Voice control interaction system of water machine

Abstract

The invention relates to a water machine voice control interaction system, which belongs to the technical field of voice recognition and intelligent control, and comprises an information management unit, an acoustic environment evaluation unit, an interaction safety evaluation unit, a comprehensive risk decision-making unit and a closed-loop correction execution unit. Data such as a signal sound pressure level, environmental noise, a user distance and room reverberation time are substituted into an acoustic recognition probability equation, and a basic recognition probability is calculated, so that the initial influence of a physical environment is quantified, a real-time acoustic robustness index is formed, and the evolution capability from passive short-term correction to active long-term optimization is formed; and the system can be ensured to keep long-term, efficient and stable operation in a continuously changing environment.

Description

Technical Field

[0001] The present invention relates to the technical field of voice recognition and intelligent control, and in particular to a water machine voice control interactive system. Background Art

[0002] In specific professional environments, such as sterile isolation wards in hospitals, the application of voice control technology faces challenges that traditional technologies cannot address. Conventional voice interaction systems typically focus on improving speech recognition accuracy. Their technical solutions primarily optimize for general acoustic environments and fail to fully consider the complex risks in specific application scenarios. This situation is primarily due to the following technical limitations: Traditional voice control systems lack effective mechanisms for assessing the additional risks inherent in interactive behaviors, such as potential infection exposure caused by user movement or additional actions due to interaction failures in medical settings. System decisions are based solely on the success rate of acoustic model matching and are uncoupled from environmental safety constraints, particularly biosafety. Acoustic conditions in medical environments are complex and variable, with interference from equipment noise, room reverberation, and human activity. Existing technologies typically employ fixed noise suppression or signal enhancement algorithms, which are poorly adaptable to real-time, dynamically changing acoustic interference and struggle to maintain robust speech recognition. In the face of interaction failures, traditional systems often resort to simple strategies such as repeated requests or interruptions. This approach lacks flexibility and fails to implement graded, intelligent corrective measures based on risk severity, resulting in suboptimal system reliability and user experience in critical application scenarios.

[0003] As a result, existing voice control technologies are difficult to trust in medical scenarios that require high reliability and safety. On the one hand, simply pursuing recognition rate may sacrifice sensor control safety. On the other hand, overly conservative interaction strategies can lead to frequent system failures, undermining their value as efficient and convenient tools, and failing to meet the urgent needs of medical staff for reliable, safe, and contactless operation in sterile environments.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide a water machine voice control interactive system to solve the problems raised in the above background technology.

[0006] The technical solution of the present invention is: comprising an information management unit, an acoustic environment assessment unit, an interactive safety assessment unit, a comprehensive risk decision unit and a closed-loop correction execution unit; The information management unit is used to collect acoustic environment data and interactive behavior data, and send the acoustic environment data to the acoustic environment assessment unit, and send the interactive behavior data to the interactive safety assessment unit; The acoustic environment assessment unit is used to perform an acoustic environment adaptability analysis on the acoustic environment data to obtain a basic recognition probability, and the interaction safety assessment unit is used to perform an interaction safety risk analysis on the interaction behavior data to obtain an operation safety index; The comprehensive risk decision unit is used to receive the basic recognition probability and the operation safety index, collect real-time acoustic robustness indicators, and track interaction failure count values; The comprehensive risk decision unit sets the interaction failure count value as an interaction failure penalty factor, modifies the basic recognition probability according to the acoustic robustness index to obtain a modified recognition probability, performs a comprehensive risk calculation based on the modified recognition probability, the operation safety index, and the interaction failure penalty factor to obtain a comprehensive risk index, compares and analyzes the comprehensive risk index with a preset risk threshold, and generates a correction instruction; The closed-loop correction execution unit is used to receive the correction instruction and execute corresponding interaction strategy adjustment.

[0007] Preferably, the acoustic environment data includes signal sound pressure level, ambient noise level, room reverberation time and user space distance; The acoustic environment assessment unit performs the acoustic environment adaptability analysis in the following steps: Calculating a signal-to-noise ratio based on the signal sound pressure level and the ambient noise level; Calculating an acoustic attenuation factor based on the user space distance, and calculating a frequency compensation factor based on the room reverberation time; The signal-to-noise ratio, the acoustic attenuation factor, and the frequency compensation factor are substituted into a preset acoustic recognition probability equation to obtain the basic recognition probability.

[0008] Preferably, the interaction behavior data includes the number of potential contacts and the duration of the interaction during the interaction process; The steps of the interactive safety assessment unit performing the interactive safety risk analysis are as follows: Multiplying the potential number of contacts by a preset contact transmission weight to obtain a contact risk value; Multiplying the interaction duration by a preset time exposure weight to obtain a time exposure risk value; The contact risk value and the time exposure risk value are added together, and the resulting sum is set as the operational safety index.

[0009] Preferably, the step of collecting the real-time acoustic robustness index by the comprehensive risk decision unit is: Extract acoustic features from real-time speech signals, calculate their weighted variance within a time window, and set the variance as an acoustic stability factor; Comparing the acoustic features with a preset reference feature baseline, calculating the degree of spectral distortion and time domain feature distortion, and setting the degree of spectral distortion and time domain feature distortion as the noise sensitivity index; Based on the real-time background noise and reverberation time, the environmental adaptation index is calculated; The acoustic stability factor, the noise sensitivity index and the environmental adaptability index together constitute the real-time acoustic robustness index.

[0010] Preferably, the step of the comprehensive risk decision unit correcting the basic recognition probability according to the real-time acoustic robustness index is: Obtaining an interaction efficiency loss factor based on a weighted summation of the acoustic stability factor, the noise sensitivity index, and the environmental adaptability index; Based on the operational safety index, a sterility assurance coefficient is calculated through a preset risk constraint function; The basic recognition probability, the interaction efficiency loss factor and the sterility assurance coefficient are multiplied, and the obtained calculation result is set as the modified recognition probability.

[0011] Preferably, the steps of the comprehensive risk decision unit performing the comprehensive risk calculation are: Comparing the corrected recognition probability with a preset speech recognition accuracy threshold to quantify a normalized recognition failure risk; Comparing the operation safety index with a preset upper limit of infection control risk to quantify and obtain a normalized infection control risk; The square sum of the normalized recognition failure risk and the normalized infection control risk is calculated, and the calculation result is multiplied by the interaction failure penalty factor, and the obtained product is set as the comprehensive risk index.

[0012] Preferably, when the comprehensive risk index is in a preset medium risk state range, the correction instruction is a first-level correction instruction; The interaction strategy adjustment performed by the closed-loop correction execution unit after receiving the first-level correction instruction includes: improving the pickup gain of the microphone array, or enabling beamforming technology to enhance directional pickup, or enhancing the noise suppression depth of the adaptive noise cancellation algorithm.

[0013] Preferably, when the comprehensive risk index exceeds a preset high-risk state threshold, the correction instruction is a secondary correction instruction; The interaction strategy adjustment performed by the closed-loop correction execution unit after receiving the secondary correction instruction includes: enabling gesture recognition as an auxiliary input channel, or activating active voice feedback to request user secondary confirmation, or running redundant recognition algorithms in parallel and performing consistency judgment on the recognition results.

[0014] Preferably, the acoustic feature is a multi-dimensional acoustic feature vector; The multidimensional acoustic feature vector includes fundamental frequency, formant frequency, Mel-frequency cepstral coefficient and short-time energy; The acoustic stability factor is comprehensively evaluated by calculating the weighted variance of each feature in the multi-dimensional acoustic feature vector within a preset time.

[0015] Preferably, the comprehensive risk decision-making unit is further used to: A comprehensive performance index is obtained by performing a weighted sum calculation based on the modified recognition probability and the operation safety index recorded in the historical period; An exponential smoothing model is used to track the comprehensive performance index and establish a performance evolution trajectory; When the performance evolution trajectory is lower than the preset performance target value, the system underlying model parameters are fine-tuned using historical interaction data.

[0016] The present invention provides a water machine voice control interactive system through improvement, which has the following improvements and advantages compared with the prior art: 1. This invention not only assesses the impact of the acoustic environment on recognition success rate but also innovatively introduces an assessment of the safety risk of interactive behavior. By substituting data such as signal sound pressure level, ambient noise, user distance, and room reverberation time into the acoustic recognition probability equation, a basic recognition probability is calculated, thereby quantifying the initial impact of the physical environment. 2. By combining the number of potential contacts and duration of interactions with pre-set contact transmission weights and time exposure weights, an operational safety index is calculated, transforming abstract infection control regulations into quantifiable risk indicators for the first time. 3. The system can collect speech signals in real time and calculate the acoustic stability factor, noise sensitivity index, and environmental adaptability index by analyzing their multidimensional acoustic feature vectors, such as fundamental frequency, formant frequency, Mel-frequency cepstral coefficients, and short-term energy. These together constitute the real-time acoustic robustness index. 4. It can calculate a comprehensive performance index based on historical data and use an exponential smoothing model to establish a performance evolution trajectory to gain insight into long-term performance trends. It can automatically trigger optimization tasks and use historical interaction data to fine-tune the underlying acoustic-sensing coupling model or robustness prediction model parameters. This evolutionary capability from passive short-term correction to active long-term optimization ensures that the system can maintain long-term, efficient, and stable operation in a constantly changing environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention will be further explained below in conjunction with the accompanying drawings and Examples: Figure 1 It is a flow chart of the system of the present invention. DETAILED DESCRIPTION

[0018] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0019] Example 1: See also Figure 1 , the present invention provides a technical solution for a water machine voice control interactive system, comprising: an information management unit, an acoustic environment assessment unit, an interactive safety assessment unit, a comprehensive risk decision unit, and a closed-loop correction execution unit; The information management unit is used to collect acoustic environment data and interactive behavior data, and send the acoustic environment data to the acoustic environment assessment unit, and send the interactive behavior data to the interactive safety assessment unit; The acoustic environment assessment unit is used to perform acoustic environment adaptability analysis on acoustic environment data to obtain basic recognition probability. The interactive safety assessment unit is used to perform interactive safety risk analysis on interactive behavior data to obtain an operation safety index. The comprehensive risk decision unit is used to receive the basic recognition probability and operation safety index, collect real-time acoustic robustness indicators, and track the interaction failure count value; Here, interaction failure is defined as the occurrence of any of the following situations: Plus One: The confidence level returned by the speech recognition engine is lower than the preset minimum confidence threshold; the user explicitly issues a cancellation instruction through voice or gesture; under the secondary correction strategy, the system actively requests the user to confirm again, but the user denies it; The comprehensive risk decision unit sets the interaction failure count value as the interaction failure penalty factor, modifies the basic recognition probability according to the acoustic robustness index to obtain a modified recognition probability, calculates the comprehensive risk based on the modified recognition probability, the operation safety index, and the interaction failure penalty factor, and compares and analyzes the comprehensive risk index with the preset risk threshold to generate a correction instruction; The closed-loop correction execution unit is used to receive correction instructions and execute corresponding interaction strategy adjustments.

[0020] In this embodiment, a water machine voice control interactive system realizes all-round, closed-loop risk control of the voice interaction process by setting up an information management unit, an acoustic environment assessment unit, an interactive safety assessment unit, a comprehensive risk decision unit and a closed-loop correction execution unit; the information management unit serves as the data entry and is responsible for collecting acoustic environment data in sterile environments such as isolation wards and interactive behavior data of medical staff, and distributing the two types of data to corresponding assessment units respectively; the acoustic environment assessment unit and the interactive safety assessment unit work in parallel, quantifying the initial risk from the two dimensions of physical environment and operational behavior, and outputting the basic recognition probability and operational safety index; the comprehensive risk decision unit serves as the data entry and is responsible for collecting acoustic environment data in sterile environments such as isolation wards and interactive behavior data of medical staff, and distributing the two types of data to corresponding assessment units respectively; the acoustic environment assessment unit and the interactive safety assessment unit work in parallel, quantifying the initial risk from the two dimensions of physical environment and operational behavior, and outputting the basic recognition probability and operational safety index respectively ... As the decision-making core of the system, it not only receives the above-mentioned initial evaluation results, but also monitors the dynamically changing acoustic robustness indicators in real time and records continuous interaction failures. This unit fuses and calculates information from all dimensions to generate a comprehensive risk index that can fully reflect the current interaction status. The closed-loop correction execution unit executes corresponding strategy adjustments based on the correction instructions generated by the comprehensive risk decision-making unit to form a complete perception-assessment-decision-correction closed loop. This system architecture ensures that voice interaction can maintain a high recognition success rate in a complex and changing medical environment while strictly complying with infection control requirements, thereby improving the safety and reliability of water machine voice control in specific scenarios.

[0021] Example 2 Acoustic environment data includes signal sound pressure level, ambient noise level, room reverberation time, and user spatial distance; The steps for the acoustic environment assessment unit to conduct acoustic environment adaptability analysis are as follows: The signal-to-noise ratio is calculated based on the signal sound pressure level and the ambient noise level; The acoustic attenuation factor is calculated based on the user space distance, and the frequency compensation factor is calculated based on the room reverberation time; Substitute the signal-to-noise ratio, acoustic attenuation factor, and frequency compensation factor into the preset acoustic recognition probability equation to obtain the basic recognition probability; To determine the key parameters in the above acoustic recognition probability equation, those skilled in the art can use the following experimental methods: Parameters The benchmark recognition probability under ideal conditions can be obtained by conducting a large number of tests on a standard speech library in an anechoic chamber, and its value is usually set between 0.98-0.99; the comprehensive sound pressure attenuation coefficient and signal-to-noise ratio adjustment coefficient By collecting real data in multiple representative acoustic environments and For example, in a typical medical setting, it might be possible to determine experimentally that The experience value range is 0.1-0.3, The experience value range is 0.5-1.5; Here, the number of potential contacts Refers to the number of times a user enters the device's preset safety distance during a single interaction task, such as the 0.5-meter range in front of the device sensed by infrared or ultrasonic sensors. Each entry into this area is considered a high-risk behavior that may cause contamination of the device surface and is counted. Interaction behavior data includes the number of potential contacts and interaction duration during the interaction process; The steps for the interactive safety assessment unit to conduct interactive safety risk analysis are as follows: Multiply the number of potential contacts by the preset contact transmission weight to obtain the contact risk value; Multiply the interaction duration by the preset time exposure weight to obtain the time exposure risk value; The contact risk value and the time exposure risk value are added together, and the resulting sum is set as the operational safety index; The above formula The weighting coefficients in the formula are not set arbitrarily, but are based on recognized infection control standards, for example, referring to hospital infection control guidelines; the weighting is set to quantify the relative risks of different behaviors; for example, if the guidelines believe that the risk of transmission from a potential physical contact is much higher than the risk of staying in the room for one more minute, then the weighting coefficients can be set to and The ratio of ; These values ​​can be calibrated by hospital infection management experts according to specific protection requirements; In this embodiment, the acoustic environment assessment unit performs an in-depth analysis of the acoustic environment data to accurately quantify the initial impact of the environment on speech recognition; the unit uses the collected signal sound pressure level and the ambient noise level to calculate the initial impact of the environment on speech recognition through the formula Calculate the signal-to-noise ratio, where is the signal-to-noise ratio, is the signal sound pressure level, is the ambient noise level), this signal-to-noise ratio is a key feature that determines speech clarity; the acoustic environment assessment unit calculates the acoustic attenuation factor and frequency compensation factor based on the user's spatial distance and room reverberation time; through a preset acoustic recognition probability equation , integrate these factors and calculate the basic recognition probability, where is the basic recognition probability, is the benchmark recognition probability under ideal conditions, is the comprehensive sound pressure attenuation coefficient, is the distance between the user and the device, is the signal-to-noise ratio adjustment coefficient, is the frequency compensation factor. At the same time, the interactive safety assessment unit quantifies the infection control risk of the interactive behavior data. This unit multiplies the potential number of contacts with the preset contact transmission weight, and multiplies the interaction duration with the preset time exposure weight. Finally, the formula The sum of the two is set as the operational safety index, where is the operational safety index, is the contact propagation weight, is the number of potential contacts, is the time exposure weight, is the duration of interaction; this process converts abstract medical standards into calculable values, providing a decision-making basis for the system to conduct voice interaction while ensuring the safety of sensory control.

[0022] Example 3 The steps for the comprehensive risk decision unit to collect real-time acoustic robustness indicators are as follows: Extract acoustic features from real-time speech signals, calculate their weighted variance within a time window, and set the variance as an acoustic stability factor; Compare the acoustic features with the preset reference feature baseline, calculate the spectral distortion and time domain feature distortion, and set it as the noise sensitivity index; Based on the real-time background noise and reverberation time, the environmental adaptation index is calculated; The acoustic stability factor, noise sensitivity index and environmental adaptability index are combined to form a real-time acoustic robustness index.

[0023] The steps for the comprehensive risk decision unit to modify the basic recognition probability based on the real-time acoustic robustness index are as follows: The interaction efficiency loss factor is obtained by weighted summation based on the acoustic stability factor, noise sensitivity index and environmental adaptability index; Based on the operational safety index, the sterility assurance coefficient is calculated through the preset risk constraint function; Perform a product operation on the basic recognition probability, the interaction efficiency loss factor and the sterility assurance coefficient, and set the calculated result as the modified recognition probability; In calculating the interaction efficiency loss factor When the weight It reflects the contribution of different robustness indicators to recognition failure. Its value can be obtained by regression analysis of historical interaction data. For example, if the mutation effect of noise is found The biggest impact on interaction failure can be given Larger weights, a typical set of weights might be .

[0024] Calculating the sterility assurance factor When the critical infection control risk threshold Set by medical institutions based on their infection control safety baseline, representing the maximum tolerable operational safety index; risk curve sensitivity parameter It is used to adjust the risk index when it exceeds the threshold. The severity of the coefficient drop, The larger the value, the more severe the system's punishment for behaviors that exceed the safety threshold. The empirical value range is between 1.5 and 3.0. In this embodiment, the comprehensive risk decision unit collects real-time acoustic robustness indicators to achieve dynamic assessment of interaction risks. This process extracts acoustic features from real-time speech signals and quantifies them from three dimensions: the acoustic stability factor is obtained by calculating the weighted variance of the acoustic features in the time window. By comparing the real-time acoustic features with the reference feature baseline, the degree of distortion of the spectrum and time domain is calculated to obtain the noise sensitivity index ; and calculate the environmental adaptation index based on real-time background noise and reverberation time ;These three indicators together constitute the real-time acoustic robustness index; Noise Sensitivity Index It can be quantified by calculating the logarithmic spectrum distance between the Mel frequency cepstral coefficient vector of the real-time speech signal and the preset reference speech MFCC vector recorded in a quiet environment. The larger the distance, the more serious the distortion. The higher the value, the better the environmental adaptation index. It can be defined as a normalization function, such as ,in is the background noise level measured in real time, is the background noise level at which the system works optimally, and the clip function limits the result to between 0 and 1; It fully reflects the quality of the current speech signal; after obtaining the robustness index, the comprehensive risk decision unit dynamically modifies the basic recognition probability based on these indicators; through the weighted summation method Calculate the interaction efficiency loss factor, where is the interaction efficiency loss factor, are the weight coefficients of each robustness index, which quantifies the recognition performance loss caused by signal quality degradation; in order to meet the requirements of coupled sensing and control, this unit is based on the operation safety index output by the interactive safety assessment unit. , through the preset risk constraint function The sterility assurance factor is calculated, where is the sterility assurance factor, is the risk curve sensitivity parameter, is the critical infection control risk threshold, this coefficient is used to constrain the interaction behavior; through the product operation , combining the basic recognition probability, the interaction efficiency loss factor and the sterility assurance coefficient to obtain a modified recognition probability that is closer to the real interaction scenario, where To correct the recognition probability, This correction process enables the system to adjust its expectations of recognition success rate in real time according to the ever-changing acoustic environment and strict sensory control standards, providing more accurate input for subsequent risk decision-making.

[0025] Example 4 The steps for the comprehensive risk decision-making unit to calculate the comprehensive risk are: Compare the corrected recognition probability with the preset speech recognition accuracy threshold to quantify the normalized recognition failure risk; Compare the operational safety index with the preset upper limit of infection control risk to quantify the normalized infection control risk; The square sum of the normalized recognition failure risk and the normalized infection control risk is calculated, and the result is multiplied by the interaction failure penalty factor. The resulting product is set as the comprehensive risk index. In the comprehensive risk index In the calculation of speech recognition accuracy threshold Set according to the criticality of the application scenario. For example, it can be set to 0.90 for query instructions and 0.95 for control instructions; the upper limit of infection control risk tolerance It is an absolutely insurmountable risk value defined by the infection control department; the interaction failure penalty coefficient This is a tuning parameter used to amplify the risk of consecutive failures. Its value can be set to 0.2-0.5 to ensure that the risk index of the system will increase significantly after 2-3 consecutive failures, thereby forcing the triggering of a higher level of correction strategy. In this embodiment, after obtaining the modified recognition probability, the comprehensive risk decision unit performs the final comprehensive risk calculation to generate a unified risk metric. The calculation process normalizes the two core risk dimensions: by comparing the modified recognition probability with the preset voice recognition accuracy threshold, the normalized recognition failure risk is quantified; at the same time, the operation safety index is compared with the preset upper limit of the sensory control risk to quantify the normalized sensory control risk. The system calculates the square sum of these two normalized risk values, which is intended to amplify the significant risk of any single dimension. The square sum result is multiplied by the interaction failure penalty factor obtained by tracking the interaction failure count value, and the resulting product is set as the comprehensive risk index, where is the comprehensive risk index, To correct the recognition probability, is the speech recognition accuracy threshold, is the operational safety index, The upper limit of infection control risk is is the interaction failure penalty coefficient, is the interaction failure count value; through this formula, the system not only balances the two major risks of recognition performance and sensor control security, but also introduces a penalty mechanism for continuous interaction failures, so that the comprehensive risk index can respond more strongly to continuous interaction failures, providing a reliable and sensitive decision-making basis for triggering the graded correction strategy.

[0026] Example 5 When the comprehensive risk index is in the preset medium risk range, the correction order is a level one correction order; After receiving the first-level correction instruction, the closed-loop correction execution unit performs the following interactive strategy adjustments: improving the pickup gain of the microphone array, or enabling beamforming technology to enhance directional pickup, or enhancing the noise suppression depth of the adaptive noise cancellation algorithm; When the comprehensive risk index exceeds the preset high-risk threshold, the correction order is a level 2 correction order; After receiving the secondary correction instruction, the closed-loop correction execution unit performs the following interaction strategy adjustments: enabling gesture recognition as an auxiliary input channel, or activating active voice feedback to request secondary confirmation from the user, or running redundant recognition algorithms in parallel and performing consistency judgment on the recognition results; Comprehensive risk index The status interval is divided into, for example, medium risk is , high risk is The thresholds are derived based on extensive offline simulation and field test calibration; these thresholds are designed to map the quantitative risk index to an actionable intervention level; When , it indicates that the recognition performance or sensor control safety of the system has deviated from the ideal state, but analysis shows that it is mainly due to the deterioration of the acoustic environment. At this time, there is a probability of more than 90% that the risk can be brought back to normal level through the first-level corrected acoustic enhancement method; when When the risk of a single dimension is too high, or the combined risk of the two is too significant, simple acoustic enhancement is no longer sufficient to ensure the reliability and security of the interaction. Therefore, a secondary correction strategy must be initiated, introducing a multimodal or strong confirmation mechanism to ensure the ultimate correctness of the operation. In this embodiment, the closed-loop correction execution unit is the operating terminal for realizing the adaptive capability of the system. It executes the graded correction strategy according to the comprehensive risk index value output by the comprehensive risk decision unit. When the comprehensive risk index is in the preset medium risk state range, for example, When the system determines that the current risk can still be controlled through acoustic optimization, it generates a first-level correction instruction; after receiving the first-level correction instruction, the closed-loop correction execution unit starts the acoustic enhancement module and executes interactive strategy adjustments including increasing the pickup gain of the microphone array by 10%-20%, enabling beamforming technology to narrow the pickup range to 60°-90° to focus on the user's sound source, or enhancing the suppression depth of the adaptive noise cancellation algorithm to 15-20dB; when the comprehensive risk index exceeds the preset high-risk state threshold, for example, When the system determines that simple acoustic enhancement is no longer sufficient to ensure the success rate and safety of the interaction, it immediately generates a secondary correction instruction; after receiving the secondary correction instruction, the closed-loop correction execution unit starts the multimodal fusion and redundancy strategy, and the executed interaction strategy adjustments include enabling gesture recognition as an auxiliary input channel, activating active voice feedback to repeat the instruction in a clear volume to request the user's second confirmation, or running multiple redundant recognition algorithms in parallel in the background and judging the consistency of the results; this hierarchical correction mechanism ensures that the system can intelligently and automatically select the most appropriate response measures according to the severity of the risk, and maximize the recovery and guarantee of the success rate and stability of voice interaction under the premise of ensuring sterility and safety.

[0027] Example 6 Acoustic features are multi-dimensional acoustic feature vectors; The multidimensional acoustic feature vector includes fundamental frequency, formant frequency, Mel-frequency cepstral coefficient and short-time energy; The acoustic stability factor is comprehensively evaluated by calculating the weighted variance of each feature in the multidimensional acoustic feature vector within a preset time; Acoustic stability factor Formula is designed to be a normalized inverse form; therefore, when the variance of each acoustic feature, that is, the instability of the speech signal, increases, the denominator increases, On the contrary, when the signal is stable and the variance is extremely small, The value of is close to 1. This design allows us to directly use the interaction efficiency loss factor χ in the subsequent calculation To intuitively quantify the loss caused by signal instability; In this embodiment, to ensure the accuracy and comprehensiveness of the acoustic robustness evaluation, the acoustic feature extracted by the system is a multi-dimensional acoustic feature vector; the vector specifically includes the fundamental frequency representing the tone information , the resonance peak frequency that reflects the characteristics of the vocal tract , Mel-frequency cepstral coefficients describing the spectral envelope, and short-time energy representing signal strength ; These features together constitute a detailed description of the physical properties of the speech signal; in calculating the acoustic stability factor When , the system does not view a feature in isolation, but comprehensively evaluates it by calculating the weighted variance of each component feature in the multidimensional acoustic feature vector within the preset time window. The calculation formula is ,in is the variance calculation, They are the weights of Mel-frequency cepstral coefficients, fundamental frequency, short-time energy and resonance peak frequency respectively; this method makes the evaluation results of the acoustic stability factor more robust, can fully reflect the clarity and smoothness of pronunciation, and provides solid underlying data support for the subsequent calculation of interaction efficiency loss.

[0028] Example 7 The integrated risk decision-making unit is also used to: Based on the correction recognition probability and operation safety index recorded in the historical period, a weighted summation is performed to obtain the comprehensive effectiveness index; An exponential smoothing model is used to track the comprehensive performance index and establish the performance evolution trajectory; When the performance evolution trajectory is lower than the preset performance target value, the system's underlying model parameters are fine-tuned using historical interaction data; The model fine-tuning here is an automated supervised learning process. The system will accumulate historical interaction data, including acoustic feature vectors, environmental parameters, and corresponding corrected recognition probabilities. , Operational Safety Index and the final interaction result success or failure label as the training set; the goal of fine-tuning is to optimize the weight parameters in the comprehensive risk decision unit, such as , k, etc.; optimization algorithms such as gradient descent can be used to minimize a long-term loss function, which is designed to penalize those historically low efficiency, i.e., low comprehensive efficiency index. For example, by assigning higher weights to data samples that lead to interaction failures, the parameters in the risk calculation model are retrained and slightly updated, so that the model can better adapt to the current environment and achieve self-evolution; In this embodiment, the comprehensive risk decision-making unit not only performs real-time risk assessment and decision-making, but also undertakes the task of optimizing the long-term performance of the system; the unit performs retrospective analysis of historical data over a long period of time; it is based on the modified recognition probability and operation safety index recorded in the historical period, and uses a weighted summation method to The comprehensive performance index is calculated, where is the comprehensive performance index, are the weights of recognition performance and sensor control safety respectively; in order to smooth short-term fluctuations and gain insights into long-term trends, the system uses an exponential smoothing model To track the comprehensive performance index, thus establishing a performance evolution trajectory, where is the current performance evolution trajectory value, The latest comprehensive performance index, is the target performance value, is the convergence rate, The system uses the stored historical interaction data, including acoustic features, environmental parameters, and interaction result labels, to fine-tune the relevant parameters in the acoustic-sensing-control coupling model or the acoustic robustness prediction model. This mechanism enables the system to learn from history and self-evolve, realizing a transition from passive short-term correction to active long-term optimization, ultimately ensuring the long-term, efficient, and stable operation of the entire voice interaction system in harsh environments such as isolation wards.

[0029] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A water machine voice control interactive system, characterized in that: include: Information management unit, acoustic environment assessment unit, interactive safety assessment unit, comprehensive risk decision-making unit, and closed-loop correction execution unit; The information management unit is used to collect acoustic environment data and interactive behavior data, and send the acoustic environment data to the acoustic environment assessment unit, and send the interactive behavior data to the interactive safety assessment unit; The acoustic environment assessment unit is used to perform an acoustic environment adaptability analysis on the acoustic environment data to obtain a basic recognition probability, and the interaction safety assessment unit is used to perform an interaction safety risk analysis on the interaction behavior data to obtain an operation safety index; The comprehensive risk decision unit is used to receive the basic recognition probability and the operation safety index, collect real-time acoustic robustness indicators, and track interaction failure count values; The comprehensive risk decision unit sets the interaction failure count value as an interaction failure penalty factor, modifies the basic recognition probability according to the acoustic robustness index to obtain a modified recognition probability, performs a comprehensive risk calculation based on the modified recognition probability, the operation safety index, and the interaction failure penalty factor to obtain a comprehensive risk index, compares and analyzes the comprehensive risk index with a preset risk threshold, and generates a correction instruction; The closed-loop correction execution unit is used to receive the correction instruction and execute corresponding interaction strategy adjustment.

2. A water machine voice control interactive system according to claim 1, characterized in that: The acoustic environment data includes signal sound pressure level, ambient noise level, room reverberation time, and user space distance; The acoustic environment assessment unit performs the acoustic environment adaptability analysis in the following steps: Calculating a signal-to-noise ratio based on the signal sound pressure level and the ambient noise level; Calculating an acoustic attenuation factor based on the user space distance, and calculating a frequency compensation factor based on the room reverberation time; The signal-to-noise ratio, the acoustic attenuation factor, and the frequency compensation factor are substituted into a preset acoustic recognition probability equation to obtain the basic recognition probability.

3. The water machine voice control interactive system according to claim 1, characterized in that: The interaction behavior data includes the number of potential contacts and the duration of the interaction during the interaction; The steps of the interactive safety assessment unit performing the interactive safety risk analysis are as follows: Multiplying the potential number of contacts by a preset contact transmission weight to obtain a contact risk value; Multiplying the interaction duration by a preset time exposure weight to obtain a time exposure risk value; The contact risk value and the time exposure risk value are added together, and the resulting sum is set as the operational safety index.

4. The water machine voice control interactive system according to claim 1, characterized in that: The steps of the comprehensive risk decision-making unit collecting the real-time acoustic robustness index are: Extract acoustic features from real-time speech signals, calculate their weighted variance within a time window, and set the variance as an acoustic stability factor; Comparing the acoustic features with a preset reference feature baseline, calculating the degree of spectral distortion and time domain feature distortion, and setting the degree of spectral distortion and time domain feature distortion as the noise sensitivity index; Based on the real-time background noise and reverberation time, the environmental adaptation index is calculated; The acoustic stability factor, the noise sensitivity index and the environmental adaptability index together constitute the real-time acoustic robustness index.

5. The water machine voice control interactive system according to claim 4, characterized in that: The step of the comprehensive risk decision unit correcting the basic recognition probability according to the real-time acoustic robustness index is as follows: Obtaining an interaction efficiency loss factor based on a weighted summation of the acoustic stability factor, the noise sensitivity index, and the environmental adaptability index; Based on the operational safety index, a sterility assurance coefficient is calculated through a preset risk constraint function; The basic recognition probability, the interaction efficiency loss factor and the sterility assurance coefficient are multiplied, and the obtained calculation result is set as the modified recognition probability.

6. The water machine voice control interactive system according to claim 1, characterized in that: The steps of the comprehensive risk decision-making unit performing the comprehensive risk calculation are: Comparing the corrected recognition probability with a preset speech recognition accuracy threshold to quantify a normalized recognition failure risk; Comparing the operation safety index with a preset upper limit of infection control risk to quantify and obtain a normalized infection control risk; The square sum of the normalized recognition failure risk and the normalized infection control risk is calculated, and the calculation result is multiplied by the interaction failure penalty factor, and the obtained product is set as the comprehensive risk index.

7. The water machine voice control interactive system according to claim 1, characterized in that: When the comprehensive risk index is in the preset medium risk state range, the correction instruction is a first-level correction instruction; The interaction strategy adjustment performed by the closed-loop correction execution unit after receiving the first-level correction instruction includes: improving the pickup gain of the microphone array, or enabling beamforming technology to enhance directional pickup, or enhancing the noise suppression depth of the adaptive noise cancellation algorithm.

8. The water machine voice control interactive system according to claim 1, characterized in that: When the comprehensive risk index exceeds a preset high-risk state threshold, the correction instruction is a secondary correction instruction; The interaction strategy adjustment performed by the closed-loop correction execution unit after receiving the secondary correction instruction includes: enabling gesture recognition as an auxiliary input channel, or activating active voice feedback to request user secondary confirmation, or running redundant recognition algorithms in parallel and performing consistency judgment on the recognition results.

9. The water machine voice control interactive system according to claim 4, characterized in that: The acoustic feature is a multi-dimensional acoustic feature vector; The multidimensional acoustic feature vector includes fundamental frequency, formant frequency, Mel-frequency cepstral coefficient and short-time energy; The acoustic stability factor is comprehensively evaluated by calculating the weighted variance of each feature in the multi-dimensional acoustic feature vector within a preset time.

10. The water machine voice control interactive system according to claim 1, characterized in that: The comprehensive risk decision-making unit is also used to: A comprehensive performance index is obtained by performing a weighted sum calculation based on the modified recognition probability and the operation safety index recorded in the historical period; An exponential smoothing model is used to track the comprehensive performance index and establish a performance evolution trajectory; When the performance evolution trajectory is lower than the preset performance target value, the system underlying model parameters are fine-tuned using historical interaction data.

Images