Voice control system for groundwater sampling equipment based on voiceprint recognition

Abstract

The present application relates to the technical field of speech recognition, in particular to a groundwater sampling device voice control system based on voiceprint recognition. The system comprises an audio acquisition module, a physical state sensor group, a dynamic adaptive filter, a voiceprint recognition module, an instruction logic verification module and a device execution mechanism. The physical state sensor group obtains the water pump rotation speed frequency, the capstan motor current value and the hydrological environment background noise spectrum. The dynamic adaptive filter generates a reverse cancellation signal based on the above physical state data, and outputs a pure voice signal by reducing noise to the original voice signal. The voiceprint recognition module extracts features, compares voiceprints and analyzes voice control instructions. The present application eliminates the masking effect of mechanical noise and hydrological noise on voice signals, improves the accuracy of voiceprint feature extraction in a strong noise environment, blocks conflicting instruction execution through time sequence interlocking comparison, avoids mechanical structure overload, and ensures the physical safety of sampling operations.

Description

Technical Field

[0001] This invention relates to the field of speech recognition technology, and more specifically to a voice control system for groundwater sampling equipment based on voiceprint recognition. Background Technology

[0002] Groundwater sampling equipment is typically controlled using physical buttons or a general-purpose voice recognition module. The general-purpose voice recognition module has a built-in fixed noise reduction model and voiceprint recognition algorithm. In actual operation, the equipment collects the operator's voice through a microphone, removes some ambient white noise through a fixed filter, extracts voice features, and compares them with a pre-set voiceprint database. If the comparison is successful, the voice semantics are parsed to generate control commands, which are then directly sent to the controllers of the winch motor and water pump to execute the actions.

[0003] The existing technologies described above employ fixed noise reduction models, which cannot adapt to the non-stationary, time-varying strong background noise generated by the superposition of water pumping, winch rotation, and groundwater flow disturbances at the groundwater sampling site. This leads to distortion in voiceprint feature extraction and triggers command rejection. Furthermore, the voice control system only focuses on the semantic content of the speech, detached from the specific mechanical and physical operating state of the sampling equipment. When the equipment receives a lowering command during the physical process of lifting the sampling tube, the control system will still directly output an execution signal, causing mechanical overload or sampling tube breakage. Existing voice control systems cannot adaptively process the dynamic coupling relationship between the underlying mechanical operating state of the sampling equipment and the complex hydrological noise environment, resulting in recognition distortion under strong noise and the potential for mechanical action conflicts that deviate from physical constraints. Summary of the Invention

[0004] The purpose of this invention is to provide a voice control system for groundwater sampling equipment based on voiceprint recognition, which can effectively solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The voice control system for groundwater sampling equipment based on voiceprint recognition includes an audio acquisition module, a physical state sensor group, a dynamic adaptive filter, a voiceprint recognition module, an instruction logic verification module, and an equipment execution mechanism. The audio acquisition module and the physical state sensor group synchronously acquire the original voice signal of the operator and the real-time physical state data of the groundwater sampling equipment. The real-time physical state data includes the pump speed frequency, the winch motor current value and the background noise spectrum of the hydrological environment. The dynamic adaptive filter receives the original speech signal and real-time physical state data, generates an inverse cancellation signal based on the pump speed frequency and the background noise spectrum of the hydrological environment, and performs noise reduction processing on the original speech signal to output a clean speech signal. The voiceprint recognition module extracts features from the clean speech signal, compares them with the pre-stored authorized personnel voiceprint template library, outputs the identity matching result, and performs semantic parsing on the clean speech signal to output voice control commands. The instruction logic verification module receives voice control instructions and real-time physical status data, determines the current physical status of the sampling tube based on the winch motor current value, and performs a timing interlock logic comparison between the action required by the voice control instructions and the current physical status of the sampling tube. When the identity matching result is authorized and the voice control command and the current physical state of the sampling tube conform to the preset security timing logic, the device actuator receives the drive control signal output by the command logic verification module and responds to the drive control signal to change the current value of the winch motor and the speed and frequency of the water pump.

[0006] Preferably, the dynamic adaptive filter includes a mechanical noise estimation sub-filter and a hydrological environment noise estimation sub-filter; The mechanical noise estimation sub-filter generates a mechanical noise prediction signal by querying a pre-stored mapping table of water pump speed and mechanical vibration noise transfer function based on the pump speed frequency. The hydrological environment noise estimation sub-filter extracts the energy distribution characteristics of the main frequency band of the water flow based on the hydrological environment background noise spectrum, and generates a hydrological environment noise prediction signal. The dynamic adaptive filter linearly superimposes the mechanical noise prediction signal and the hydrological environment noise prediction signal to generate the inverse cancellation signal, and adds the inverse cancellation signal to the original speech signal in the time domain to output the clean speech signal.

[0007] Preferably, the physical state sensor group includes a hydrophone array disposed outside the inlet of the groundwater sampling device; The hydrophone array collects the vibration signal of the water body in the well where the groundwater sampling equipment is located, and converts the water body vibration signal into a hydroacoustic electrical signal. The physical state sensor group performs a fast Fourier transform on the underwater acoustic signal to extract the frequency domain energy distribution matrix within a preset frequency band, and uses the frequency domain energy distribution matrix as the background noise spectrum of the hydrological environment. The physical state sensor group encapsulates the hydrological environment background noise spectrum, the water pump speed frequency, and the winch motor current value into a data frame and sends it to the dynamic adaptive filter.

[0008] Preferably, the voiceprint recognition module extracts the Mel frequency cepstral coefficient features and the fundamental frequency envelope features of the pure speech signal, and concatenates the Mel frequency cepstral coefficient features and the fundamental frequency envelope features into a high-dimensional voiceprint feature vector; The voiceprint recognition module calculates the cosine similarity between the high-dimensional voiceprint feature vector and each template vector in the authorized operator's voiceprint template library, and selects the maximum cosine similarity as the identity matching result. The voiceprint recognition module inputs the clean speech signal into a pre-trained end-to-end speech recognition network, outputs a text sequence, and extracts the voice control command from the text sequence according to a preset keyword hitting rule.

[0009] Preferably, the instruction logic verification module pre-stores a correspondence table between the current value range of the winch motor and the physical state of the sampling tube, and the correspondence table includes the lifting current range, the lowering current range and the quiescent current range. The instruction logic verification module queries the current range to which the current value of the winch motor belongs in the corresponding relationship table, and determines the physical state of the sampling tube corresponding to the current range as the current physical state. The instruction logic verification module parses the target action contained in the voice control instruction, queries the pre-stored directed graph of action state transitions, and determines whether there is a directed edge connection when executing the target action from the current physical state. If there is a directed edge connection, it is determined that it conforms to the safe timing logic.

[0010] Preferably, the device actuator includes a winch motor drive circuit and a water pump frequency converter; The device actuator unpacks the drive control signal and extracts the target winch motor current setpoint and the target water pump speed frequency setpoint. The winch motor drive circuit receives the target winch motor current setpoint, adjusts the pulse width modulation duty cycle of the internal insulated gate bipolar transistor, and adjusts the winch motor current value to the target winch motor current setpoint. The water pump frequency converter receives the target water pump speed frequency setpoint, adjusts the three-phase AC power frequency output to the water pump motor, and adjusts the water pump speed frequency to the target water pump speed frequency setpoint.

[0011] Preferably, the dynamic adaptive filter further includes an error feedback regulator; The error feedback regulator acquires the clean speech signal, calculates the short-time energy and short-time zero-crossing rate of the clean speech signal, compares the short-time energy with a preset energy threshold, and compares the short-time zero-crossing rate with a preset zero-crossing rate threshold. When the short-time energy is greater than the preset energy threshold and the short-time zero-crossing rate is greater than the preset zero-crossing threshold, the error feedback regulator generates a filter offset compensation coefficient. The mechanical noise estimation sub-filter and the hydrological environment noise estimation sub-filter update the weighting coefficients of the mechanical noise transfer function mapping table and the energy distribution characteristics of the main frequency band of the water flow based on the filter offset compensation coefficient.

[0012] Preferably, the hydrophone array includes multiple hydrophone nodes that are evenly spaced along the depth direction of the wellbore; The physical state sensor group acquires the hydroacoustic signals collected by each hydrophone node, calculates the cross-correlation function of the hydroacoustic signals between two adjacent hydrophone nodes, and calculates the spatial propagation direction vector of water noise based on the peak time delay of the cross-correlation function. The physical state sensor group constructs a beamforming guide vector based on the spatial propagation direction vector, and uses the beamforming guide vector to perform array weighted summation of the hydroacoustic electrical signals of each hydrophone node to suppress pump coupling noise propagating along the wellbore axis, and outputs the suppressed water vibration signal.

[0013] Preferably, the instruction logic verification module is equipped with a state transition delay register; When it is determined that the target action is performed from the current physical state with a directed edge connection, the instruction logic verification module obtains the duration for which the winch motor current value is maintained in the current range, and writes the duration into the state transition delay register. The instruction logic verification module queries the pre-stored mapping table of physical states and minimum stable durations, extracts the minimum stable duration corresponding to the current physical state, and compares whether the duration is greater than the minimum stable duration. If it is greater, it finally determines that it conforms to the safety timing logic and outputs the drive control signal.

[0014] Preferably, the voiceprint recognition module has a pre-stored syntax tree for groundwater sampling instructions; The syntax tree contains device action nodes and device parameter nodes. The device action nodes include lift, drop, and stop. The device parameter nodes include depth values ​​and flow values. The voiceprint recognition module inputs the text sequence into a finite state sensor constructed based on the syntax tree. According to the state transition path of the finite state sensor, it performs legality verification and error correction and replacement on the characters in the text sequence, filters out illegal characters that do not conform to the node connection relationship of the syntax tree, outputs the corrected target text sequence, and extracts the voice control command from the target text sequence according to the keyword hitting rule.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention acquires the pump speed frequency, winch motor current value, and hydrological environment background noise spectrum through a physical state sensor group. A dynamic adaptive filter generates an inverse cancellation signal based on the above physical state data to reduce noise in the original speech signal. This changes the traditional fixed noise reduction model, which cannot adapt to the superposition of equipment mechanical noise and hydrological environment noise, and reduces the masking effect of non-stationary time-varying noise on the speech signal. The instruction logic verification module determines the current physical state of the sampling tube based on the winch motor current value and compares the actions required by the voice control command with the current physical state through time-series interlocking logic. This changes the existing technology's method of directly executing instructions without considering the equipment's mechanical state, blocking instruction outputs that contradict the current physical logic and avoiding mechanical overload.

[0016] 2. The hydrophone array obtains the spatial propagation direction vector by calculating the cross-correlation function of the hydroacoustic signals of adjacent nodes, and uses the beamforming steering vector for array weighted summation to suppress pump coupling noise propagating along the borehole axis; the error feedback regulator generates filter offset compensation coefficients based on the short-time energy and short-time zero-crossing rate of the clean speech signal, updates the weight coefficients of the sub-filters, and corrects the offset error in the filtering process; the state transition delay register ensures the command response after the smooth transition of the equipment state by comparing the duration of the winch motor current maintained in a specific range with the minimum stable duration; the finite state sensor, combined with a dedicated instruction syntax tree, performs legality verification and error correction on the text sequence, filtering out illegal characters and eliminating interference from non-standard instructions. Attached Figure Description

[0017] Figure 1 This is the overall system control flowchart of the present invention; Figure 2 This is a flowchart of the dynamic adaptive filtering and noise reduction process of the present invention; Figure 3 This is a flowchart of the hydrological environmental noise spectrum extraction process of the present invention; Figure 4 This is a flowchart of the voiceprint recognition and command parsing process of the present invention; Figure 5 This is a flowchart of the instruction logic verification and timing interlocking of the present invention; Figure 6 This is a flowchart of the device actuator drive control of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] Please refer to Figure 1 This embodiment provides a voice control system for groundwater sampling equipment based on voiceprint recognition, including an audio acquisition module, a physical state sensor group, a dynamic adaptive filter, a voiceprint recognition module, an instruction logic verification module, and an equipment execution mechanism. The audio acquisition module and the physical state sensor group employ a synchronous acquisition mechanism triggered by the same source clock to synchronously acquire the operator's original voice signal and the real-time physical state data of the groundwater sampling equipment. The real-time physical state data includes the pump speed frequency, winch motor current value, and the hydrological environment background noise spectrum. The dynamic adaptive filter receives the synchronously output original voice signal and real-time physical state data. Based on the pump speed frequency and hydrological environment background noise spectrum in the real-time physical state data, it generates an inverse cancellation signal with the opposite phase to the superimposed noise in the original voice signal. After time-domain noise reduction processing, a clean voice signal is output. The voiceprint recognition module receives the clean voice signal, extracts voiceprint features, and performs template matching with a pre-stored authorized operator voiceprint template library, outputting the identity matching result. Simultaneously, the voiceprint recognition module performs semantic parsing processing on the clean voice signal and outputs the corresponding voice control commands. The instruction logic verification module receives voice control commands from the voiceprint recognition module and real-time physical status data synchronously output by the physical status sensor group. Based on the winch motor current value in the real-time physical status data, it determines the current physical status of the sampling tube. It then performs a timing interlock logic comparison between the device actions required by the voice control command and the current physical status of the sampling tube, outputting the comparison result. When the identity matching result indicates an authorized operator, and the voice control command and the current physical status of the sampling tube conform to the preset safety timing logic, the device actuator receives the drive control signal output by the instruction logic verification module. In response to the drive control signal, it adjusts the output parameters, changing the winch motor current value and the water pump speed frequency, thus completing the voice control process for the groundwater sampling equipment.

[0020] Specifically, the synchronous acquisition mechanism between the audio acquisition module and the physical state sensor group uses a common crystal oscillator to provide the clock signal. The sampling clock of the audio acquisition module and the sampling clock of the physical state sensor group are output by frequency division from the same crystal oscillator. The sampling frequency of the audio acquisition module is set to an integer multiple of the sampling frequency of the physical state sensor group. After each physical state data sampling cycle is completed, a frame synchronization signal for audio acquisition is triggered synchronously to ensure that each frame of the original voice signal has a unique timestamp identifier corresponding to the real-time physical state data at the corresponding moment, avoiding phase deviation in noise reduction processing caused by timing asynchrony. The pump speed frequency is obtained by calculating the number of zero-crossings per unit time using the zero-crossing signal of the three-phase AC power of the pump motor in the sampling device. The winch motor current value is obtained by analog-to-digital conversion of the voltage signal across the series sampling resistor in the winch motor drive circuit. The background noise spectrum of the hydrological environment is obtained by frequency domain conversion of the water vibration signal in the well.

[0021] A dynamic adaptive filter receives the original speech signal and synchronized real-time physical state data, extracts the pump speed frequency and hydrological environment background noise spectrum, performs time-domain prediction on the mechanical noise component corresponding to the pump speed frequency and the environmental noise component corresponding to the hydrological environment background noise spectrum, and generates an inverse cancellation signal with the same amplitude but opposite phase to the noise component superimposed on the original speech signal. This inverse cancellation signal is then linearly superimposed on the original speech signal in the time domain to cancel out the noise component in the original speech signal, outputting a clean speech signal. The time-domain expression of the original speech signal is:

[0022] in, This is the original speech signal at the nth sampling point. For the clean target speech component at the nth sampling point, The superimposed noise component is the noise component at the nth sampling point. The superimposed noise component includes the mechanical noise component generated by the operation of the water pump and the hydrological environmental noise component generated by the vibration of the water in the well.

[0023] The time-domain expression of the inverse cancellation signal generated by the dynamic adaptive filter is: Inverse cancellation signal and superimposed noise component The time-domain expression of the pure speech signal output after time-domain superposition processing, where the amplitudes are the same but the phases are opposite, is:

[0024] in, For the clean speech signal at the nth sampling point, when hour, and The time-domain waveforms are consistent, thus completing the noise cancellation process.

[0025] The voiceprint recognition module receives a clean speech signal and first performs preprocessing, including pre-emphasis, framing, and windowing, to extract voiceprint features. These features are then matched against each template in a pre-stored authorized operator voiceprint template library to obtain a matching score. This score is compared to a preset matching threshold. If the matching score is greater than or equal to the threshold, the output identity matching result is "authorized"; otherwise, it is "unauthorized." Simultaneously, the voiceprint recognition module performs speech recognition processing on the preprocessed clean speech signal, converting it into a corresponding text sequence. Based on preset keyword matching rules, it extracts action commands and parameter commands related to the control of the groundwater sampling equipment from the text sequence and combines them into a complete voice control command. During voiceprint feature extraction, Mel-frequency cepstral coefficients are used as the core feature parameter. Following the standard conversion relationship between Mel-frequency and linear frequency, the linear frequency domain speech signal is converted to the Mel-frequency domain, which conforms to human hearing characteristics, and the corresponding Mel-frequency cepstral coefficient features are extracted.

[0026] The instruction logic verification module receives the winch motor current value from the synchronized real-time physical status data, queries the pre-stored correspondence between the current range and the physical state of the sampling tube, and determines the current physical state of the sampling tube, which includes lifting, lowering, and stationary states. Simultaneously, the instruction logic verification module parses the target actions contained in the voice control commands, including lifting the sampling tube, lowering the sampling tube, stopping the equipment, and starting / stopping the water pump. The module queries the pre-stored directed graph of action state transitions to determine if a valid state transition path exists for executing the target action from the current physical state. If a valid state transition path exists, the module determines that the voice control command and the current physical state of the sampling tube conform to the preset safety timing logic; otherwise, it determines that they do not conform and blocks the output of the drive control signal.

[0027] The equipment actuator receives the drive control signal output by the instruction logic verification module, performs protocol unpacking on the drive control signal, extracts the target winch motor current setpoint and the target water pump speed frequency setpoint, outputs the target winch motor current setpoint to the winch motor drive circuit, adjusts the winch motor's operating current to the target setpoint, and controls the lifting, lowering or stationary action of the sampling tube; outputs the target water pump speed frequency setpoint to the water pump inverter, adjusts the water pump motor's power supply frequency to the target setpoint, and controls the water pump's speed and pumping flow rate.

[0028] Table 1. Real-time Physical Status Data Acquisition Parameters and Data Frame Encapsulation Format of Groundwater Sampling Equipment Timestamp 32bit 100Hz The unique time identifier for synchronous acquisition is generated by a clock counter from the same source. 0x00 Water pump speed frequency 16bit 100Hz The current operating speed and frequency of the water pump motor, in Hz. 0x04 Winch motor current value 16bit 100Hz Real-time current sampling value of the winch motor drive circuit, in mA. 0x06 Hydrological Environment Background Noise Spectrum 128bit 100Hz The frequency domain energy distribution matrix of the water vibration signal inside the well is divided into 16 frequency bands, with 8 bits of energy value for each band. 0x08 Frame Check Code 16bit 100Hz Cyclic redundancy check (CRC) codes in data frames are used to verify the integrity of data transmission. 0x18 In this embodiment, the physical state sensor group encapsulates the acquired real-time physical state data into fixed-length data frames according to the format defined in the table above. Each data frame corresponds to a synchronization timestamp and is synchronously transmitted to the dynamic adaptive filter and instruction logic verification module along with the original speech signal frame with the same timestamp acquired by the audio acquisition module. This ensures that the physical state data and speech signal used in the noise reduction and logic verification processes are in the same time dimension, avoiding processing errors caused by timing deviations. Each parameter within the data frame is set with a fixed bit width and offset address to ensure consistency in data transmission and parsing. The frame checksum is used to verify whether errors occur during data transmission. When the verification fails, the corresponding data frame is discarded, and a previous valid data frame is used as a temporary replacement to avoid system malfunctions caused by data errors.

[0029] In this embodiment, a synchronous acquisition mechanism triggered by a shared clock source enables the synchronous acquisition of the original voice signal and the real-time physical status data of the equipment, ensuring the timing consistency of subsequent processing. A dynamic adaptive filter, combined with the real-time physical status data of the equipment, generates a reverse cancellation signal to reduce noise in the original voice signal, adapting to the time-varying noise environment of the groundwater sampling site. A voiceprint recognition module completes the identification of the operator and semantic parsing of voice commands, realizing access control for voice control. An instruction logic verification module, combined with the real-time physical status of the equipment, performs a timing interlock logic comparison of the voice control instructions, ensuring compatibility between the control instructions and the physical status of the equipment. The equipment's operating status is adjusted by the device's actuator responding to legitimate drive control signals. The technical solution of this embodiment fully realizes the voice control of groundwater sampling equipment, solving the problems in existing technologies where fixed noise reduction models cannot adapt to time-varying noise environments and where control instructions are detached from the physical status of the equipment, leading to mechanical structural risks.

[0030] In one alternative embodiment, refer to Figure 2 and Figure 3 The dynamic adaptive filter comprises a mechanical noise estimation sub-filter and a hydrological environment noise estimation sub-filter. The mechanical noise estimation sub-filter receives the pump speed frequency from real-time physical state data, queries a pre-stored mapping table of pump speed and mechanical vibration noise transfer function, obtains the mechanical vibration noise transfer function at the corresponding speed frequency, and generates a mechanical noise prediction signal based on the transfer function. The hydrological environment noise estimation sub-filter receives the hydrological environment background noise spectrum from real-time physical state data, extracts the energy distribution characteristics of the main frequency band of water flow in the spectrum, and generates a hydrological environment noise prediction signal based on the energy distribution characteristics. The dynamic adaptive filter linearly superimposes the mechanical noise prediction signal and the hydrological environment noise prediction signal to generate a final inverse cancellation signal. This inverse cancellation signal is then added to the original speech signal in the time domain to output a clean speech signal.

[0031] Specifically, the reverse cancellation signal is generated by the linear superposition of the mechanical noise prediction signal and the hydrological environment noise prediction signal, and its time-domain expression is as follows:

[0032] in, For the mechanical noise prediction signal of the nth sampling point, This is the predicted hydrological environmental noise signal for the nth sampling point. The weighting coefficients for the mechanical noise prediction signal are... Let be the weighting coefficients of the hydrological environmental noise prediction signal, and satisfy . .

[0033] Table 2 Mapping Parameters of Pump Speed ​​Frequency and Mechanical Vibration Noise Transfer Function 0-10 [0,0] [-0.92,-0.85] 20-100 0.8 10-20 [0,0] [-0.95,-0.88] 40-150 1.1 20-30 [0,0] [-0.97,-0.91] 60-200 1.3 30-40 [0,0] [-0.98,-0.93] 80-250 1.5 40-50 [0,0] [-0.99,-0.95] 100-300 1.7 In this embodiment, the pre-stored mapping table of pump speed and mechanical vibration noise transfer function is set according to the parameters in the table. After receiving the real-time pump speed frequency, the mechanical noise estimation sub-filter first determines the interval to which the speed frequency belongs, extracts the zero-point parameters, pole parameters, and gain coefficient of the transfer function within the corresponding interval, constructs the corresponding discrete-time transfer function, and uses the pump speed frequency as the input signal. Through time-domain recursive operation of the transfer function, a mechanical noise prediction signal at the corresponding speed frequency is generated. The transfer function adopts a second-order infinite-length unit impulse response filter structure to ensure that the time-domain waveform of the mechanical noise prediction signal matches the time-domain waveform of the mechanical noise component generated by the actual pump operation, thereby improving the noise cancellation effect.

[0034] The physical state sensor group includes a hydrophone array positioned outside the inlet of the groundwater sampling equipment. The hydrophone array collects water vibration signals within the wellbore where the groundwater sampling equipment is located and converts these signals into hydroacoustic electrical signals. The physical state sensor group performs a Fast Fourier Transform (FFT) on the hydroacoustic electrical signals to extract the frequency domain energy distribution matrix within a preset frequency band. This frequency domain energy distribution matrix is ​​used as the hydrological environmental background noise spectrum. The physical state sensor group encapsulates the hydrological environmental background noise spectrum, along with the pump speed frequency and winch motor current value, into a data frame and sends it to a dynamic adaptive filter. The frequency domain conversion of the hydroacoustic electrical signals is achieved through a Fast Fourier Transform, converting the time-domain hydroacoustic electrical signals into frequency-domain complex values. The energy value at the corresponding frequency point is obtained by taking the square of the modulus of the frequency-domain complex value, and these values ​​are combined to form the frequency domain energy distribution matrix within the preset frequency band, i.e., the hydrological environmental background noise spectrum.

[0035] The hydrophone array comprises multiple hydrophone nodes evenly spaced along the well depth. A physical state sensor array acquires the hydroacoustic signals collected by each hydrophone node, calculates the cross-correlation function between adjacent hydrophone nodes, calculates the spatial propagation direction vector of the water noise based on the peak time delay of the cross-correlation function, constructs a beamforming steering vector based on the spatial propagation direction vector, and uses the beamforming steering vector to perform array-weighted summation of the hydroacoustic signals from each hydrophone node to suppress pump coupling noise propagating along the well axis, outputting a suppressed water vibration signal. By calculating the cross-correlation function of the hydroacoustic signals from adjacent hydrophone nodes, the time delay value corresponding to the peak value of the cross-correlation function is obtained. This time delay value is the time difference between the two hydrophone nodes receiving the water noise signal; based on this time difference and the spacing between the hydrophone nodes, the spatial propagation direction vector of the water noise signal is calculated.

[0036] The formula for calculating the array weighted sum is:

[0037] Where M is the total number of nodes in the hydrophone array. These are the weighting coefficients corresponding to the i-th hydrophone node, calculated from the beamforming steering vector. The time-domain sequence of the underwater acoustic electrical signal collected by the i-th hydrophone node. This is the water vibration signal output after weighted summation of the array. By setting the values ​​of the weighting coefficients, the main lobe of the beamforming is aligned with the incident direction of the water environment noise, and the null is aligned with the incident direction of the pump coupling noise propagating along the wellbore axis, thus suppressing the pump coupling noise. Based on the angle between the incident direction of the noise signal and the wellbore axis, the spacing between adjacent hydrophone nodes, and the wavelength of the noise signal, a beamforming steering vector corresponding to the incident direction is constructed.

[0038] The dynamic adaptive filter also includes an error feedback regulator. This regulator acquires the clean speech signal output by the dynamic adaptive filter, calculates its short-time energy and short-time zero-crossing rate, and compares the short-time energy with a preset energy threshold and the short-time zero-crossing rate with a preset zero-crossing rate threshold. When both the short-time energy and the short-time zero-crossing rate exceed the preset energy threshold and the preset zero-crossing rate, the error feedback regulator generates a filter offset compensation coefficient. The mechanical noise estimation sub-filter and the hydrological environment noise estimation sub-filter update the weighting coefficients corresponding to the mechanical noise transfer function mapping table and the weighting coefficients corresponding to the energy distribution characteristics of the main frequency band of the water flow based on the filter offset compensation coefficient. The short-time energy of the m-th frame of the clean speech signal is calculated, using the sum of the squares of the time-domain sampled values ​​of that frame as the short-time energy value.

[0039] Calculate the short-time zero-crossing rate of the clean speech signal in the m-th frame, count the number of times the time-domain waveform of the speech signal in that frame crosses zero, and use the number of zero-crossings per unit frame length as the short-time zero-crossing rate value.

[0040] The formula for calculating the filter offset compensation coefficient is:

[0041] in, Let be the filter offset compensation coefficient corresponding to the m-th frame. The preset step size factor, The preset energy threshold, The preset zero-crossing rate threshold, Let be the short-time energy of the clean speech signal in the m-th frame, and be the short-time zero-crossing rate of the clean speech signal in the m-th frame. and When the filter offset compensation coefficient is positive, it is used to increase the weight coefficient of the corresponding sub-filter; when or At this time, the filter offset compensation coefficient is 0, and no weight coefficient update is performed. The formula for updating the weight coefficients of the sub-filter is:

[0042] in, The weighting coefficients before the update. The updated weighting coefficients, These are the offset compensation coefficients for filtering. By updating the weighting coefficients, the offset error caused by changes in noise characteristics during the filtering process is corrected, thereby improving the noise reduction performance of the dynamic adaptive filter.

[0043] In this embodiment, mechanical noise estimation sub-filters and hydrological environment noise estimation sub-filters are used to predict mechanical noise components and hydrological environment noise components, respectively, generating corresponding noise prediction signals. These signals are then linearly superimposed to obtain an inverse cancellation signal, achieving targeted cancellation of different types of noise components. A hydrophone array collects water vibration signals within the wellbore, and a fast Fourier transform is used to obtain the hydrological environment background noise spectrum, providing an accurate frequency domain basis for hydrological environment noise prediction. Cross-correlation calculations of the hydrophone array yield the spatial propagation direction vector of the noise, and a beamforming steering vector is constructed for array weighted summation, effectively suppressing pump coupling noise propagating along the wellbore axis and improving the accuracy of the hydrological environment background noise spectrum. An error feedback regulator calculates the short-time energy and short-time zero-crossing rate of the clean speech signal, generating filter offset compensation coefficients. These coefficients dynamically update the weight coefficients of the sub-filters, correcting offset errors during the filtering process and further enhancing the adaptability of the dynamic adaptive filter to time-varying noise environments.

[0044] In one alternative embodiment, refer to Figure 4 The voiceprint recognition module extracts the Mel-frequency cepstral coefficient features and fundamental frequency envelope features from the clean speech signal, and concatenates the Mel-frequency cepstral coefficient features and fundamental frequency envelope features into a high-dimensional voiceprint feature vector. The voiceprint recognition module calculates the cosine similarity between the high-dimensional voiceprint feature vector and each template vector in the authorized operator's voiceprint template library, and selects the maximum cosine similarity as the identity matching result. The voiceprint recognition module inputs the clean speech signal into a pre-trained end-to-end speech recognition network, outputs a text sequence, and extracts voice control commands from the text sequence according to preset keyword hitting rules.

[0045] Specifically, in the extraction of Mel frequency cepstral coefficient features, the pre-emphasis processing uses a first-order finite-length unit impulse response filter with a pre-emphasis coefficient set to 0.97 to enhance the high-frequency components of the speech signal. Frame segmentation employs a frame length of 256 sampling points and a frame shift of 128 sampling points, with a Hamming window superimposed on each frame to suppress spectral leakage. Each frame after windowing undergoes a fast Fourier transform to obtain a linear spectrum. This linear spectrum is then passed through a Mel filter bank to obtain the Mel spectrum. The logarithm of the Mel spectrum is followed by a discrete cosine transform (DCT) to extract the first 12 dimensions of the DCT coefficients, which serve as the static Mel frequency cepstral coefficient features. Simultaneously, the first and second differences of the static features are calculated and combined to form a 36-dimensional Mel frequency cepstral coefficient feature.

[0046] In the extraction of fundamental frequency envelope features, the fundamental frequency is the fundamental tone frequency of the speech signal, corresponding to the frequency of human vocal cord vibration. During extraction, autocorrelation is performed on each frame of clean speech signal, and the time delay value corresponding to the peak value of the autocorrelation function is calculated and converted into a fundamental frequency value. The extraction range of the fundamental frequency is set to 80Hz-450Hz, covering the fundamental frequency range of adult males and females. The fundamental frequency values ​​of consecutive frames are smoothed to obtain the fundamental frequency envelope features. Statistical features such as the mean, variance, extreme values, and slope of the fundamental frequency envelope are extracted and combined into a 12-dimensional fundamental frequency envelope feature. In the concatenation of high-dimensional voiceprint feature vectors, the 36-dimensional Mel-frequency cepstral coefficient features are concatenated with the 12-dimensional fundamental frequency envelope features to obtain a 48-dimensional high-dimensional voiceprint feature vector. During the concatenation process, the feature values ​​of each dimension are normalized to ensure that the feature values ​​of different dimensions are within the same numerical range, avoiding excessive influence of dimensions with large numerical differences on the matching results.

[0047] Calculate the cosine similarity between the high-dimensional voiceprint feature vector and each template vector in the authorized operator's voiceprint template library. The cosine similarity value ranges from [-1, 1]. The closer the value is to 1, the higher the similarity between the two vectors. Select the maximum value among all similarities and compare it with a preset similarity threshold. When the maximum value is greater than or equal to the preset similarity threshold, the identity matching result is authorized; otherwise, it is unauthorized.

[0048] The fundamental frequency value is extracted using the autocorrelation function. The autocorrelation function of the m-th frame of the clean speech signal is calculated, and the time delay corresponding to the peak value of the autocorrelation function is the pitch period. Based on the sampling frequency and pitch period of the speech signal, the fundamental frequency value of the m-th frame is calculated. The time delay corresponding to the peak value of the autocorrelation function is... The fundamental frequency is the period of the fundamental tone. The formula for calculating the fundamental frequency value is:

[0049] in, The sampling frequency of the speech signal. is the fundamental frequency value of the m-th frame signal, in Hz.

[0050] The voiceprint recognition module inputs clean speech signals into a pre-trained end-to-end speech recognition network and outputs a text sequence. The end-to-end speech recognition network employs a structure combining convolutional neural networks (CNNs), recurrent neural networks (RNNs), and a connectionist temporal classification loss function. The CNN layers extract local frequency domain features of the speech signal, with the input being a Mel spectrogram, the dimension of which is [time step, number of Mel bands]. Each CNN layer consists of three 2D convolutional layers, each with a 3×3 kernel size, a stride of 1, padding of 1, and a linear rectified function as the activation function. Each convolutional layer is followed by a max-pooling layer with a 2×2 pooling window and a stride of 2, used to reduce feature dimensionality. The RNN layers employ a bidirectional gated recurrent unit structure, consisting of two layers, each with 256 hidden units, used to extract temporal contextual features of the speech signal and output the character probability distribution corresponding to each time step. Connectionist temporal classification loss function is used to align the temporal features of speech with the characters of the text sequence. It eliminates the need for pre-alignment of the speech signal and text at the frame level, enabling end-to-end speech recognition and outputting the corresponding text sequence.

[0051] The voiceprint recognition module has a pre-stored syntax tree for groundwater sampling instructions. The syntax tree contains equipment action nodes and equipment parameter nodes. The equipment action nodes include lifting, lowering, and stopping, and the equipment parameter nodes include depth and flow rates. The voiceprint recognition module inputs a text sequence into a finite state sensor built on the syntax tree. Based on the state transition path of the finite state sensor, it performs legality verification and error correction on the characters in the text sequence, filters out illegal characters that do not conform to the connection relationship of the syntax tree nodes, and outputs the corrected target text sequence. Based on the keyword matching rules, it extracts voice control commands from the target text sequence.

[0052] Table 3. Definitions of syntax tree nodes and state transition rules for groundwater sampling instructions. root node S Instruction start node none Action Node none Action Node A1 Improve action nodes Root node S Parameter node P1, End node E {Lifting, raising, hoisting} Action Node A2 Deploy action nodes Root node S Parameter node P1, End node E {Descending, descending, exploring} Action Node A3 Shutdown Action Node Root node S End node E {Stop, halt, emergency stop} Action Node A4 Water pump start-up node Root node S Parameter node P2, End node E {Start the water pump, start the pump, and pump water} Action Node A5 Water pump stop node Root node S End node E {Stop water pump, turn off pump, stop pump} Parameter node P1 Depth parameter node Action nodes A1, A2 End node E {meters, meters, depth} Parameter node P2 Traffic parameter node Action node A4 End node E {Increase per minute, flow rate, tier} End node E End of instruction node Action nodes, parameter nodes none none In this embodiment, the groundwater sampling instruction syntax tree is constructed according to the node definitions in the table above. The finite state sensor constructs state transition paths based on the node connection relationships of this syntax tree. Each node corresponds to a state of the finite state sensor, and there are valid transition paths between the states of valid parent and child nodes. After the voiceprint recognition module inputs the text sequence output by speech recognition into the finite state sensor, it starts from the initial state corresponding to the root node and performs state transitions sequentially according to the character order of the text sequence. When a character in the text sequence matches the keyword set of the corresponding node, a valid state transition is completed. When a character in the text sequence cannot match the keyword set corresponding to the valid child node of the current node, the character is determined to be an illegal character and is filtered out. When a character in the text sequence has an edit distance of less than or equal to 2 with the keyword set of the valid child node, error correction and replacement are performed, replacing the character with the corresponding valid keyword to ensure the validity of the text sequence. After the verification and error correction processing of the finite state sensor, the target text sequence that conforms to the node connection relationships of the syntax tree is output, effectively filtering out illegal characters related to non-sampling control and eliminating interference from non-standard instructions.

[0053] The preset keyword matching rules include action keyword matching and parameter keyword matching. Action keywords correspond to action nodes in the syntax tree, and parameter keywords correspond to parameter nodes in the syntax tree. The voiceprint recognition module performs word segmentation on the target text sequence, extracts action keywords and parameter keywords, and combines the device actions corresponding to the action keywords with the device parameters corresponding to the parameter keywords to generate complete voice control commands. When the target text sequence is "lower 10 meters", the action keyword extracted after word segmentation is "lower", corresponding to the lower action node, and the parameter keyword is "10 meters", corresponding to the depth parameter node. The combined voice control command is "lower the sampling tube to a depth of 10 meters". When the target text sequence is "stop", the action keyword extracted after word segmentation is "stop", corresponding to the stop action node, with no parameter keywords. The combined voice control command is "stop the equipment".

[0054] In this embodiment, the Mel-frequency cepstral coefficients and fundamental frequency envelope features of the clean speech signal are extracted and concatenated into a high-dimensional voiceprint feature vector, which comprehensively represents the voiceprint characteristics of the operator and improves the accuracy of identity matching. By calculating the cosine similarity between the high-dimensional voiceprint feature vector and the voiceprint template vector, the operator's identity is authenticated, ensuring the operation permission management of the voice control system. Through a pre-trained end-to-end speech recognition network, the clean speech signal is converted into a text sequence, realizing the semantic parsing of the speech signal. Through a finite-state sensor built based on a groundwater sampling-specific instruction syntax tree, the text sequence is validated for legality and corrected for errors, effectively filtering out illegal characters and eliminating interference from non-standard instructions, thus improving the accuracy and reliability of voice control instruction extraction.

[0055] In one alternative embodiment, refer to Figure 5 and Figure 6 The instruction logic verification module has a pre-stored table of correspondences between the current value range of the winch motor and the physical state of the sampling tube. The table includes the lifting current range, the lowering current range, and the quiescent current range. The instruction logic verification module queries the current range to which the current value of the winch motor belongs in the table and determines the physical state of the sampling tube corresponding to the current range as the current physical state. The instruction logic verification module parses the target action contained in the voice control command, queries the pre-stored directed graph of action state transitions, and determines whether there is a directed edge connection when executing the target action from the current physical state. If there is a directed edge connection, it is determined that it conforms to the safe timing logic.

[0056] The instruction logic verification module is equipped with a state transition delay register. When it is determined that the target action to be executed from the current physical state has a directed edge connection, the instruction logic verification module obtains the duration for which the winch motor current value is maintained within the current range and writes the duration into the state transition delay register. The instruction logic verification module queries the pre-stored mapping table between physical states and minimum stable durations, extracts the minimum stable duration corresponding to the current physical state, and compares whether the duration is greater than the minimum stable duration. If it is greater, it finally determines that it meets the safety timing logic and outputs the drive control signal.

[0057] Table 4. Correspondence between the physical state of the sampling tube and the current range of the winch motor, and definition of the directed edge for state transition. S0 0-50 static state Upgrade, Decentralize, Shutdown S0→S1, S0→S2, S0→S0 200 S1 51-200 Improve status stationary, stopped S1→S0、S1→S0 300 S2 201-350 Decentralization status stationary, stopped S2→S0、S2→S0 300 S3 >350 Overload condition shutdown S3→S0 100 In this embodiment, the correspondence between the current range of the winch motor and the physical state of the sampling tube is set according to the table above. When the sampling tube is in different physical states, the load torque of the winch motor is different, and the corresponding drive current value is in different ranges. When the sampling tube is stationary, the winch motor has no load rotation requirement, and the drive current value is in the range of 0-50mA; when the sampling tube is in the lifting state, the winch motor needs to overcome the gravity of the sampling tube and the friction force between the well wall and the borehole, and the drive current value is in the range of 51-200mA; when the sampling tube is in the lowering state, the winch motor needs to overcome the gravitational acceleration of the sampling tube and provide a reverse braking torque, and the drive current value is in the range of 201-350mA; when the sampling tube is in an overload state such as jamming, the load torque of the winch motor increases sharply, and the drive current value is greater than 350mA. The directed graph for action state transitions is constructed based on the directed edge labels in the table above. Each directed edge corresponds to a valid state transition path. Only state transitions with corresponding directed edges are considered valid; those without are considered invalid, thus blocking the output of the drive control signal. When the current physical state of the sampling tube is in the raised state S1, the allowed target actions are stationary and stopped, and the corresponding valid state transition path is S1→S0. If the received voice control command includes a lowering action, the corresponding state transition path is S1→S2. This path does not have a corresponding directed edge, so it is determined to violate the safe timing logic, blocking the execution of the command to avoid mechanical overload or sampling tube breakage.

[0058] The formula for calculating the duration for which the winch motor current value remains within its specified current range is as follows:

[0059] in, This refers to the duration for which the current value remains within the corresponding range, measured in milliseconds (ms). This is the current timestamp, in milliseconds (ms). This is the start timestamp of the current value entering the corresponding interval, in milliseconds. The instruction logic verification module writes the calculated duration into the state transition delay register. The state transition delay register uses a 32-bit register structure and supports recording a maximum duration of 4,294,967,295 milliseconds. When the current value switches to a new interval, the value of the state transition delay register is cleared to zero, and the recording of the duration of the new interval begins again.

[0060] The instruction logic verification module queries the pre-stored mapping table between physical states and minimum stabilization durations, extracts the minimum stabilization duration corresponding to the current physical state, and compares the duration recorded in the state transition delay register with the minimum stabilization duration. Only when the duration exceeds the minimum stabilization duration is the voice control instruction deemed to conform to safe timing logic, and the drive control signal is output. The minimum stabilization duration is set to ensure that the device's physical state completes a stable transition before responding to new control instructions, avoiding mechanical structural shocks caused by receiving new instructions during state transitions. When the sampling tube switches from the stationary state S0 to the raised state S1, it needs to wait for a duration greater than the minimum stabilization duration of 300ms to ensure that the raising action of the sampling tube runs smoothly before responding to new stationary or stop instructions.

[0061] The equipment actuator includes a winch motor drive circuit and a water pump frequency converter. The equipment actuator unpacks the drive control signal and extracts the target winch motor current setpoint and the target water pump speed frequency setpoint. The winch motor drive circuit receives the target winch motor current setpoint and adjusts the pulse width modulation duty cycle of the internal insulated gate bipolar transistor to adjust the winch motor current value to the target winch motor current setpoint. The water pump frequency converter receives the target water pump speed frequency setpoint and adjusts the three-phase AC frequency output to the water pump motor to adjust the water pump speed frequency to the target water pump speed frequency setpoint.

[0062] The winch motor drive circuit uses a proportional-integral closed-loop control algorithm to adjust the pulse width modulation duty cycle. The adjustment formula is as follows:

[0063] in, The pulse width modulation duty cycle is the value of the nth control cycle. The pulse width modulation duty cycle is the value of the (n-1)th control cycle. This is the proportionality coefficient. The integral coefficient is... Set the target winch motor current value. This is the current feedback value of the winch motor in the nth control cycle. This is the integral term of the current error. Through a proportional-integral control algorithm, the pulse width modulation duty cycle is adjusted in real time to stabilize the actual current value of the winch motor at the target setpoint, achieving precise control of the physical state of the sampling tube.

[0064] The water pump frequency converter uses a constant voltage-frequency ratio control algorithm to regulate the output voltage and frequency. The control formula is as follows:

[0065] in, This represents the effective value of the inverter output phase voltage in the nth control cycle, in V. This is a preset constant voltage frequency ratio coefficient, in units of V / Hz; The target pump speed frequency is given in Hz for the nth control cycle. A constant pressure-frequency ratio control algorithm ensures that the main magnetic flux of the pump motor remains constant, avoiding weak or over-magnetic phenomena, thus achieving smooth adjustment of the pump speed frequency and controlling the pump flow rate.

[0066] The drive control signal uses the same source timestamp identifier as the real-time physical state data frame to ensure that the drive control signal and the real-time physical state of the equipment are in the same time dimension. The drive control signal contains the target winch motor current setpoint, the target water pump speed and frequency setpoint, and a check code. The equipment actuator performs cyclic redundancy check on the drive control signal. If the check passes, the corresponding target setpoint is extracted. If the check fails, the drive control signal is discarded to maintain the current operating state of the equipment and avoid equipment malfunctions caused by data errors.

[0067] In this embodiment, a pre-stored correspondence table between the current value range of the winch motor and the physical state of the sampling tube is used to accurately determine the current physical state of the sampling tube, providing an accurate physical state basis for the logical verification of the command. A directed graph of action state transitions is used to determine whether the state transition corresponding to the voice control command is legal, achieving a time-locked logic comparison between the control command and the physical state of the equipment, blocking command outputs that contradict the current physical logic. A state transition delay register records the duration for which the current value remains within the corresponding range, comparing the duration with the minimum stable duration to ensure a smooth command response after a smooth transition of the equipment state, avoiding mechanical structural impact during state transitions. Proportional-integral closed-loop control of the winch motor drive circuit enables precise adjustment of the winch motor current value, controlling the operating state of the sampling tube. Constant voltage-frequency ratio control of the water pump inverter enables smooth adjustment of the water pump speed frequency, controlling the water pump flow rate and ensuring accurate execution of legal voice control commands.

Claims

1. A voice control system for groundwater sampling equipment based on voiceprint recognition, characterized in that, It includes an audio acquisition module, a physical state sensor group, a dynamic adaptive filter, a voiceprint recognition module, an instruction logic verification module, and a device execution mechanism; The audio acquisition module and the physical state sensor group synchronously acquire the original voice signal of the operator and the real-time physical state data of the groundwater sampling equipment. The real-time physical state data includes the pump speed frequency, the winch motor current value and the background noise spectrum of the hydrological environment. The dynamic adaptive filter receives the original speech signal and real-time physical state data, generates an inverse cancellation signal based on the pump speed frequency and the background noise spectrum of the hydrological environment, and performs noise reduction processing on the original speech signal to output a clean speech signal. The voiceprint recognition module extracts features from the clean speech signal, compares them with the pre-stored authorized personnel voiceprint template library, outputs the identity matching result, and performs semantic parsing on the clean speech signal to output voice control commands. The instruction logic verification module receives voice control instructions and real-time physical status data, determines the current physical status of the sampling tube based on the winch motor current value, and performs a timing interlock logic comparison between the action required by the voice control instructions and the current physical status of the sampling tube. When the identity matching result is authorized and the voice control command and the current physical state of the sampling tube conform to the preset security timing logic, the device actuator receives the drive control signal output by the command logic verification module and responds to the drive control signal to change the current value of the winch motor and the speed and frequency of the water pump.

2. The voice control system for groundwater sampling equipment based on voiceprint recognition according to claim 1, characterized in that, The dynamic adaptive filter includes a mechanical noise estimation sub-filter and a hydrological environment noise estimation sub-filter; The mechanical noise estimation sub-filter generates a mechanical noise prediction signal by querying a pre-stored mapping table of water pump speed and mechanical vibration noise transfer function based on the pump speed frequency. The hydrological environment noise estimation sub-filter extracts the energy distribution characteristics of the main frequency band of the water flow based on the hydrological environment background noise spectrum, and generates a hydrological environment noise prediction signal. The dynamic adaptive filter linearly superimposes the mechanical noise prediction signal and the hydrological environment noise prediction signal to generate the inverse cancellation signal, and adds the inverse cancellation signal to the original speech signal in the time domain to output the clean speech signal.

3. The voice control system for groundwater sampling equipment based on voiceprint recognition according to claim 1, characterized in that, The physical state sensor group includes a hydrophone array disposed outside the inlet of the groundwater sampling device. The hydrophone array collects the vibration signal of the water body in the well where the groundwater sampling equipment is located, and converts the water body vibration signal into a hydroacoustic electrical signal. The physical state sensor group performs a fast Fourier transform on the underwater acoustic signal to extract the frequency domain energy distribution matrix within a preset frequency band, and uses the frequency domain energy distribution matrix as the background noise spectrum of the hydrological environment. The physical state sensor group encapsulates the hydrological environment background noise spectrum, the water pump speed frequency, and the winch motor current value into a data frame and sends it to the dynamic adaptive filter.

4. The voice control system for groundwater sampling equipment based on voiceprint recognition according to claim 1, characterized in that, The voiceprint recognition module extracts the Mel frequency cepstral coefficient features and the fundamental frequency envelope features of the pure speech signal, and concatenates the Mel frequency cepstral coefficient features and the fundamental frequency envelope features into a high-dimensional voiceprint feature vector; The voiceprint recognition module calculates the cosine similarity between the high-dimensional voiceprint feature vector and each template vector in the authorized operator's voiceprint template library, and selects the maximum cosine similarity as the identity matching result. The voiceprint recognition module inputs the clean speech signal into a pre-trained end-to-end speech recognition network, outputs a text sequence, and extracts the voice control command from the text sequence according to a preset keyword hitting rule.

5. The voice control system for groundwater sampling equipment based on voiceprint recognition according to claim 1, characterized in that, The instruction logic verification module has a pre-stored table of correspondence between the current value range of the winch motor and the physical state of the sampling tube. The table includes the lifting current range, the lowering current range, and the quiescent current range. The instruction logic verification module queries the current range to which the current value of the winch motor belongs in the corresponding relationship table, and determines the physical state of the sampling tube corresponding to the current range as the current physical state. The instruction logic verification module parses the target action contained in the voice control instruction, queries the pre-stored directed graph of action state transitions, and determines whether there is a directed edge connection when executing the target action from the current physical state. If there is a directed edge connection, it is determined that it conforms to the safe timing logic.

6. The voice control system for groundwater sampling equipment based on voiceprint recognition according to claim 1, characterized in that, The device actuator includes a winch motor drive circuit and a water pump frequency converter; The device actuator unpacks the drive control signal and extracts the target winch motor current setpoint and the target water pump speed frequency setpoint. The winch motor drive circuit receives the target winch motor current setpoint, adjusts the pulse width modulation duty cycle of the internal insulated gate bipolar transistor, and adjusts the winch motor current value to the target winch motor current setpoint. The water pump frequency converter receives the target water pump speed frequency setpoint, adjusts the three-phase AC power frequency output to the water pump motor, and adjusts the water pump speed frequency to the target water pump speed frequency setpoint.

7. The voice control system for groundwater sampling equipment based on voiceprint recognition according to claim 2, characterized in that, The dynamic adaptive filter also includes an error feedback regulator; The error feedback regulator acquires the clean speech signal, calculates the short-time energy and short-time zero-crossing rate of the clean speech signal, compares the short-time energy with a preset energy threshold, and compares the short-time zero-crossing rate with a preset zero-crossing rate threshold. When the short-time energy is greater than the preset energy threshold and the short-time zero-crossing rate is greater than the preset zero-crossing threshold, the error feedback regulator generates a filter offset compensation coefficient. The mechanical noise estimation sub-filter and the hydrological environment noise estimation sub-filter update the weighting coefficients of the mechanical noise transfer function mapping table and the energy distribution characteristics of the main frequency band of the water flow based on the filter offset compensation coefficient.

8. The voice control system for groundwater sampling equipment based on voiceprint recognition according to claim 3, characterized in that, The hydrophone array includes multiple hydrophone nodes that are evenly spaced along the depth direction of the wellbore. The physical state sensor group acquires the hydroacoustic signals collected by each hydrophone node, calculates the cross-correlation function of the hydroacoustic signals between two adjacent hydrophone nodes, and calculates the spatial propagation direction vector of water noise based on the peak time delay of the cross-correlation function. The physical state sensor group constructs a beamforming guide vector based on the spatial propagation direction vector, and uses the beamforming guide vector to perform array weighted summation of the underwater acoustic signals of each of the hydrophone nodes, outputting the suppressed water vibration signal.

9. The voice control system for groundwater sampling equipment based on voiceprint recognition according to claim 5, characterized in that, The instruction logic verification module is equipped with a state transition delay register. When it is determined that the target action is performed from the current physical state with a directed edge connection, the instruction logic verification module obtains the duration for which the winch motor current value is maintained in the current range, and writes the duration into the state transition delay register. The instruction logic verification module queries the pre-stored mapping table of physical states and minimum stable durations, extracts the minimum stable duration corresponding to the current physical state, and compares whether the duration is greater than the minimum stable duration. If it is greater, it finally determines that it conforms to the safety timing logic and outputs the drive control signal.

10. The voice control system for groundwater sampling equipment based on voiceprint recognition according to claim 4, characterized in that, The voiceprint recognition module has a pre-stored syntax tree for groundwater sampling instructions. The syntax tree contains device action nodes and device parameter nodes. The device action nodes include lift, drop, and stop. The device parameter nodes include depth values ​​and flow values. The voiceprint recognition module inputs the text sequence into a finite state sensor constructed based on the syntax tree, performs legality verification and error correction and replacement on the characters in the text sequence according to the state transition path of the finite state sensor, outputs the corrected target text sequence, and extracts the voice control command from the target text sequence according to the keyword hitting rules.

Images