the neural network model is iteratively trained according to the training subset, and the iteratively trained neural network model is evaluated according to the test subset to obtain the pre-trained neural network model.

[0086] It is understandable that by collecting and using historical data, deep learning of complex voiceprint features is achieved, allowing the model to capture the potential rules between deviation factors and abnormal voiceprints. The pre-trained neural network model has the ability to efficiently process real-time data and can quickly output voiceprint recognition results, providing a basis for early warning of pipeline abnormalities. The training and evaluation mechanism of the solution ensures the generalization ability and stability of the model and avoids recognition bias caused by overfitting. It effectively improves the automation level of voiceprint recognition.

[0087] In the above embodiment, by dividing the area to be detected into multiple sub-detection areas, combining the voiceprint data collected by the acoustic sensor, using denoising processing to improve the signal quality, and generating a preliminary identification mark based on the voiceprint spectrum, a fine-grained analysis of the monitoring area is achieved. For the case where the spectrum is initially identified as suspected abnormal, the abnormal area is further refined by calculating the spectrum data deviation factor, constructing a spectrum deviation factor set, and introducing a pre-trained neural network model to identify and analyze the final optimized abnormal deviation factor. Not only is the noise resistance performance and the recognition accuracy of complex abnormal states improved, but also through regional processing and intelligent analysis, accurate positioning and optimized judgment of abnormal areas are achieved, and intelligent early warning reminders are realized. The efficiency and reliability of oil and gas pipeline abnormality detection are improved.

[0088] In another preferred embodiment based on the above embodiment, refer to Figure 2 As shown, this embodiment provides a voiceprint recognition system based on machine learning, which is used to apply the above-mentioned voiceprint recognition method based on machine learning, including:

[0089] The collection unit is configured to determine the area to be detected and divide the area to be detected into a plurality of sub-detection areas; collect voiceprint data of all acoustic sensors in each sub-detection area within a preset time period, denoise the voiceprint data, obtain a voiceprint spectrum according to the denoised voiceprint data, and generate a preliminary identification mark for the area to be detected according to the voiceprint spectrum, wherein the preliminary identification mark includes a spectrum abnormality mark, a frequency normal mark, and a spectrum suspected abnormality mark;

[0090] The processing unit is configured to extract sub-detection areas that meet preset conditions when the preliminary identification mark is a suspected abnormal spectrum mark, analyze the voiceprint spectra in the remaining sub-detection areas, and calculate the spectrum data deviation factor in the remaining sub-detection areas based on the analysis results, wherein the preset condition is that the voiceprint spectra in the sub-detection areas all meet the spectrum range, or the preset condition is that the voiceprint spectra in the sub-detection areas do not meet the spectrum range;

[0091] An integration unit is configured to extract all spectrum data deviation factors, numerically sort all spectrum data deviation factors, classify all spectrum data deviation factors according to the sorting result, and obtain a plurality of sub-detection area spectrum deviation factor sets;

[0092] an optimization unit, configured to obtain a suspected abnormal deviation factor according to a set of spectrum deviation factors of all sub-detection areas, count the number of second areas of the sub-detection areas that meet the preset conditions, and optimize the suspected abnormal deviation factor according to the number of second areas to obtain a final suspected abnormal deviation factor of the area to be detected;

[0093] The early warning unit is configured to input the final suspected abnormal deviation factor into a pre-trained neural network model to output a voiceprint recognition result, and issue an early warning reminder based on the voiceprint recognition result.

[0094] It is understandable that by dividing the area to be detected into multiple sub-detection areas, combining the voiceprint data collected by the acoustic sensor, using denoising processing to improve the signal quality, and generating a preliminary identification mark based on the voiceprint spectrum, a fine-grained analysis of the monitoring area is achieved. For the case where the spectrum is initially identified as suspected abnormal, the abnormal area is further refined by calculating the spectrum data deviation factor, constructing a spectrum deviation factor set, and introducing a pre-trained neural network model to identify and analyze the final optimized abnormal deviation factor. Not only is the noise resistance performance and the recognition accuracy of complex abnormal states improved, but also through regional processing and intelligent analysis, accurate positioning and optimized judgment of abnormal areas are achieved, and intelligent early warning reminders are realized. Improve the efficiency and reliability of oil and gas pipeline abnormality detection.

[0095] 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.

[0096] 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.

[0097] 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.

[0098] 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.

[0099] 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 voiceprint recognition method based on machine learning, characterized in that: include: Determine an area to be detected, and divide the area to be detected into a plurality of sub-detection areas; Collect voiceprint data of all acoustic sensors in each of the sub-detection areas within a preset time period, perform denoising on the voiceprint data, obtain a voiceprint spectrum according to the denoised voiceprint data, and generate a preliminary identification mark for the area to be detected according to the voiceprint spectrum, wherein the preliminary identification mark includes a spectrum abnormality mark, a frequency normal mark, and a spectrum suspected abnormality mark; When the preliminary identification mark is a suspected abnormal spectrum mark, extract the sub-detection area that meets the preset conditions, analyze the voiceprint spectrum in the remaining sub-detection areas, and calculate the spectrum data deviation factor in the remaining sub-detection areas based on the analysis results, wherein the preset condition is that the voiceprint spectrum in the sub-detection area meets the spectrum range, or the preset condition is that the voiceprint spectrum in the sub-detection area does not meet the spectrum range; Extracting all spectrum data deviation factors, numerically sorting all spectrum data deviation factors, classifying all spectrum data deviation factors according to the sorting results, and obtaining a plurality of sub-detection area spectrum deviation factor sets; Obtaining a suspected abnormal deviation factor according to the set of spectrum deviation factors of all the sub-detection areas, counting the number of second areas of the sub-detection areas that meet the preset condition, and optimizing the suspected abnormal deviation factor according to the number of the second areas to obtain a final suspected abnormal deviation factor of the area to be detected; The final suspected abnormal deviation factor is input into a pre-trained neural network model to output a voiceprint recognition result, and an early warning reminder is issued according to the voiceprint recognition result.

2. The voiceprint recognition method based on machine learning according to claim 1, characterized in that: When a preliminary identification mark is generated for the area to be detected according to the voiceprint spectrum, it includes: Collecting non-abnormal voiceprint data under rated flow in the area to be detected, obtaining rated voiceprint spectrum according to the non-abnormal voiceprint data, comparing each voiceprint spectrum with the rated voiceprint spectrum, and generating a preliminary identification mark for the area to be detected according to the comparison result; When all voiceprint spectra are within the rated voiceprint spectrum range, a frequency normal mark is generated for the area to be detected; When all voiceprint spectra are not within the rated voiceprint spectrum range, generating a spectrum anomaly mark for the area to be detected; When one or more voiceprint spectra are within the rated voiceprint spectrum range, and one or more voiceprint spectra are not within the rated voiceprint spectrum range, a spectrum suspected abnormality mark is generated for the area to be detected.

3. The voiceprint recognition method based on machine learning according to claim 2, characterized in that: Analyzing the voiceprint spectrum in the remaining sub-detection area and calculating the spectrum data deviation factor in the remaining sub-detection area based on the analysis result includes: The voiceprint spectra in the remaining sub-detection areas are used as a verification set, the voiceprint spectra in the verification set that are within the rated voiceprint spectrum range are included in a first verification set, and the voiceprint spectra in the verification set that are not within the rated voiceprint spectrum range are included in a second verification set; Calculate the difference between each spectrum amplitude in the first verification set and the maximum spectrum amplitude in the rated voiceprint spectrum, extract the first maximum spectrum amplitude difference and the first minimum spectrum amplitude difference in the first verification set, calculate the difference between each spectrum amplitude in the second verification set and the minimum spectrum amplitude in the rated voiceprint spectrum, and extract the second maximum spectrum amplitude difference and the second minimum spectrum amplitude difference in the second verification set; The spectrum data deviation factors in the remaining sub-detection areas are calculated according to the first maximum spectrum amplitude difference, the first minimum spectrum amplitude difference, the second maximum spectrum amplitude difference and the second minimum spectrum amplitude difference.

4. The voiceprint recognition method based on machine learning according to claim 3 is characterized in that: When calculating the spectrum data deviation factor in the remaining sub-detection area according to the first maximum spectrum amplitude difference, the first minimum spectrum amplitude difference, the second maximum spectrum amplitude difference and the second minimum spectrum amplitude difference, it includes: The spectrum data deviation factor in the sub-detection area is calculated according to the following formula: ； Wherein, P is the spectrum data deviation factor in the sub-detection area, m1 is the number of spectrum amplitudes in the first verification set, fi is the i-th spectrum amplitude difference in the first verification set, fmin is the first minimum spectrum amplitude difference, ymax is the first maximum spectrum amplitude difference, For all The maximum value in, m2 is the number of spectrum amplitudes in the second verification set, bj is the jth spectrum amplitude difference in the second verification set, bmin is the second minimum spectrum amplitude difference, bmax is the second maximum spectrum amplitude difference, For all The maximum value in .

5. The voiceprint recognition method based on machine learning according to claim 4 is characterized in that: Classifying and processing all the spectrum data deviation factors according to the sorting result to obtain a plurality of sub-detection area spectrum deviation factor sets includes: The spectrum data deviation factor in the largest sub-detection area and the spectrum data deviation factors in the adjacent sub-detection areas form an initial deviation factor set, and the spectrum data deviation factor in the third sub-detection area is used as the spectrum data deviation factor in the sub-detection area to be classified; Calculate a first sum of the spectrum data deviation factors in two sub-detection areas in the initial deviation factor set, and calculate a first difference between the spectrum data deviation factor in the sub-detection area to be classified and the first sum, and determine whether the first difference is greater than or equal to a preset factor difference; if so, classify the spectrum data deviation factor in the sub-detection area to be classified into the initial deviation factor set; if not, use the initial deviation factor set as a sub-detection area spectrum deviation factor set, and form a second sub-detection area spectrum deviation factor set based on the spectrum data deviation factor in the sub-detection area to be classified and the spectrum data deviation factor in the fourth sub-detection area; In this way, multiple sets of spectrum deviation factors of sub-detection areas are obtained.

6. The voiceprint recognition method based on machine learning according to claim 1, characterized in that: When obtaining the suspected abnormal deviation factor according to the spectrum deviation factor set of all the sub-detection areas, it includes: Calculating a set and value of spectrum data deviation factors within a sub-detection area in the spectrum deviation factor set of each sub-detection area, and determining a set mean according to the set and value; All sets and values ​​smaller than the set mean are classified into a first combination, and all sets and values ​​larger than the set mean are classified into a second combination, and the suspected abnormal deviation factor is calculated based on the first combination and the second combination.

7. The voiceprint recognition method based on machine learning according to claim 6, characterized in that: When calculating the suspected abnormal deviation factor according to the first combination and the second combination, it includes: ； Among them, Y is the suspected abnormal deviation factor, q1 is the number of set sum values ​​in the first combination, q2 is the number of set sum values ​​in the second combination, hr is the rth set sum value in the second combination, and hmax is the maximum set sum value in the second combination.

8. The voiceprint recognition method based on machine learning according to claim 7, characterized in that: When the suspected abnormal deviation factor is optimized according to the number of the second regions, it includes: The number of areas in which the voiceprint spectra in the sub-detection areas that meet the preset conditions all meet the spectrum range is counted and recorded as the first area number, and the difference between the total number of sub-detection areas and the first area number is recorded as the second area number, a quantity ratio is obtained according to the second area number and the first area number, and an optimization coefficient is determined according to the quantity ratio to optimize the suspected abnormal deviation factor, the optimization coefficient is proportional to the quantity ratio, and the value range of the optimization coefficient is (1, 1.2). During optimization, the product of the optimization coefficient and the suspected abnormal deviation factor is obtained to obtain the final suspected abnormal deviation factor of the area to be detected.

9. The voiceprint recognition method based on machine learning according to claim 8, characterized in that: The final suspected abnormal deviation factor is input into a pre-trained neural network model to output a voiceprint recognition result, and an early warning reminder is issued according to the voiceprint recognition result, including: Collecting historical molding data, wherein the historical molding data includes historical suspected abnormal deviation factors and historical identification results; Sampling the historical forming data according to a preset ratio to obtain a training subset and a test subset; A pre-selected neural network model is obtained, and the neural network model is iteratively trained according to the training subset, and the iteratively trained neural network model is evaluated according to the test subset to obtain the pre-trained neural network model.

10. A voiceprint recognition system based on machine learning, used for applying the voiceprint recognition method based on machine learning as claimed in any one of claims 1 to 9, characterized in that: include: A collection unit is configured to determine an area to be detected and divide the area to be detected into a plurality of sub-detection areas; Collect voiceprint data of all acoustic sensors in each of the sub-detection areas within a preset time period, perform denoising on the voiceprint data, obtain a voiceprint spectrum according to the denoised voiceprint data, and generate a preliminary identification mark for the area to be detected according to the voiceprint spectrum, wherein the preliminary identification mark includes a spectrum abnormality mark, a frequency normal mark, and a spectrum suspected abnormality mark; a processing unit configured to, when the preliminary identification mark is a suspected abnormal spectrum mark, extract sub-detection areas that meet preset conditions, analyze the voiceprint spectra in the remaining sub-detection areas, and calculate the spectrum data deviation factors in the remaining sub-detection areas based on the analysis results, wherein the preset condition is that the voiceprint spectra in the sub-detection areas all meet the spectrum range, or the preset condition is that the voiceprint spectra in the sub-detection areas do not meet the spectrum range; an integration unit configured to extract all spectrum data deviation factors, numerically sort all spectrum data deviation factors, classify all spectrum data deviation factors according to the sorting result, and obtain a plurality of sub-detection area spectrum deviation factor sets; an optimization unit, configured to obtain a suspected abnormal deviation factor according to the set of spectrum deviation factors of all the sub-detection areas, count the number of second areas of the sub-detection areas that meet the preset condition, and optimize the suspected abnormal deviation factor according to the number of the second areas to obtain a final suspected abnormal deviation factor of the area to be detected; The early warning unit is configured to input the final suspected abnormal deviation factor into a pre-trained neural network model to output a voiceprint recognition result, and issue an early warning reminder based on the voiceprint recognition result.

Images