An embedded fuzzy voice control method and system based on language operator quantization

Abstract

The application relates to an embedded fuzzy voice control method and system based on language operator quantification, and relates to the field of embedded voice control, which comprises the following steps: collecting a voice instruction of a user; performing voice recognition on the voice instruction to generate corresponding text instructions; performing semantic analysis on the text instructions to extract control variables, action directions and language operators; determining a basic fuzzy set based on the control variables and the action directions; determining a deformation operator according to the language operator; combining the deformation operator and the basic fuzzy set to generate an input fuzzy quantity, and collecting a current physical state value; and obtaining an aggregated fuzzy set based on the input fuzzy quantity, the current physical state value and a fuzzy rule library; performing defuzzification calculation on the aggregated fuzzy set to determine a physical control signal value, and driving a physical device to perform corresponding actions according to the physical control signal value. The application has the effect of meeting the actual needs of users for the delicacy of device adjustment.

Description

Technical Field

[0001] This invention relates to the field of embedded voice control, and in particular to an embedded fuzzy voice control method and system based on language operator quantization. Background Technology

[0002] Embedded fuzzy voice control refers to a human-computer interaction method that integrates voice recognition and fuzzy logic control technology into embedded devices, enabling users to operate the devices through natural language commands.

[0003] In existing technologies, embedded voice control mainly adopts keyword matching or slot filling-based solutions. This approach uses automatic speech recognition technology to convert user speech into text, then extracts preset keywords from the text and matches them with a device control library to execute corresponding actions. When a user says "turn up the air conditioner temperature," the system can recognize keywords such as "air conditioner," "temperature," and "turn up," and execute a fixed-step operation to increase the temperature setting by 1 or 2 degrees.

[0004] When users use vague commands with degree adverbs, existing technologies can usually only identify the core action command, while ignoring the language operators used to modify the degree of the action. This makes it impossible for users to make fine-grained control through natural language, resulting in a rigid interactive experience that fails to meet users' actual needs for fine-grained device adjustment and needs to be improved. Summary of the Invention

[0005] To meet users' actual needs for finer device adjustments, this invention provides an embedded fuzzy speech control method and system based on language operator quantization.

[0006] In a first aspect, the present invention provides an embedded fuzzy speech control method based on language operator quantization, employing the following technical solution: An embedded fuzzy speech control method based on language operator quantization includes: Collect user voice commands; Speech recognition is performed on voice commands to generate corresponding text commands; Semantic parsing of text instructions is performed to extract control variables, action direction, and language operators used to modify the degree of action; The basic fuzzy set is determined based on control variables and action direction; Determine the transformation operator based on the language operator; The transformation operator and the basic fuzzy set are combined to generate the input fuzzy quantity, and the current physical state value of the controlled equipment is collected; Fuzzy inference is performed based on the input fuzzy quantity, the current physical state value, and the preset fuzzy rule base to obtain an aggregated fuzzy set; The aggregated fuzzy set is defuzzified to determine the physical control signal value, and the physical device is driven to perform corresponding actions based on the physical control signal value.

[0007] By adopting the above technical solution, semantic parsing of voice commands is performed to extract control variables, action direction, and language operators used to modify the degree of action. The language operators are quantized into transformation operators and combined with the basic fuzzy set to generate input fuzzy quantities. Then, fuzzy inference and defuzzification calculations are performed in combination with the current physical state value to finally generate accurate physical control signal values. This enables embedded devices to understand and respond to fuzzy voice commands with degree adverbs, realize fine-grained adjustment of the device, meet users' actual needs for control precision, and improve the naturalness and intelligence level of human-computer interaction.

[0008] Optionally, methods for generating aggregated fuzzy sets are also included: The current physical state value is fuzzified to generate a fuzzy quantity for the current state. The rule excitation intensity and rule consequent fuzzy set are determined based on the input fuzzy quantity, the current state fuzzy quantity, and the preset fuzzy rules. Based on the excitation intensity of the rule, a truncated fuzzy set of the rule consequent is truncated to generate a truncated fuzzy set; All truncated fuzzy sets are superimposed and aggregated to obtain an aggregated fuzzy set.

[0009] By adopting the above technical solution, the current physical state value is fuzzified to generate the current state fuzzy quantity. The input fuzzy quantity and fuzzy rules are combined to determine the rule excitation intensity and the rule consequent fuzzy set. The consequent fuzzy set is truncated and then superimposed and aggregated to realize the effective activation and synthesis of fuzzy rules. This enables the fuzzy inference process to comprehensively consider user instructions and the current equipment state, thereby improving the accuracy and adaptability of control decisions.

[0010] Optionally, the specific method for the defuzzification calculation includes: Determining precise numerical points based on aggregated fuzzy sets; Determine the percentage control increment based on precise numerical points; The register update value is calculated by combining the percentage control increment with the preset hardware timer period value; The physical control signal values ​​are generated based on the register update values.

[0011] By adopting the above technical solution, precise numerical points are determined from the aggregated fuzzy set, converted into percentage control increments, and then the register update value is calculated in combination with the hardware timer period to finally generate physical control signals. This achieves a precise mapping from fuzzy inference results to specific hardware control parameters, ensuring that control instructions can be accurately executed on embedded devices.

[0012] Optionally, user intent verification methods may also be included: After driving the physical device to perform corresponding actions based on the physical control signal value, the user's silent state data is collected during the preset silent period; The initial confidence level of user concern regarding the execution result is determined based on silent state data. When the initial confidence level is lower than the preset confidence level threshold, a reverse fine-tuning signal is generated based on the physical control signal value and the preset reverse amplitude coefficient. The system drives the physical device to perform a reverse action based on the reverse fine-tuning signal, and collects the user's corrective actions within a preset observation period after the reverse action is performed. The corrective actions are used to determine the user's true satisfaction with the physical control signal values, and the user intent model is updated based on the true satisfaction values.

[0013] By adopting the above technical solution, the initial confidence level of attention is determined by the user status data during the silent period. When the confidence level is insufficient, reverse fine-tuning is actively applied to probe the user's response. The true satisfaction level is determined based on the corrective actions during the observation period, and the user intent model is updated. This achieves proactive verification of user intent and adaptive optimization of the model, thereby improving the accuracy of human-computer interaction and user experience.

[0014] Optionally, a method for determining the initial confidence level of attention may also be included: Collect user gaze duration, head orientation angle, and facial expression features; The user's attention concentration coefficient is calculated based on the duration of gaze and the angle of head orientation. The facial expression attention coefficient is obtained based on facial expression features; The initial attention confidence score is obtained by weighting and fusing the attention concentration coefficient and the facial expression attention coefficient.

[0015] By adopting the above technical solution, and by integrating gaze duration, head orientation angle, and facial expression features, the attention concentration coefficient and facial expression attention coefficient are comprehensively calculated, and the initial attention confidence is obtained by weighting. This enables multimodal quantitative evaluation of the user's attention level and improves the accuracy of intent verification.

[0016] Optionally, a method for generating the inverse fine-tuning signal is also included: Collect ambient light and noise levels of the environment in which the controlled equipment is located; The level of perception masking is determined based on ambient brightness and ambient noise values. The base inverse amplitude value is determined based on the perceived masking level; The initial reverse fine-tuning amount is calculated based on the physical control signal value and the basic reverse amplitude value. Collect users' historical correction sensitivity data; Determine the individual sensitivity coefficient for each user based on historical correction sensitivity data; The final reverse fine-tuning amount is obtained by combining the initial reverse fine-tuning amount and the individual sensitivity coefficient; The inverse fine-tuning signal is generated based on the final inverse fine-tuning amount.

[0017] By adopting the above technical solution, the perception masking level is determined by ambient brightness and noise, and then the basic reverse amplitude is determined. Combined with the individual sensitivity coefficient obtained by the user's historical correction sensitivity data, the initial reverse fine-tuning amount is personalized, generating a reverse fine-tuning signal that matches the environmental conditions and user sensitivity, thereby improving the effectiveness of exploratory actions and user acceptance.

[0018] Optional methods include those for perceiving unconscious intentions. After driving the physical equipment to perform corresponding actions based on the physical control signal value, the baseline physiological parameters of the user are collected within the preset stable period; A small-amplitude reverse disturbance signal is generated based on the physical control signal value and the preset perturbation amplitude coefficient; The physical device is driven to perform a micro-reverse action based on the micro-amplitude reverse perturbation signal, and the user's real-time physiological response parameters are collected during the execution of the micro-reverse action. The alertness response coefficient is determined based on the difference between real-time physiological response parameters and baseline physiological parameters. When the alertness response coefficient exceeds the preset alertness threshold, the alertness response coefficient is associated with the physical control signal value and stored in the user intent model.

[0019] By adopting the above technical solution, by applying a small reverse perturbation and monitoring the difference between the user's physiological response parameters and the benchmark value, the alertness response coefficient is quantified. When the alertness response exceeds the threshold, the user's unconscious physiological response is associated with and stored with the current control signal, thereby realizing the perception and learning of the user's unconscious intention and enriching the connotation of the user intention model.

[0020] Optionally, a method for determining the alertness response coefficient may also be included: Pupil and respiratory change sequences were extracted from real-time physiological response parameters. Based on baseline physiological parameters, baseline pupillary and respiratory values ​​are obtained; The pupil change sequence was compared with the baseline pupil value to obtain the pupil fluctuation peak and pupil response delay; The respiratory change sequence was compared with the baseline respiratory value to obtain the respiratory fluctuation amplitude and respiratory recovery time; The alertness response coefficient is obtained by combining the peak pupillary fluctuation, pupillary response delay, respiratory fluctuation amplitude, and respiratory recovery time.

[0021] By adopting the above technical solution, multidimensional physiological characteristic parameters of pupil and respiration are extracted, including pupil fluctuation peak, pupil response delay, respiration fluctuation amplitude and respiration recovery time. The alertness response coefficient is calculated in a comprehensive manner to achieve a refined quantitative assessment of the user's unconscious alertness response and improve the sensitivity and accuracy of intention perception.

[0022] Optionally, an adaptive update method for the perturbation amplitude coefficients is also included: Collect the historical alert coefficient and corresponding historical amplitude coefficient during the execution of historical micro-amplitude reverse disturbance actions; The direction of amplitude adjustment for each perturbation is determined by comparing the historical alert coefficient with the preset ideal response range. The amplitude adjustment value is generated based on the direction of amplitude adjustment; The updated amplitude coefficient is calculated by combining the historical amplitude coefficient and the amplitude coefficient adjustment value, and then stored in the user intent model.

[0023] By adopting the above technical solution, the deviation between the alertness coefficient and the ideal response range during historical perturbations is analyzed to determine the direction of amplitude adjustment and generate adjustment values. The perturbation amplitude coefficient is then adaptively updated, enabling the perturbation intensity to be dynamically optimized according to the user's actual physiological response characteristics, thereby improving the personalization and effectiveness of unconscious intention perception.

[0024] Secondly, this application provides an embedded fuzzy speech control system based on language operator quantization, which adopts the following technical solution: An embedded fuzzy speech control system based on language operator quantization includes: The acquisition module is used to acquire voice commands and current physical status values; The memory is used to store the program that implements an embedded fuzzy speech control method based on language operator quantization; The processor is used to load and execute programs stored in memory.

[0025] In summary, this application includes at least one of the following beneficial technical effects: 1. By semantically parsing voice commands, control variables, action directions, and language operators used to modify the degree of action are extracted. The language operators are quantized into transformation operators and combined with the basic fuzzy set to generate input fuzzy quantities. Then, fuzzy inference and defuzzification calculations are performed in combination with the current physical state value to finally generate accurate physical control signal values. This enables embedded devices to understand and respond to fuzzy voice commands with degree adverbs, realize fine-grained adjustment of the device, meet the user's actual needs for control precision, and improve the naturalness and intelligence of human-computer interaction. 2. By using user status data during the silent period to determine the initial confidence level, when the confidence level is insufficient, proactively apply reverse fine-tuning to probe user response. Based on the corrective actions during the observation period, determine the true satisfaction level and update the user intent model, thereby achieving proactive verification of user intent and adaptive optimization of the model, improving the accuracy of human-computer interaction and user experience. 3. By applying a small reverse perturbation and monitoring the difference between the user's physiological response parameters and the baseline value, the alertness response coefficient is quantified. When the alertness response exceeds the threshold, the user's unconscious physiological response is associated with and stored with the current control signal, thereby realizing the perception and learning of the user's unconscious intention and enriching the connotation of the user intention model. Attached Figure Description

[0026] Figure 1 This is a flowchart of an embedded fuzzy speech control method based on language operator quantization. Detailed Implementation

[0027] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0028] Reference Figure 1 This application discloses an embedded fuzzy speech control method based on language operator quantization, including the following steps: S10: Collect user's voice commands.

[0029] Voice commands refer to the raw audio signals issued by a user through a microphone or other voice input device, containing the user's control intention. These signals carry semantic information about the user's desired control of a device (such as lights, air conditioning, curtains, etc.), particularly the degree modifiers they contain. This step involves real-time monitoring and capturing of the user's voice input via the audio acquisition module of an embedded device.

[0030] S11: Perform speech recognition on the voice command to generate the corresponding text command.

[0031] Text commands refer to parsable text sequences composed of words, converted from user voice commands using automatic speech recognition technology. This step utilizes a lightweight speech recognition engine on an embedded platform to perform the conversion.

[0032] S12: Perform semantic parsing on the text instructions to extract control variables, action direction, and language operators used to modify the degree of action.

[0033] Control variables refer to physical parameters that the user intends to adjust.

[0034] Action direction refers to the trend of the change in the variable that the user expects to control, such as increasing to correspond to growth, decreasing to correspond to reduction, turning on to correspond to startup, and turning off to correspond to stop.

[0035] Linguistic operators are degree adverbs used to modify the degree of an action, such as "slightly", "a little bit", "very", "extremely", "significantly", etc.

[0036] This step involves segmenting and semantically analyzing the text instructions to identify and extract the three core elements mentioned above. The specific natural language processing methods used are common knowledge in this field and will not be elaborated here.

[0037] S13: Determine the basic fuzzy set based on control variables and action direction.

[0038] A basic fuzzy set refers to a membership function pre-defined in a fuzzy logic system for a specific control variable and action direction, representing the basic intent of that action. For example, for the intent of "increasing brightness," the system pre-stores a basic fuzzy set representing "brightening," whose membership function defines the degree of belonging of different brightness increment values ​​to the fuzzy concept of "brightening." This basic fuzzy set is pre-defined and stored in the system by those skilled in the art based on the physical characteristics of the controlled equipment and their control experience.

[0039] S14: Determine the transformation operator based on the language operator.

[0040] Transformation operators are mathematical operators used to perform morphological transformations on the membership functions of basic fuzzy sets, and are used to quantify degree modifiers in natural language into mathematical transformations in fuzzy logic.

[0041] The specific determination method is as follows: The system has a pre-stored operator mapping table, which associates different linguistic operators with corresponding variant operator types. For example, words indicating emphasis, such as "very" and "extremely," are mapped to concentration operators, while words indicating reduction, such as "slightly," "a little," and "slightly," are mapped to dilation operators. The controller, based on the linguistic operators extracted in S12, queries this operator mapping table to match and determine the variant operator used for this control. This operator mapping table is pre-set by those skilled in the art based on linguistic knowledge and control experience.

[0042] S15: Combine the deformation operator and the basic fuzzy set to generate the input fuzzy quantity and collect the current physical state value of the controlled equipment.

[0043] Input fuzzy quantity refers to a new fuzzy set generated by applying a transformation operator to a basic fuzzy set, which accurately reflects the degree of user intent. For example, when the language operator is identified as "very", the centralized operation μ is applied. very (x) = (μ(x)) 2 This generates a relatively steep input fuzzy value representing a "significant increment"; when the language operator is identified as "slight" or "very slight", a diffusion operation is applied: μ slight(x) = (μ(x)) 0.5 This generates a relatively smooth input fuzzy value that represents a "small increment".

[0044] The current physical state value refers to the actual operating parameters of the controlled device acquired in real time by sensors, such as the current brightness percentage and the current temperature value. This step generates fuzzy input representing the user's intent and obtains physical feedback representing the current state, preparing the preconditions for fuzzy inference.

[0045] S16: Perform fuzzy inference based on the input fuzzy quantity, the current physical state value, and the preset fuzzy rule base to obtain an aggregated fuzzy set.

[0046] A fuzzy rule base refers to a pre-defined set of rules consisting of several fuzzy conditional statements in the form of "if...then...", used to describe the logical relationship between input fuzzy quantities and output control quantities. For example, "If the current brightness is too high and the user command is to brighten it slightly, then the output increment is minimal."

[0047] Aggregated fuzzy sets refer to the complex fuzzy regions representing the final control decisions obtained by substituting input fuzzy quantities and current physical state values ​​into a fuzzy rule base through a fuzzy inference engine for rule matching, rule activation, and result aggregation. This step specifically employs the Mamdani fuzzy inference model and includes sub-steps such as fuzzification, rule evaluation, truncation operations, and superposition aggregation, the specific processes of which will be described in detail in subsequent steps. This fuzzy rule base is pre-defined and stored in the system by those skilled in the art based on the control characteristics and safety requirements of the controlled equipment.

[0048] S17: Perform defuzzification calculation on the aggregated fuzzy set to determine the physical control signal value, and drive the physical device to perform corresponding actions based on the physical control signal value.

[0049] The physical control signal value refers to the precise numerical value that can be recognized by the hardware actuator by transforming the fuzzy region of the aggregated fuzzy set through defuzzification calculation. The specific steps of the defuzzification calculation will be explained in detail in S30 to S33, and will not be repeated here.

[0050] The physical control signal value is sent to the actuator (such as LED driver circuit, motor driver, etc.) to drive the physical device to perform the corresponding action according to the user's instructions.

[0051] It also includes methods for generating aggregated fuzzy sets: S20: Fuzzify the current physical state value to generate a fuzzy quantity for the current state.

[0052] The current state fuzzy value refers to the fuzzy representation of the degree of membership of a precise physical value collected by a sensor (e.g., current brightness of 80%, normalized to 0.8) in various fuzzy sets (e.g., "low", "medium", "high") obtained by mapping the value to a fuzzy universe through a preset membership function. For example, the current brightness value of 0.8 might be mapped to a "high brightness" fuzzy set with a membership degree of 0.7, and simultaneously mapped to a "medium brightness" fuzzy set with a membership degree of 0.3. This step is completed through a fuzzification interface, and the specific shape of the membership function (e.g., triangle, trapezoid, Gaussian) is preset by those skilled in the art based on the characteristics of the controlled equipment and the control accuracy requirements.

[0053] S21: Determine the rule excitation intensity and rule consequent fuzzy set based on the input fuzzy quantity, the current state fuzzy quantity, and the preset fuzzy rules.

[0054] Fuzzy rules refer to pre-defined fuzzy conditional statements in the form of "IF (prefix) THEN (consequence)" used to describe the logical mapping relationship between input fuzzy quantities and output control quantities. For example, a rule can be expressed as: "IF current brightness IS high AND instruction increment IS small THEN output increment IS zero". This fuzzy rule library is pre-set and stored in the system by those skilled in the art based on the control characteristics and safety requirements of the controlled equipment.

[0055] Rule activation intensity refers to the intensity coefficient of a rule being activated, calculated by matching the input fuzzy quantity and the current state fuzzy quantity with the antecedent of each fuzzy rule.

[0056] The specific calculation method is as follows: the minimum value of the membership degree of each antecedent condition is taken as the excitation intensity of the rule using the minimum value of the minimum value of the membership degree of each antecedent condition in the Mamdani inference model. For example, for the above rule, if the membership degree of "current brightness is high" is 0.7 and the membership degree of "instruction increment is small" is 0.5, then the excitation intensity of the rule is w=min(0.7, 0.5)=0.5.

[0057] The rule consequent fuzzy set refers to the fuzzy set defined in the consequent of each activated fuzzy rule, representing the suggested output result of that rule. This rule consequent fuzzy set is obtained directly by querying the fuzzy rule library, based on the fuzzy variables described in the rule consequent (e.g., "output increment IS zero") and their predefined membership functions. The shapes (e.g., triangular, trapezoidal, Gaussian) and parameters of these membership functions are pre-defined by those skilled in the art based on the control characteristics and output range of the controlled equipment and stored in the system.

[0058] S22: Based on the excitation intensity of the rule, perform a truncation operation on the fuzzy set of the rule consequent to generate a truncated fuzzy set.

[0059] Truncated fuzzy sets refer to fuzzy sets generated by pruning or scaling the corresponding rule consequent fuzzy sets based on the excitation intensity of each rule. Specifically, the membership function of the rule consequent fuzzy set is truncated using the excitation intensity as an upper limit. The portion of the membership value below the excitation intensity is retained, while the portion above the excitation intensity is flattened to the excitation intensity value. This operation reflects the contribution of each rule to the final decision; rules with higher excitation intensity retain a larger fuzzy set region after truncation.

[0060] S23: Overlay and aggregate all truncated fuzzy sets to obtain an aggregated fuzzy set.

[0061] An aggregated fuzzy set refers to the set of all truncated fuzzy sets after truncation, which are then combined to form an irregular geometric region representing the comprehensive decision-making result. This aggregated fuzzy set is denoted as μ. agg (y), where y is a variable in the output universe of discourse. This step completes the result aggregation in fuzzy inference, providing input for subsequent defuzzification calculations. The aggregation operation uses the maximum value method, that is, taking the maximum membership value of all truncated fuzzy sets at each point as the membership value of the aggregation result.

[0062] Specific methods for defuzzification calculation include: S30: Determine precise numerical points based on aggregated fuzzy sets.

[0063] Precise numerical points refer to the single numerical values ​​that represent the center or optimal representation of an irregular geometric region of an aggregated fuzzy set, obtained through defuzzification algorithms. Specifically, the centroid method is used to calculate the geometric centroid of the aggregated fuzzy set. The centroid method is achieved by calculating the weighted average of the region covered by the aggregated fuzzy set; its discretization formula is as follows: , where y i μ represents the number of sampling points on the output universe of discourse (e.g., the percentage of the PWM increment). agg (y) i y is the membership value corresponding to this sampling point. crisp For precise numerical points, the specific implementation of this centroid method is common knowledge in this field and will not be elaborated here.

[0064] S31: Determine the percentage control increment based on precise numerical points.

[0065] Percentage control increment refers to the percentage increase or decrease required relative to the current physical state value after mapping the precise numerical point calculated by S30 to the percentage change range of the control variable. For example, the precise numerical point y calculated using the center of gravity method... crispIf the value is 5.2, the corresponding percentage control increment is 5.2%, meaning that the control quantity needs to be increased by 5.2% based on the current value. This mapping relationship is achieved through a preset scaling factor or linear transformation. The specific mapping parameters are set in advance by those skilled in the art based on the control range and accuracy requirements of the controlled equipment.

[0066] S32: Combines percentage control increments and preset hardware timer period values ​​to calculate register update values.

[0067] The hardware timer period value refers to the preset value of the automatic reload register (ARR) of the timer used to generate PWM waveforms or other periodic control signals. This value determines the period or resolution of the control signal. The hardware timer period value is preset by those skilled in the art and will not be elaborated here.

[0068] The register update value refers to the specific value that needs to be written to the hardware control register. The calculation method is: Register update value = Current register value + (Hardware timer period value × Percentage control increment). For example, if the current duty cycle register CCR value is 800, the hardware timer period value ARR is 1000, and the percentage control increment is 5.2%, then the register update value = 800 + 1000 × 5.2% = 852.

[0069] S33: Generate physical control signal values ​​based on register update values.

[0070] The register update value calculated by S32 is written to the corresponding hardware control register. The hardware timer automatically generates a PWM waveform with the corresponding duty cycle based on this value. After being amplified by the driver circuit, this waveform drives the physical device (such as LED lights, motors, etc.) to perform corresponding actions according to the user's instructions. For example, after writing the register update value 852 to the PWM register, the duty cycle of the output PWM waveform becomes 85.2%, and the brightness of the light increases smoothly, completing one closed-loop control. The method for generating this hardware control signal is common knowledge in the field and will not be elaborated here.

[0071] It also includes user intent verification methods: S40: After driving the physical device to perform corresponding actions based on the physical control signal value, collect the user's silent state data within the preset silent period.

[0072] The silent period refers to the waiting time after the system has completed a control action and before it initiates any exploratory operation, typically set to 3 to 5 seconds. The purpose of this silent period is to give the user time to react and observe whether the user will proactively provide feedback on the execution result. The duration of this silent period is preset by those skilled in the art based on average user reaction time and application scenario requirements.

[0073] Silent state data refers to user behavior data collected by visual sensors (such as cameras) during silent periods. This includes information such as the user's gaze duration, head orientation angle, facial expression features, and body posture. This data is used to initially determine whether the user has noticed the device's execution result.

[0074] S41: Determine the user's initial confidence level regarding the execution result based on silent state data.

[0075] Initial attention confidence refers to a numerical indicator calculated based on user behavior characteristics during the silent period, quantifying the degree of user attention to the execution result. This value ranges from 0 to 1; a higher value indicates that the user is more likely to have paid attention to the device's execution result. The specific method for determining initial attention confidence will be explained in detail in subsequent sections S50 to S53, and will not be elaborated upon here.

[0076] S42: When the initial confidence level is lower than the preset confidence level threshold, a reverse fine-tuning signal is generated based on the physical control signal value and the preset reverse amplitude coefficient.

[0077] The confidence threshold is a critical value used to determine whether a user is paying sufficient attention to the execution result; a value below this threshold indicates that the user may not have noticed the change in the device. This threshold is preset by those skilled in the art based on the actual application scenario and the expected level of user attention, for example, it may be set to 0.6.

[0078] The reverse amplitude coefficient is a preset positive number less than 1, used to ensure that the amplitude of the reverse fine-tuning is less than the original control amplitude, avoiding user discomfort. This coefficient is preset by those skilled in the art based on ergonomic principles and experimental data, for example, set to between 0.2 and 0.4.

[0079] A reverse fine-tuning signal is a control command used to drive a physical device to perform a small-amplitude, tentative action that is opposite in direction to the original action. The specific method for generating the reverse fine-tuning signal will be explained in detail in S60 to S67, and will not be repeated here.

[0080] S43: Drive the physical device to perform a reverse action based on the reverse fine-tuning signal, and collect the user's corrective actions within the preset observation period after the reverse action is performed.

[0081] The observation period refers to a waiting time after the reverse action is performed, typically set to 3 to 5 seconds, to observe whether the user will make a corrective action. The length of this observation period is preset by those skilled in the art based on the average user reaction time.

[0082] Corrective actions refer to proactive feedback behaviors that users may exhibit during the observation period to express dissatisfaction or a desire for further adjustments. These include issuing voice commands again, manually pressing buttons, operating the touchscreen, and using gesture control. This step monitors and collects these corrective actions in real time through voice recognition, button detection, and visual recognition.

[0083] S44: Determine the user's true satisfaction value with the physical control signal value based on the corrective action, and update the user intent model based on the true satisfaction value.

[0084] The true satisfaction score is a quantitative value representing the user's satisfaction with the original execution result, determined by comprehensively considering whether the user took corrective actions during the observation period, as well as the magnitude and direction of those actions. Specifically, if the user took corrective actions during the observation period (e.g., saying "adjust the brightness"), it indicates that the user cares about the original execution result, and that the original result may not have fully met their expectations. In this case, the true satisfaction score is set to a lower value (e.g., 0.3). If the user did not take any corrective actions during the observation period, it indicates that the user may be satisfied with the original execution result, or may not have noticed the change. In this case, the true satisfaction score is set to a moderate value (e.g., 0.7). This judgment logic and specific values ​​are predetermined by those skilled in the art based on experimental data.

[0085] A user intent model is a personalized database that stores users' historical interaction data and satisfaction feedback, used to continuously optimize the system's understanding of users' personalized preferences. This model includes parameters such as the user's sensitivity coefficients to different language operators, historical true satisfaction values, and individual correction sensitivity. By continuously updating this model, the system can gradually improve the quantification accuracy of degree words such as "a little bit" and "very" for the same user.

[0086] It also includes a method for determining the initial confidence level: S50: Collects the user's gaze duration, head orientation angle, and facial expression features.

[0087] Eye-dwell time refers to the cumulative duration for which a user's gaze remains focused on or near the device, measured using a camera combined with eye-tracking technology. A longer duration indicates a higher level of user attention to the device's status.

[0088] The head orientation angle refers to the spatial angle between the user's facial normal and the device's orientation, calculated from the camera image using facial pose estimation technology. The smaller this angle, the more directly the user is facing the device, and the greater the likelihood of their attention being focused on it.

[0089] Facial expression features refer to the categories of user facial expressions (such as calm, confused, satisfied, surprised, etc.) and their intensity values ​​extracted from camera images through facial action unit analysis or deep learning models. These features are used to infer the user's subconscious reaction to the current state.

[0090] S51: Calculate the user's attention concentration coefficient based on gaze duration and head orientation angle.

[0091] The attention concentration coefficient is a numerical indicator reflecting the degree to which a user's attention is focused on a device, calculated by combining gaze duration and head orientation angle. The specific calculation method is as follows: gaze duration is normalized to obtain a gaze score, and head orientation angle is mapped to a matching score (the smaller the angle, the higher the score). The two are then weighted and summed. The weights of gaze duration and head orientation angle are predetermined by those skilled in the art based on experimental data; for example, gaze weight is 0.6, and head orientation weight is 0.4.

[0092] S52: Obtain the expression attention coefficient based on facial expression features.

[0093] The facial expression attention coefficient is a numerical indicator reflecting the user's level of attention to the current state, obtained by mapping facial expression features. The system has a pre-stored facial expression attention mapping table, which was established in advance by those skilled in the art based on psychological research and experimental data. By matching the identified facial expression features with the mapping table, the corresponding facial expression attention coefficient can be obtained.

[0094] S53: Weighted fusion of attention concentration coefficient and facial expression attention coefficient to obtain initial attention confidence.

[0095] The attention concentration coefficient and facial expression attention coefficient are each assigned a preset fusion weight, and then weighted and summed to obtain the final initial attention confidence score. The principle for setting the fusion weights is that the attention concentration coefficient reflects explicit attention behavior, while the facial expression attention coefficient reflects implicit psychological states, and the two are complementary. For example, if the weight of the attention concentration coefficient is set to α = 0.6 and the weight of the facial expression attention coefficient is set to β = 0.4, then the initial attention confidence score = α × attention concentration coefficient + β × facial expression attention coefficient. These fusion weights are preset by those skilled in the art based on the effectiveness of the two types of indicators in actual application scenarios.

[0096] It also includes methods for generating inverse fine-tuning signals: S60: Collect the ambient brightness and ambient noise values ​​of the environment where the controlled equipment is located.

[0097] Ambient brightness refers to the lighting intensity of the surrounding environment, collected by a photosensor. This value reflects the strength of the current ambient light and affects the user's visual perception of changes in brightness.

[0098] Ambient noise level refers to the background sound intensity of the environment surrounding the device, as captured by a microphone. This value reflects the level of noise in the current environment and affects the user's ability to perceive auditory feedback or the sound of the device operating.

[0099] S61: Determine the level of perception masking based on ambient brightness and ambient noise values.

[0100] Perception masking level is a quantitative indicator determined based on a combination of ambient brightness and ambient noise, reflecting the degree of interference of the current environment with subtle changes in user perception. The specific determination method is as follows: the system has a pre-stored ambient masking mapping table, which defines corresponding perception masking levels based on different combinations of ambient brightness and ambient noise ranges. The controller queries this mapping table based on the ambient brightness and ambient noise values ​​collected by the S60 to match and determine the current perception masking level. This mapping table is pre-set by those skilled in the art based on ergonomic principles and experimental data.

[0101] S62: Determine the base inverse amplitude value based on the perceived masking level.

[0102] The basic reverse amplitude value refers to the initial amplitude parameter of the reverse action, obtained by matching it from a preset amplitude mapping table based on the perceived masking level. A higher perceived masking level indicates greater environmental interference with user perception, requiring a larger reverse amplitude to ensure the user can perceive the reverse action. Specifically, the system has a pre-stored amplitude mapping table, which associates different basic reverse amplitude values ​​with different perceived masking levels. The controller queries this mapping table based on the perceived masking level determined in step S61 to obtain the basic reverse amplitude value. This mapping table is preset by those skilled in the art based on experimental data.

[0103] S63: Calculate the initial reverse fine-tuning amount based on the physical control signal value and the basic reverse amplitude value.

[0104] The initial reverse fine-tuning amount refers to a preliminary reverse action amplitude value obtained by multiplying the physical control signal value by the basic reverse amplitude value and taking the opposite direction. The specific calculation method is: Initial reverse fine-tuning amount = -(physical control signal value × basic reverse amplitude value).

[0105] S64: Collect the user's historical correction sensitivity data.

[0106] Historical correction sensitivity data refers to statistical information about a user's corrective responses to reverse probing actions over multiple past interactions. This data includes the magnitude of the reverse probing action each time, whether the user made a correction, the delay time of the correction, and the magnitude of the correction. This data is retrieved from the user intent model and used to analyze the perceptual sensitivity characteristics of individual users.

[0107] S65: Determine the individual sensitivity coefficient for each user based on historical correction sensitivity data.

[0108] The individual sensitivity coefficient is a personalized coefficient reflecting a user's sensitivity to subtle changes, derived from the analysis of their historical corrective behavior. Specifically, it is determined by statistically analyzing historical corrective sensitivity data, calculating the user's correction probability at different magnitudes of adverse reactions, establishing a magnitude correction probability curve, and taking the reciprocal of the magnitude value corresponding to a 50% correction probability as the individual sensitivity coefficient. Users with high sensitivity (who frequently correct small changes) correspond to higher sensitivity coefficients, while users with low sensitivity correspond to lower sensitivity coefficients. This calculation method is pre-defined by those skilled in the art based on statistical principles.

[0109] S66: Combine the initial reverse fine-tuning amount and the individual sensitivity coefficient to obtain the final reverse fine-tuning amount.

[0110] The final reverse fine-tuning amount refers to the magnitude of the reverse action after personalized adjustment for the user, obtained by calculating the initial reverse fine-tuning amount and the individual sensitivity coefficient. The specific calculation method is: Final reverse fine-tuning amount = Initial reverse fine-tuning amount × Individual sensitivity coefficient.

[0111] S67: Generate a reverse fine-tuning signal based on the final reverse fine-tuning amount.

[0112] The reverse fine-tuning signal refers to a control command containing the final reverse fine-tuning amount information, used to drive the physical device to perform a reverse action. Specifically, it is generated by converting the final reverse fine-tuning amount into a control command with the same format as the physical control signal value in S17, encapsulating it into a control signal format recognizable by the actuator, and sending it to the corresponding drive module. This signal will be used to perform the reverse action in S43.

[0113] It also includes methods for perceiving unconscious intentions: S70: After driving the physical device to perform corresponding actions based on the physical control signal value, it collects the user's baseline physiological parameters within the preset stable period.

[0114] The settling period refers to the short waiting time after the device completes its action and before any minor disturbance is applied to the system, typically set to 2 to 3 seconds. The purpose of this settling period is to allow the user to recover from the stimulation of the device's action and to collect physiological data that reflects the user's true baseline. The duration of this settling period is preset by those skilled in the art based on the user's physiological response time constant.

[0115] Baseline physiological parameters refer to the set of data collected during a stable period using non-contact sensors (such as millimeter-wave radar, high-definition cameras, and thermal imagers) that reflect the user's physiological baseline in a calm state. These parameters include baseline pupil diameter, baseline respiratory rate, baseline heart rate, and baseline muscle tension.

[0116] S71: Generates a small-amplitude reverse disturbance signal based on the physical control signal value and the preset perturbation amplitude coefficient.

[0117] The perturbation amplitude coefficient refers to a preset parameter for the amplitude of a small perturbation that is much smaller than the user's conscious perception threshold, typically set between 0.05 and 0.1. This coefficient is preset by those skilled in the art based on ergonomic principles and experimental data to ensure that the amplitude of the perturbation is below the absolute threshold of human senses, but sufficient to trigger a subconscious physiological response.

[0118] A micro-amplitude reverse disturbance signal is a control command used to drive a physical device to perform an extremely small reverse action that is almost imperceptible to the user. Specifically, it is generated by multiplying the physical control signal value by a disturbance amplitude coefficient to obtain the amplitude value of the micro-amplitude disturbance, and then taking the opposite direction to the original action.

[0119] S72: Drive the physical device to perform a micro-reverse action based on the micro-amplitude reverse perturbation signal, and collect the user's real-time physiological response parameters during the execution of the micro-amplitude reverse action.

[0120] Real-time physiological response parameters refer to physiological data sequences of the same type as those in the S70, but varying over time, synchronously acquired via non-contact sensors during the execution of micro-amplitude reverse movements. These parameters include real-time pupil diameter change sequences, real-time respiratory rate change sequences, and real-time heart rate change sequences. This data is continuously recorded at a high sampling rate to capture the user's subconscious responses to micro-perturbations.

[0121] S73: The alertness response coefficient is determined based on the difference between real-time physiological response parameters and baseline physiological parameters.

[0122] The alertness response coefficient is a numerical indicator that quantifies the user's subconscious level of alertness, calculated by comparing real-time physiological response parameters with baseline physiological parameters. A higher coefficient indicates a stronger subconscious response to micro-perturbations, suggesting a potential focus on the original execution result. The specific determination method is explained in detail in subsequent steps S80 to S84.

[0123] S74: When the alertness response coefficient exceeds the preset alertness threshold, the alertness response coefficient is associated with the physical control signal value and stored in the user intent model.

[0124] The alertness threshold is a critical value used to determine whether a user's subconscious reaction is sufficiently significant. This threshold is pre-set by those skilled in the art based on extensive experimental data and statistical analysis, for example, 0.7. When the alertness reaction coefficient exceeds this threshold, it indicates that although the user failed to consciously perceive the micro-perturbation, their subconscious mind is concerned about the original execution result. At this point, the alertness reaction coefficient is associated with the current physical control signal value and stored in the user intent model for subsequent personalized parameter optimization, such as adjusting the perturbation amplitude coefficient or updating the user sensitivity model.

[0125] It also includes methods for determining the alertness response coefficient: S80: Extract pupil change sequences and respiratory change sequences from real-time physiological response parameters.

[0126] A pupil change sequence refers to a sequence of pupil diameter measurements recorded sequentially over time by an eye-tracking camera or a high-frame-rate camera during the execution of a micro-amplitude reverse motion. This sequence reflects the dynamic response of the pupil to micro-perturbations. The sampling frequency is preset by a person skilled in the art based on reaction speed requirements, for example, 50 frames per second.

[0127] A respiratory change sequence refers to a sequence of respiratory rate measurements continuously recorded chronologically by millimeter-wave radar during the execution of a micro-amplitude reverse motion. This sequence reflects the dynamic response of the respiratory pattern to micro-perturbations.

[0128] S81: Based on baseline physiological parameters, baseline pupillary and respiratory values ​​are obtained.

[0129] The baseline pupil value refers to the average pupil diameter value extracted from the baseline physiological parameters collected from S70 during the stable period. This value serves as the baseline for subsequent comparisons of pupil changes, and is specifically calculated as the arithmetic mean of all pupil diameter measurements during the stable period.

[0130] The baseline respiratory rate refers to the average respiratory rate during the stable period extracted from the baseline physiological parameters collected from the S70. This value serves as the baseline for subsequent comparisons of respiratory changes, and is specifically calculated as the arithmetic mean of all respiratory rate measurements during the stable period.

[0131] S82: Compare the pupil change sequence with the baseline pupil value to obtain the pupil fluctuation peak and pupil response delay.

[0132] The pupillary oscillation peak value refers to the maximum positive deviation (pupil dilation) or maximum negative deviation (pupil constriction) relative to a reference pupillary value in a pupillary change sequence. Specifically, it is determined by calculating the difference between each measured value in the pupillary change sequence and the reference pupillary value, and taking the maximum absolute value of these differences as the pupillary oscillation peak value. The larger this peak value, the stronger the pupil's response to micro-perturbations.

[0133] Pupil response delay refers to the time difference between the start of a slight reverse action and the moment when a significant deviation from the baseline pupil value (exceeding a preset pupil change threshold) first appears in the pupil change sequence. A shorter delay indicates a faster subconscious reaction. The pupil change threshold is preset by a person skilled in the art based on the measurement noise level.

[0134] S83: Compare the respiratory change sequence with the baseline respiratory value to obtain the respiratory fluctuation amplitude and respiratory recovery time.

[0135] Respiratory fluctuation amplitude refers to the maximum fluctuation value of a respiratory change sequence relative to a baseline respiratory value. Specifically, it is determined by calculating the absolute value of the difference between each measured value in the respiratory change sequence and the baseline respiratory value, and taking the maximum value as the respiratory fluctuation amplitude. The larger this amplitude, the more significantly the respiratory system is affected by micro-perturbations.

[0136] The respiratory recovery time refers to the length of time from the moment when the respiratory change sequence first deviates significantly from the baseline respiratory value (exceeding a preset respiratory change threshold) to the moment when the respiratory change sequence recovers to a stable range near the baseline respiratory value (remaining within the preset recovery threshold for multiple consecutive frames). The longer this time, the longer the effect of the micro-perturbation on the respiratory system lasts. The respiratory change threshold and recovery threshold are preset by those skilled in the art based on the measurement noise level and physiological recovery characteristics.

[0137] S84: The alertness response coefficient is obtained by combining the peak pupil fluctuation, pupil response delay, respiratory fluctuation amplitude, and respiratory recovery time.

[0138] The alertness response coefficient is a quantitative indicator that comprehensively reflects the user's subconscious level of alertness, calculated by integrating the four characteristic parameters mentioned above. The specific determination method is as follows: The system has a pre-stored alertness response coefficient calculation model. This model normalizes the pupillary fluctuation peak value, pupillary response delay, respiratory fluctuation amplitude, and respiratory recovery time, and assigns preset fusion weights to each parameter (e.g., pupillary fluctuation peak value weight 0.3, pupillary response delay weight 0.2, respiratory fluctuation amplitude weight 0.3, and respiratory recovery time weight 0.2). Then, a weighted sum is performed to obtain the alertness response coefficient. The pupillary response delay is negatively correlated with the alertness response coefficient (the shorter the delay, the higher the coefficient), while the other three parameters are positively correlated with the alertness response coefficient. This calculation model and fusion weights are preset by those skilled in the art based on physiological research findings and experimental data.

[0139] It also includes an adaptive update method for the perturbation amplitude coefficients: S90: Collect the historical alert coefficient and corresponding historical amplitude coefficient during the execution of historical micro-amplitude reverse disturbance actions.

[0140] Historical alertness coefficients refer to the alertness response coefficient values ​​recorded and stored in the user intent model through steps S70 to S74 during multiple past minor reverse perturbation actions. These coefficients reflect the intensity of the user's subconscious response triggered by each minor perturbation.

[0141] The historical amplitude coefficient refers to the perturbation amplitude coefficient value used in each micro-amplitude reverse perturbation action, corresponding to the aforementioned historical alert coefficient. This data is also stored in the user intent model, forming a one-to-one data pair with the historical alert coefficient.

[0142] S91: Determine the direction of amplitude adjustment for each perturbation based on the comparison between the historical alert coefficient and the preset ideal response range.

[0143] The ideal response range refers to a preset, expected range of alertness response coefficients used to assess whether the amplitude of each perturbation is appropriate. This range is preset by those skilled in the art based on extensive experimental data and expected subconscious response intensity, for example, set to [0.6, 0.8]. When the alertness response coefficient is below the lower limit of this range, it indicates that the perturbation amplitude is too small and has not triggered a sufficient subconscious response; when the alertness response coefficient is above the upper limit of this range, it indicates that the perturbation amplitude is too large and there is a risk of it being consciously perceived by the user; when the alertness response coefficient falls within the range, it indicates that the amplitude is appropriate.

[0144] The amplitude adjustment direction refers to the trend of the perturbation amplitude coefficient that needs to be adjusted, determined based on the comparison results between the previous alert coefficient and the ideal response range. The specific determination method is as follows: if the alert coefficient is lower than the lower limit of the range, the amplitude adjustment direction is "increase"; if the alert coefficient is higher than the upper limit of the range, the amplitude adjustment direction is "decrease"; if the alert coefficient falls within the range, the amplitude adjustment direction is "unchanged".

[0145] S92: Generate amplitude coefficient adjustment value based on amplitude adjustment direction.

[0146] The amplitude coefficient adjustment value refers to the specific numerical change used to correct the perturbation amplitude coefficient, calculated based on a determined adjustment direction. The specific generation method is as follows: the system has a pre-stored adjustment step size table, which associates corresponding adjustment step size values ​​with different adjustment directions. For example, when the adjustment direction is "increase," the adjustment step size is +0.01; when the adjustment direction is "decrease," the adjustment step size is -0.01; and when the adjustment direction is "unchanged," the adjustment step size is 0. This adjustment step size is preset by those skilled in the art based on the system's convergence speed and stability requirements. For multiple historical data sets, the average or cumulative value of each adjustment step size can be taken as the final amplitude coefficient adjustment value.

[0147] S93: Combine historical amplitude coefficients and amplitude coefficient adjustment values ​​to calculate the updated amplitude coefficients and store them in the user intent model.

[0148] The updated amplitude coefficient is the new coefficient obtained by adding the adjusted amplitude coefficient value calculated in this step to the historical amplitude coefficient. The specific calculation method is: Updated amplitude coefficient = Historical amplitude coefficient + Adjusted amplitude coefficient value. This updated amplitude coefficient will replace the original perturbation amplitude coefficient and be used to generate subsequent micro-amplitude reverse perturbation signals, thereby achieving personalized adaptive optimization of the perturbation amplitude and ensuring that it is always within the optimal range of "below the conscious perception threshold but sufficient to trigger a subconscious reaction." The updated coefficient is stored in the user intent model for use in step S71.

[0149] Based on the same inventive concept, embodiments of the present invention provide an embedded fuzzy speech control system based on language operator quantization, comprising: The data acquisition module is used to collect voice commands, current physical state values, silence state data, correction actions, gaze dwell time, head orientation angle, facial expression features, ambient brightness values, ambient noise values, historical correction sensitivity data, baseline physiological parameters, real-time physiological response parameters, historical alertness coefficients, and historical amplitude coefficients. The memory is used to store the program that implements an embedded fuzzy speech control method based on language operator quantization; The processor is used to load and execute programs stored in memory.

[0150] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0151] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An embedded fuzzy speech control method based on language operator quantization, characterized in that, include: Collect user voice commands; Speech recognition is performed on voice commands to generate corresponding text commands; Semantic parsing of text instructions is performed to extract control variables, action direction, and language operators used to modify the degree of action; The basic fuzzy set is determined based on control variables and action direction; Determine the transformation operator based on the language operator; The transformation operator and the basic fuzzy set are combined to generate the input fuzzy quantity, and the current physical state value of the controlled equipment is collected; Fuzzy inference is performed based on the input fuzzy quantity, the current physical state value, and the preset fuzzy rule base to obtain an aggregated fuzzy set; The aggregated fuzzy set is defuzzified to determine the physical control signal value, and the physical device is driven to perform corresponding actions based on the physical control signal value.

2. The embedded fuzzy speech control method based on language operator quantization according to claim 1, characterized in that, It also includes methods for generating aggregated fuzzy sets: The current physical state value is fuzzified to generate a fuzzy quantity for the current state. The rule excitation intensity and rule consequent fuzzy set are determined based on the input fuzzy quantity, the current state fuzzy quantity, and the preset fuzzy rules. Based on the excitation intensity of the rule, a truncated fuzzy set of the rule consequent is truncated to generate a truncated fuzzy set; All truncated fuzzy sets are superimposed and aggregated to obtain an aggregated fuzzy set.

3. The embedded fuzzy speech control method based on language operator quantization according to claim 1, characterized in that, The specific methods for defuzzification calculation include: Determining precise numerical points based on aggregated fuzzy sets; Determine the percentage control increment based on precise numerical points; The register update value is calculated by combining the percentage control increment with the preset hardware timer period value; The physical control signal values ​​are generated based on the register update values.

4. The embedded fuzzy speech control method based on language operator quantization according to claim 1, characterized in that, It also includes user intent verification methods: After driving the physical device to perform corresponding actions based on the physical control signal value, the user's silent state data is collected during the preset silent period; The initial confidence level of user concern regarding the execution result is determined based on silent state data. When the initial confidence level is lower than the preset confidence level threshold, a reverse fine-tuning signal is generated based on the physical control signal value and the preset reverse amplitude coefficient. The system drives the physical device to perform a reverse action based on the reverse fine-tuning signal, and collects the user's corrective actions within a preset observation period after the reverse action is performed. The corrective actions are used to determine the user's true satisfaction with the physical control signal values, and the user intent model is updated based on the true satisfaction values.

5. The embedded fuzzy speech control method based on language operator quantization according to claim 4, characterized in that, It also includes a method for determining the initial confidence level: Collect user gaze duration, head orientation angle, and facial expression features; The user's attention concentration coefficient is calculated based on the duration of gaze and the angle of head orientation. The facial expression attention coefficient is obtained based on facial expression features; The initial attention confidence score is obtained by weighting and fusing the attention concentration coefficient and the facial expression attention coefficient.

6. The embedded fuzzy speech control method based on language operator quantization according to claim 4, characterized in that, It also includes methods for generating inverse fine-tuning signals: Collect ambient light and noise levels of the environment in which the controlled equipment is located; The level of perception masking is determined based on ambient brightness and ambient noise values. The base inverse amplitude value is determined based on the perceived masking level; The initial reverse fine-tuning amount is calculated based on the physical control signal value and the basic reverse amplitude value. Collect users' historical correction sensitivity data; Determine the individual sensitivity coefficient for each user based on historical correction sensitivity data; The final reverse fine-tuning amount is obtained by combining the initial reverse fine-tuning amount and the individual sensitivity coefficient; The inverse fine-tuning signal is generated based on the final inverse fine-tuning amount.

7. The embedded fuzzy speech control method based on language operator quantization according to claim 1, characterized in that, It also includes methods for perceiving unconscious intentions: After driving the physical equipment to perform corresponding actions based on the physical control signal value, the baseline physiological parameters of the user are collected within the preset stable period; A small-amplitude reverse disturbance signal is generated based on the physical control signal value and the preset perturbation amplitude coefficient; The physical device is driven to perform a micro-reverse action based on the micro-amplitude reverse perturbation signal, and the user's real-time physiological response parameters are collected during the execution of the micro-reverse action. The alertness response coefficient is determined based on the difference between real-time physiological response parameters and baseline physiological parameters. When the alertness response coefficient exceeds the preset alertness threshold, the alertness response coefficient is associated with the physical control signal value and stored in the user intent model.

8. The embedded fuzzy speech control method based on language operator quantization according to claim 7, characterized in that, It also includes methods for determining the alertness response coefficient: Pupil and respiratory change sequences were extracted from real-time physiological response parameters. Based on baseline physiological parameters, baseline pupillary and respiratory values ​​are obtained; The pupil change sequence was compared with the baseline pupil value to obtain the pupil fluctuation peak and pupil response delay; The respiratory change sequence was compared with the baseline respiratory value to obtain the respiratory fluctuation amplitude and respiratory recovery time; The alertness response coefficient is obtained by combining the peak pupillary fluctuation, pupillary response delay, respiratory fluctuation amplitude, and respiratory recovery time.

9. The embedded fuzzy speech control method based on language operator quantization according to claim 8, characterized in that, It also includes an adaptive update method for the perturbation amplitude coefficients: Collect the historical alert coefficient and corresponding historical amplitude coefficient during the execution of historical micro-amplitude reverse disturbance actions; The direction of amplitude adjustment for each perturbation is determined by comparing the historical alert coefficient with the preset ideal response range. The amplitude adjustment value is generated based on the direction of amplitude adjustment; The updated amplitude coefficient is calculated by combining the historical amplitude coefficient and the amplitude coefficient adjustment value, and then stored in the user intent model.

10. An embedded fuzzy speech control system based on language operator quantization, characterized in that, include: The acquisition module is used to acquire voice commands and current physical status values; A memory for storing a program that implements an embedded fuzzy speech control method based on language operator quantization as described in any one of claims 1 to 9; The processor is used to load and execute programs stored in memory.

Images