Voiceprint monitoring system

Abstract

The invention provides a voiceprint monitoring system. The voiceprint monitoring system comprises a sound sensor, a data acquisition module, a security terminal, a security gateway and a server platform, wherein the sound sensor is connected with the data acquisition module through a cable, a power interface on the other side of the data acquisition module is connected with the power adapter, the data acquisition module is connected to the security terminal through a network cable, and the security terminal encrypts and transmits data to the security gateway in a 4G / 5G or wired mode. And the security gateway decrypts the data and then transmits the data to the server platform for analysis, judgment and display. The voiceprint monitoring system is high in voiceprint data acquisition and transmission safety and accuracy, and the acquired data can be remotely transmitted to a unified large server platform to be analyzed, judged and displayed.

Description

Technical Field

[0001] The present invention relates to the technical field of voiceprint monitoring, and in particular to a voiceprint monitoring system. Background Art

[0002] Voiceprint is a sound wave signal with certain information content collected by electroacoustic instruments. The voiceprint of power equipment has a certain periodicity and stability, and contains unique information of the equipment. For example, the sounds emitted by transformers and generators are different, and electroacoustic instruments can identify them by collecting the voiceprints of the two. The sounds emitted by a normally operating power equipment and a faulty power equipment are also different. It is possible to determine whether the power equipment is faulty by collecting the voiceprint signal of the power equipment. The voiceprint signal emitted by the power equipment is different depending on the type of fault. For example, taking the transformer as an example, if a 50HZ "buzzing" voiceprint signal is emitted, it means that the transformer is operating normally; if the transformer emits a "crackling" sound, it means that there is a current short circuit inside the transformer.

[0003] In the prior art, a Chinese patent with publication number CN116705039A discloses an AI-based voiceprint monitoring system and method for electric power equipment, wherein the system includes a voiceprint acquisition module for collecting voiceprint signals; a data processing module for acquiring voiceprint signals and performing preprocessing; a voiceprint recognition module for identifying the target electric power equipment and the voiceprint signal data corresponding to the target electric power equipment; a fault identification module for determining whether the target electric power equipment has a fault and identifying the type of fault; and a fault warning module for issuing an alarm to the manager. The invention identifies multiple fault types and main fault types of the target electric power equipment through wavelet transform. At the same time, the voiceprint acquisition module controls the moving direction and acquisition frequency of the sound sensor, thereby reducing the influence of interference signals and improving the accuracy of identifying fault types. However, the data collected by the sound sensor in the voiceprint monitoring system cannot be transmitted over long distances to a unified large server platform for analysis, judgment and display, and the accuracy of voiceprint data collection and transmission also needs to be further improved. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention proposes a voiceprint monitoring system, which has high accuracy and security in voiceprint data collection and transmission. The collected data can be transmitted over a long distance to a unified large server platform for analysis, judgment and display.

[0005] To realize the above technical scheme, the present invention provides a voiceprint monitoring system, including: a sound sensor for detecting sudden abnormal sounds appearing in the environment; a data acquisition module for converting sound data into feature vectors, extracting short-time Fourier transform, Mel-frequency cepstrum coefficients, impact spectrum and harmonic spectrum of the audio to form multi-channel features, accurately identifying multiple targets, and processing artificial intelligence algorithms; a security terminal for encrypting and transmitting the analyzed data; a security gateway for decrypting the encrypted data and transmitting it to a server platform; a server platform for storing a sample library of the analysis unit, which is used for saving, querying, comparing, importing, exporting and deleting voiceprint maps and audio files; wherein the sound sensor is connected to the data acquisition module through a cable, the power interface on the other side of the data acquisition module is connected to the power adapter, the data acquisition module is connected to the security terminal through a network cable, the security terminal encrypts and transmits the data to the security gateway through 4G / 5G or wired mode, and the security gateway decrypts and transmits the data to the server platform for analysis, judgment and display.

[0006] Preferably, the sound sensor adopts a three-dimensional microphone array sound sensor to collect sudden abnormal sounds at any position in the three-dimensional space, and obtain three-dimensional information of the horizontal azimuth, vertical azimuth and the distance between the sound source and the microphone array reference point, which is then converted into electrical signals and transmitted to the data acquisition module.

[0007] Preferably, the data acquisition module includes: a programmable input and output unit, which is an interface for docking with external circuits to complete the driving and matching requirements of input / output signals under different electrical characteristics, and can adapt to different electrical standards and I / O physical characteristics through flexible software configuration, adjust the size of the driving current, and change the pull-up and pull-down resistors; a configurable logic block, each configurable logic block includes a configurable switch matrix, which is composed of 4 or 6 input selection circuits and triggers, and is used to process combinational logic, shift registers or RAM; a digital clock management module, which is used to provide accurate clock synthesis, reduce jitter, and realize filtering functions; an embedded block RAM, which is used to store data; a data conversion module, which converts sound data into feature vectors, extracts short-time Fourier transform, Mel frequency cepstrum coefficients, impulse spectrum and harmonic spectrum of audio, and forms multi-channel features; an algorithm module, which analyzes and calculates the multi-channel features converted by the data conversion module by loading an artificial intelligence algorithm.

[0008] Preferably, the algorithm module analyzes and calculates the multi-channel features in the following manner:

[0009] S1. Yes Take the initial value s (0) =s, then perform k Gibbs samplings, and then cycle through the subsequent steps S2, S3, and S4 k times;

[0010] S2, the tth cycle, using P(h|s (t-1) ) Sample h (t-1) ;

[0011] S3, the tth cycle, using P(s|h (t-1) ) Sampling (t) ;

[0012] S4. Each sampling actually obtains the approximate value of the partial derivative in a gradient ascent iteration, which is abbreviated as Δw ij , Δa j , Δb i ;

[0013] S5. Continuously iterate and update the three parameters. The iterative update formula is:

[0014] Δw ij =Δw ij +[P(h i =1|s (0) )x (0) -P(h i =1|s (k) x j (k) )]

[0015] Δa j =Δa j +[x j (0) -x j (k) ]

[0016] Δb i =Δb i +[P(h i =1|s (0) )-P(h i =1|s (k) )].

[0017] Preferably, the server platform performs the following operations before comparing the voiceprint graph and the audio file with the sample library of the analysis unit:

[0018] (1) Convert the audio signal into a spectrum through fast Fourier transform;

[0019] (2) Train the audio data through the AI ​​audio separator to learn the characteristics of different sound sources;

[0020] (3) When processing new audio, the AI ​​audio separator analyzes the input spectrogram and predicts the performance of each sound source at different time periods;

[0021] (4) The output separated signal is inversely transformed to be converted back into a time domain audio signal.

[0022] Preferably, the server platform stores the voiceprint sample data in PCM encoding format and in WAV format.

[0023] Preferably, the voiceprint sample storage in the server platform includes three modes: cyclic rolling storage, extraction storage, and manual storage. The audio segments generated by the cycle are extracted and stored as sample data before deletion.

[0024] The beneficial effects of a voiceprint monitoring system provided by the present invention are:

[0025] (1) The present invention transmits the collected voiceprint data to the data acquisition module through a cable through a sound sensor, and then converts the sound data into a feature vector through the data acquisition module, extracts the short-time Fourier transform, Mel-frequency cepstrum coefficients, impact spectrum and harmonic spectrum of the audio to form a multi-channel feature, accurately identifies multiple targets, and processes the artificial intelligence algorithm, and then transmits the analyzed data to the security terminal for encrypted transmission, and then the security terminal encrypts and transmits the data to the security gateway through 4G / 5G or wired methods, and decrypts the encrypted data through the security gateway and transmits it to the server platform, and finally analyzes, judges and displays the decrypted voiceprint map and audio file through the sample library of the analysis unit stored in the server platform, thereby realizing the long-distance secure transmission and accurate analysis, judgment and display of the voiceprint collection data.

[0026] (2) The present invention uses a three-dimensional microphone array sound sensor to accurately collect sudden abnormal sounds at any position in the three-dimensional space, thereby improving the breadth and accuracy of sound collection data.

[0027] (3) Through the design of the data acquisition module, the present invention can accurately extract the short-time Fourier transform, Mel-frequency cepstrum coefficients, impulse spectrum and harmonic spectrum in the voiceprint data to form multi-channel features, and improve the accuracy and robustness of voiceprint data feature extraction through the set intelligent iterative algorithm.

[0028] (4) The server platform of the present invention can adaptively process various types of audio by using an AI audio separator to train audio data, and improve the accuracy and efficiency of separation through learning. By then comparing the audio data with the sample library of the analysis unit, the fault type can be quickly and accurately determined. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary persons in the art without creative work are within the scope of protection of the present invention.

[0031] Embodiment: A voiceprint monitoring system.

[0032] Reference Figure 1 As shown, a voiceprint monitoring system includes: multiple sound sensors 1, data acquisition module 2, security terminal 3, security gateway 4, server platform 5, the position and connection relationship between each component is: each sound sensor 1 is connected to the data acquisition module 2 through a cable, the power interface on the other side of the data acquisition module 2 is connected to the power adapter, the data acquisition module 2 is connected to the security terminal 3 through a network cable, the security terminal 3 encrypts and transmits the data to the security gateway 4 through 4G / 5G or wired mode, and the security gateway 4 decrypts and transmits the data to the server platform 5 for analysis, judgment and display. Among them:

[0033] The sound sensor 1 is used to detect sudden abnormal sounds occurring in the environment. In this embodiment, the sound sensor 1 adopts a three-dimensional microphone array sound sensor to collect sudden abnormal sounds at any position in the three-dimensional space, and obtains three-dimensional information of the horizontal azimuth angle, vertical azimuth angle and distance between the sound source and the microphone array reference point, and then converts the information into electrical signals and transmits them to the data acquisition module 2. The present invention adopts a three-dimensional microphone array sound sensor to accurately collect sudden abnormal sounds at any position in the three-dimensional space, thereby improving the breadth and accuracy of the sound collection data.

[0034] The function of data acquisition module 2 is to convert the pre-processed sound data into feature vectors to facilitate the processing of artificial intelligence algorithms. It extracts short-time Fourier transform, Mel frequency cepstrum coefficients, impact spectrum and harmonic spectrum of audio to form multi-channel features to more accurately identify multiple targets. The data acquisition module 2 includes: a programmable input and output unit, which is an interface for connecting with external circuits to complete the driving and matching requirements of input / output signals under different electrical characteristics. It can adapt to different electrical standards and I / O physical characteristics through flexible software configuration, adjust the size of the driving current, and change the pull-up and pull-down resistors; a configurable logic block, each of which includes a configurable switch matrix, which is composed of 4 or 6 input selection circuits and triggers, and is used to process combinational logic, shift registers or RAM; a digital clock management module, which is used to provide accurate clock synthesis, reduce jitter, and realize filtering functions; an embedded block RAM, which is used to store data; a data conversion module, which converts sound data into feature vectors, extracts short-time Fourier transform, Mel frequency cepstrum coefficients, impulse spectrum and harmonic spectrum of audio, and forms multi-channel features; an algorithm module, which analyzes and calculates the multi-channel features converted by the data conversion module by loading an artificial intelligence algorithm.

[0035] Speech extraction features are extremely important in the recognition process. Selecting appropriate features can improve the final recognition results. The present invention targets the features of the voiceprint of power equipment, and the algorithm module analyzes and calculates the multi-channel features in the following way:

[0036] S1. Yes Take the initial value s (0) =s, then perform k Gibbs samplings, and then cycle through the subsequent steps S2, S3, and S4 k times;

[0037] S2, the tth cycle, using P(h|s (t-1) ) Sample h (t-1) ;

[0038] S3, the tth cycle, using P(s|h (t-1) ) Sampling (t) ;

[0039] S4. Each sampling actually obtains the approximate value of the partial derivative in a gradient ascent iteration, which is abbreviated as Δw ij , Δa j , Δb i ;

[0040] S5. Continuously iterate and update the three parameters. The iterative update formula is:

[0041] Δw ij =Δw ij +[P(h i=1|s (0) )x (0) -P(h i =1|s (k) x j (k) )]

[0042] Δa j =Δa j +[x j (0) -x j (k) ]

[0043] Δb i =Δb i +[P(h i =1|s (0) )-P(h i =1|s (k) )].

[0044] The present invention designs a data acquisition module and targets the characteristics of the voiceprint of power equipment. An iterative algorithm suitable for extracting the voiceprint features of power equipment is loaded into the algorithm module. The short-time Fourier transform, Mel-frequency cepstrum coefficients, impulse spectrum and harmonic spectrum in the voiceprint data of power equipment can be accurately extracted to form multi-channel features. The accuracy and robustness of voiceprint data feature extraction are improved by setting an intelligent iterative algorithm.

[0045] The function of the security terminal 3 is to encrypt the analyzed data for transmission. The voiceprint data needs to be encrypted by the security terminal 3 during long-distance transmission to prevent it from being deciphered.

[0046] The function of the security gateway 4 is to decrypt the encrypted data and transmit it to the server platform 5. Through the cooperation between the security terminal 3 and the security gateway 4, the security of the data transmission process is improved.

[0047] The function of the server platform 5 is to save, query, compare, import, export and delete the transmitted voiceprint graph and audio files, and can use the sample library of the analysis unit stored in the server platform 5 to compare, analyze and display the transmitted voiceprint graph and audio files. The voiceprint sample data storage format uses PCM encoding and is stored in wav format; the sampling rate of each channel is not less than 48kHz, and the bit depth is not less than 16bit; the voiceprint sample storage combines three methods: cyclic rolling storage, extraction storage and manual storage, and extracts audio clips as sample data for storage before deleting the data generated by the cycle.

[0048] The server platform 5 of the present invention performs the following operations before comparing the voiceprint graph and the audio file with the sample library of the analysis unit:

[0049] (1) Convert the audio signal into a spectrum through fast Fourier transform;

[0050] (2) Train the audio data through the AI ​​audio separator to learn the characteristics of different sound sources;

[0051] (3) When processing new audio, the AI ​​audio separator analyzes the input spectrogram and predicts the performance of each sound source at different time periods;

[0052] (4) The output separated signal is inversely transformed to be converted back into a time domain audio signal.

[0053] The server platform 5 of the present invention can adaptively process various types of audio by training audio data using an AI audio separator, and improve the accuracy and efficiency of separation through learning, and then compare it with the sample library of the analysis unit to quickly and accurately determine the type of fault.

[0054] The present invention transmits the collected voiceprint data to the data acquisition module 2 through a cable through the sound sensor 1, and then converts the sound data into a feature vector through the data acquisition module 2, extracts the short-time Fourier transform, Mel-frequency cepstrum coefficients, impact spectrum and harmonic spectrum of the audio to form a multi-channel feature, accurately identifies multiple targets, and processes the artificial intelligence algorithm, and then transmits it to the security terminal 3 to encrypt and transmit the analyzed data, and then the security terminal 3 encrypts and transmits the data to the security gateway 4 through 4G / 5G or wired mode, and decrypts the encrypted data through the security gateway 4 and transmits it to the server platform 5, and finally analyzes, judges and displays the decrypted voiceprint map and audio file through the sample library of the analysis unit stored in the server platform 5, thereby realizing the long-distance secure transmission and accurate analysis, judgment and display of the voiceprint collection data.

[0055] The above description is only a preferred embodiment of the present invention, but the present invention should not be limited to the contents disclosed in the embodiment and the drawings, so any equivalent or modification completed without departing from the spirit disclosed in the present invention shall fall within the scope of protection of the present invention.

Claims

1. A voiceprint monitoring system, characterized in that include: Sound sensor, used to detect sudden abnormal sounds in the environment; The data acquisition module converts sound data into feature vectors, extracts audio short-time Fourier transform, Mel frequency cepstrum coefficients, impulse spectrum and harmonic spectrum, forms multi-channel features, accurately identifies multiple targets, and processes them using artificial intelligence algorithms; The security terminal encrypts and transmits the analyzed data; The security gateway decrypts the encrypted data and transmits it to the server platform; The server platform stores the sample library of the analysis unit and is used to save, query, compare, import, export and delete voiceprints and audio files; Among them, the sound sensor is connected to the data acquisition module through a cable, the power interface on the other side of the data acquisition module is connected to the power adapter, the data acquisition module is connected to the security terminal through a network cable, and the security terminal encrypts and transmits the data to the security gateway via 4G / 5G or wired method. After decryption, the security gateway transmits the data to the server platform for analysis, judgment and display.

2. The voiceprint monitoring system according to claim 1, characterized in that: The sound sensor uses a three-dimensional microphone array sound sensor to collect sudden abnormal sounds at any position in the three-dimensional space, and obtain three-dimensional information of the horizontal azimuth, vertical azimuth and distance between the sound source and the microphone array reference point, which is then converted into electrical signals and transmitted to the data acquisition module.

3. The voiceprint monitoring system according to claim 1, characterized in that: The data acquisition module comprises: Programmable input and output unit, used as an interface to connect with external circuits, to complete the driving and matching requirements of input / output signals under different electrical characteristics. It can adapt to different electrical standards and I / O physical characteristics, adjust the size of the driving current, and change the pull-up and pull-down resistances through flexible software configuration; Configurable logic blocks, each of which contains a configurable switch matrix consisting of 4 or 6 input selection circuits and flip-flops for processing combinational logic, shift registers or RAM; Digital clock management module, used to provide accurate clock synthesis, reduce jitter, and implement filtering functions; Embedded block RAM for data storage; The data conversion module converts the sound data into feature vectors, extracts the short-time Fourier transform, Mel-frequency cepstrum coefficients, impulse spectrum and harmonic spectrum of the audio, and forms multi-channel features; The algorithm module analyzes and calculates the multi-channel features converted by the data conversion module by loading the artificial intelligence algorithm.

4. The voiceprint monitoring system according to claim 3, characterized in that: The algorithm module analyzes and calculates the multi-channel features in the following manner: S1. Yes Take the initial value s (0) =s, then perform k Gibbs samplings, and then cycle through the subsequent steps S2, S3, and S4 k times; S2, the tth cycle, using P(h|s (t-1) ) Sample h (t-1) ; S3, the tth cycle, using P(s|h (t-1) ) Samples (t) ; S4. Each sampling obtains the approximate value of the partial derivative in the gradient ascent iteration process, which is abbreviated as Δw ij , Δa j , Δb i ; S5. Continuously iterate and update the three parameters. The iterative update formula is: Δw ij =Δw ij +[P(h i =1|s (0) )x (0) -P(h i =1|s (k) x j (k) )] Δa j =Δa j +[x j (0) -x j (k) ] Δb i =Δb i +[P(h i =1|s (0) )-P(h i =1|s (k) )]。 5. The voiceprint monitoring system according to claim 1, characterized in that: The server platform will perform the following operations before comparing the voiceprint and audio files with the sample library of the analysis unit: (1) Convert the audio signal into a spectrum through fast Fourier transform; (2) Train the audio data through the AI ​​audio separator to learn the characteristics of different sound sources; (3) When processing new audio, the AI ​​audio separator analyzes the input spectrogram and predicts the performance of each sound source at different time periods; (4) The output separated signal is inversely transformed to be converted back into a time domain audio signal.

6. The voiceprint monitoring system according to claim 1, characterized in that: The server platform stores the voiceprint sample data in PCM encoding and WAV format.

7. The voiceprint monitoring system according to claim 6, characterized in that: The voiceprint sample storage in the server platform includes three modes: cyclic rolling storage, extraction storage, and manual storage. The audio segments generated by the cycle are extracted and stored as sample data before deletion.

Images