Massage chair voice interaction control system

Abstract

The application discloses a massage chair voice interaction control system and relates to the technical field of massage chair voice interaction, which comprises a multi-source sensing data acquisition module, a voice signal preprocessing module, a voice interaction data processing module and an execution control management module, wherein the voice interaction data processing module comprises a credibility unit, an instruction duration unit and a multi-target massage parameter analysis unit. The three units are combined to realize quantitative evaluation and differential response of voice instruction quality. The credibility unit fuses multi-source information such as semantics, acoustic environment and pronunciation stability to evaluate instruction credibility. The instruction duration unit evaluates the proportion of effective signals in a voice segment. The multi-target massage parameter analysis unit comprehensively controls and adjusts massage intensity, regional priority and hot compress temperature.

Description

Technical Field

[0001] This invention relates to the field of voice interaction technology for massage chairs, specifically to a voice interaction control system for massage chairs. Background Technology

[0002] As a home health appliance with an increasingly widespread adoption, massage chairs have evolved from simple mechanical kneading in the early days to comprehensive health devices that integrate multiple functions such as kneading, tapping, airbag compression, and heat therapy. In terms of control methods, massage chairs have evolved from physical buttons and remote controls to touch screens and mobile apps. In recent years, with the maturity of voice recognition technology and the decrease in the cost of offline voice modules, voice interaction control has become a standard feature of mid-to-high-end massage chairs.

[0003] The basic working mode of voice-controlled massage chairs currently on the market is as follows: the user issues a voice command, the built-in voice recognition module converts the voice into text, and the system drives the corresponding actuator to execute the preset massage program according to the predefined command action mapping table. The essence of this mode is a linear mapping between voice commands and fixed actions. If the command is recognized, it will be executed; if it is not recognized, it will not be executed.

[0004] Furthermore, the one-dimensional evaluation of command quality during voice interaction leads to a lack of flexibility in execution decisions. Existing systems mostly process voice commands in two states: recognition and non-recognition, without distinguishing the confidence level of the command. The same command, whether clearly spoken in a quiet bedroom or casually said in a noisy living room, will be executed by the system with the same force and manner. This binary decision-making mechanism severely degrades the user experience in noisy environments, while also wasting the system's ability to provide more accurate responses when the command is clear.

[0005] Furthermore, speech activity detection may not be able to adapt to varying acoustic environments. Traditional speech activity detection often uses fixed energy thresholds or zero-crossing rate thresholds. Once set, these thresholds can result in significant differences in performance under different signal-to-noise ratios. In quiet environments, fixed thresholds can accurately detect the start and end boundaries of speech, but in environments with background noise, the energy of the noise segment may exceed the threshold, causing speech activity detection to misjudge noise as speech or elongate the speech boundary, resulting in the system receiving pseudo-instructions mixed with a large amount of noise. Existing systems may lack adaptive adjustment mechanisms. Summary of the Invention

[0006] The purpose of this invention is to provide a voice interaction control system for a massage chair, which solves the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a voice interaction control system for a massage chair, comprising a voice interaction data processing module and an execution control management module;

[0008] The voice interaction data processing module is used to receive preprocessed credibility association data, instruction duration association data, and massage control association data;

[0009] The voice interaction data processing module:

[0010] The confidence coefficient of the instruction is output based on the confidence of the automatic speech recognition engine, the mean square energy of the speech segment, the mean square energy of the background noise, the system clock timestamp of the speech start time detected by the VAD algorithm, the time of the most recent exhalation peak detected by the chair back pressure sensor, and a fixed coefficient in the confidence-related data.

[0011] Based on the signal-to-noise ratio, confidence threshold, actual speech rate of the user's current voice command, and standard speech rate of the command template in the command duration correlation data, and combined with dynamic energy consumption benchmark value, the effective command duration weight is output.

[0012] Based on the time interval and forgetting half-life from the last effective voice command in the massage control associated data, the interaction history forgetting factor is analyzed. The massage intensity coefficient is calculated by using the system's preset minimum perceptible intensity, the system's preset maximum safe intensity, and the interaction history forgetting factor. The region priority probability distribution and the target temperature at the current moment are output by using the preference weight of the i-th region and the target temperature of the last decision, combined with the command credibility coefficient and the effective command duration weight.

[0013] The execution control management module:

[0014] It is used to obtain the massage intensity coefficient, the probability distribution of regional priority and the target temperature at the current moment, and to execute the control management module to control kneading, airbags and heat application.

[0015] Optionally, the voice interaction data processing module includes:

[0016] Credibility unit, instruction duration unit, and multi-target massage parameter analysis unit;

[0017] The voice interaction data processing module outputs:

[0018] The instruction credibility coefficient is used to distinguish the credibility level of instructions and has the ability to distinguish instruction quality. It is the entry quality label of the massage chair voice interaction control system.

[0019] The effective command duration weight is used to control the system's level of trust in the command length, connecting acoustic quality assessment and execution management decisions;

[0020] The massage intensity coefficient, the regional priority probability distribution, and the target temperature at the current moment determine the magnitude of the massage force, the direction of the massage force, and the intensity level of the heat application, respectively.

[0021] Optionally, the processing procedure for the credibility unit is as follows:

[0022] S11. After performing endpoint detection, acoustic feature extraction and decoding on the speech through the automatic speech recognition engine, the confidence score of the automatic speech recognition engine is output to provide a semantic-level trust basis.

[0023] S12. The mean square energy of the speech segment is obtained by calculating the square mean of the amplitudes of all sampling points within the speech segment, so as to quantify the strength of the user's speech signal.

[0024] S13. The mean square energy of the background noise is obtained by calculating the square mean of the amplitudes of all sampling points within the silent segment before the instruction, so as to quantify the intensity of the current environmental noise.

[0025] S14. The energy ratio is obtained by calculating the ratio of the mean square energy of the speech segment to the sum of the mean square energies of the speech segment and the background noise, so as to assess the degree of pollution of the speech signal by the background noise.

[0026] S15. The chair back pressure sensor detects the periodic changes in chest cavity pressure caused by breathing fluctuations and detects the local minimum point of the pressure waveform to obtain the moment when the chair back pressure sensor detected the most recent expiratory peak.

[0027] S16. The respiratory phase factor is obtained by dividing the difference between the system clock timestamp at the start of speech and the time when the chair back pressure sensor detected the most recent peak of exhalation by a fixed coefficient, in order to assess the impact of the user's speaking phase in the respiratory cycle on vocal cord stability.

[0028] S17. Multiply the confidence level of the automatic speech recognition engine, the energy ratio, and the breathing phase factor to output the command confidence coefficient.

[0029] Optionally, the processing procedure for the instruction duration unit is as follows:

[0030] S21. The signal-to-noise ratio is obtained by calculating the logarithmic form of the ratio of the mean square energy of the speech segment to the mean square energy of the background noise, so as to measure the relative intensity of the effective component and the background noise component in the speech signal.

[0031] S22. By taking the difference between the signal-to-noise ratio and a preset threshold as input, the signal-to-noise ratio factor is obtained to evaluate the degree of influence of the signal-to-noise ratio on the effective proportion of speech segments.

[0032] S23. By using the difference between the instruction credibility coefficient and the confidence threshold as input, the confidence factor is obtained to evaluate the degree of influence of instruction credibility on the effective proportion of speech segments.

[0033] S24. By calculating the ratio of the number of syllables in the text to the total duration of the speech segment, the actual speech rate of the user's current voice command is obtained, so as to quantify the speed at which the user speaks.

[0034] S25. By calculating the ratio of the difference between the user's actual speech rate and the standard speech rate of the instruction template to the standard speech rate, a speech rate imbalance coefficient is obtained to assess the degree of abnormality in the user's pronunciation state.

[0035] S26. Multiply the signal-to-noise ratio factor, confidence factor, and speech rate imbalance coefficient to output the effective instruction duration weight.

[0036] Optionally, the processing procedure of the multi-target massage parameter analysis unit is as follows:

[0037] S31. By calculating the exponential decay form of the ratio of the time interval from the last effective voice command to the forgetting half-life, the interaction history forgetting factor is obtained to quantify the degree of memory retention of user habits by the system.

[0038] S32. By introducing the minimum perceptible intensity, maximum safe intensity, and interaction history forgetting factor preset by the system, and combining the linear weighted result of the instruction credibility coefficient and the effective instruction duration weight, the massage intensity coefficient is output.

[0039] S33. By introducing preference weights for each massage area and combining the product of instruction credibility coefficient, effective instruction duration weight, and interaction history forgetting factor as the temperature scaling factor of the normalized exponential function, the region priority probability distribution is output.

[0040] S34. By introducing the product of the instruction credibility coefficient, the effective instruction duration weight, and the interaction history forgetting factor, and combining it with the target temperature of the previous decision, the target temperature at the current moment is output.

[0041] Optionally, the execution control management module includes a kneading motor control unit, an airbag control unit, and a heat therapy control unit.

[0042] Optionally, the kneading motor control unit specifically comprises:

[0043] Controlled objects: one kneading DC brushless motor on each side;

[0044] Mapping method: The massage intensity coefficient value is linearly mapped to the motor PWM duty cycle;

[0045] Specific control process:

[0046] The timer PWM module of the main control chip outputs a square wave with a specific frequency and duty cycle.

[0047] The square wave is amplified by the motor driver chip and then drives the kneading motor.

[0048] The kneading motor speed sensor provides feedback on the actual speed, which is then compared with the target speed. The controller then uses a closed-loop adjustment to regulate the PWM duty cycle, bringing the actual speed closer to the target value.

[0049] Optionally, the airbag control unit specifically comprises:

[0050] Controlled objects: one neck airbag, one shoulder airbag on each side, and one waist airbag, for a total of four independent air paths, each supplied with air by a solenoid valve and an air pump;

[0051] Mapping method: The regional priority probability distribution represents the proportion of working time of each region within a control cycle;

[0052] The specific control process is as follows: First, the control cycle is set, and the working time of each area is allocated based on the probability distribution of regional priority. Within one cycle, the solenoid valves of the three areas are opened in turn in a polling manner, and the opening time is allocated proportionally. During the opening of the solenoid valve, the air pump works at a constant power to inflate the airbag of the corresponding area to the set pressure.

[0053] The heat therapy control unit is specifically:

[0054] Controlled objects: Flexible heating pads distributed in the neck, shoulder, and waist areas;

[0055] Mapping method: The target temperature at the current moment is mapped to the target value of the temperature controller;

[0056] Specific control process:

[0057] An NTC thermistor is attached next to each heating element to read the current temperature value;

[0058] The PID controller uses the current target temperature as the target value and the current temperature value as the feedback value to calculate the PWM control quantity.

[0059] The PWM signal drives the switching transistor to control the on-time ratio of the heating element;

[0060] Once the current temperature reaches the target temperature and stabilizes, the PID controller enters the heat preservation state.

[0061] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0062] I. This invention constructs a multi-source information fusion instruction credibility evaluation model by multiplying the confidence level output by the speech recognition engine, the ratio of speech segment energy to noise segment energy, and the breathing phase factor through a credibility unit. The speech recognition engine confidence level provides a semantic-level trust basis, reflecting the degree of confirmation of the recognition result by the speech recognition engine. The ratio of speech segment energy to noise segment energy provides acoustic-level correction, reflecting the separation degree of the speech signal relative to background noise. The breathing phase factor detects the user's breathing waveform through a chair back pressure sensor, assesses the stability of the user's vocal cords while speaking, and reflects the user's current breathing phase. Whether the instruction is delivered stably during speech, the calculation logic of the product of these three factors determines that insufficient information in any one dimension will significantly reduce the overall credibility. The speech recognition engine may understand the instruction, but it cannot fully trust it in a noisy environment. Similarly, it cannot fully trust it in a quiet environment if the speech recognition engine has low confidence. Even if both are good, the intelligibility of the instruction will still decrease if the user speaks with deep inhalation, causing airflow interference to the vocal cords. The introduction of this unit enables the system to distinguish between high-quality instructions with good environment and good pronunciation and low-quality instructions with poor environment or poor pronunciation. This provides a reliable quantitative basis for the differentiated response of the subsequent actuator and solves the technical problem of the lack of a multi-source information fusion evaluation mechanism for instruction credibility.

[0063] II. This invention, through an instruction persistence unit, constructs a multi-dimensional model for evaluating the proportion of effective signal in detected speech segments by multiplying the signal-to-noise ratio (SNR) factor, confidence factor, and speech rate mismatch coefficient. The SNR factor evaluates the boundary clarity of the speech signal based on the SNR. A high SNR indicates good separation between speech and noise, and reliable start and end boundaries detected by speech activity detection. A low SNR makes it difficult to distinguish between speech and noise, and speech activity detection may misclassify noise segments as speech. The confidence factor performs secondary verification of the speech activity detection results based on the instruction confidence output by the confidence unit. The speech rate mismatch coefficient evaluates the degree of speech rate mismatch. The integrity of the temporal structure of a speech segment is crucial. When the speech rate is normal, the syllable distribution of the speech segment is uniform and the temporal structure is complete. When the speech rate is too fast or too slow, the temporal structure of the speech segment is abnormal, which may contain compressed or dragging invalid components. The logic of multiplying three factors determines that the speech segment detected by speech activity detection is considered to be almost entirely effective only when all three conditions are met: clear signal boundaries, reliable instructions, and normal pronunciation temporal structure. If any one of these conditions is not met, the effective percentage will be significantly reduced. This unit enables the system to quantitatively evaluate the quality of speech segments, rather than accepting or rejecting all the detection results of speech activity detection. It solves the technical problem that the effective signal percentage of speech activity detection cannot be adaptively evaluated.

[0064] Third, this invention, through a multi-objective massage parameter analysis unit, achieves refined adaptive adjustment of massage intensity, regional priority probability distribution, and target temperature of the hot compress through the combined effect of the interaction history forgetting factor, instruction credibility, and effective instruction duration weight. The regional priority probability distribution maps the product of instruction credibility, effective instruction duration weight, and interaction history forgetting factor to the probability distribution of each region through a normalized exponential function. When the product of the three factors is high, the output of the normalized exponential function tends to be an extreme distribution, and the system accurately allocates massage resources to the user-specified region according to the preset regional preference weight. When the product of the three factors is low, the output of the normalized exponential function tends to be a uniform distribution, and the system does not favor any specific region but uniformly cares for all regions. The target temperature of the hot compress is implemented through a discrete state machine with hysteresis. The system now features a temperature setting safety switching mechanism based on the product of three factors: instruction reliability, effective instruction duration weight, and interaction history forgetting factor. When the product is high, the system provides high-temperature heat therapy combined with high-intensity massage for deep soothing effects. When the product is medium, the system provides medium-temperature heat therapy for gentle soothing. When the product is low, the system maintains a low temperature to avoid unnecessary heating due to unreliable instructions. The hysteresis mechanism prevents the temperature from repeatedly jumping around near the setting boundary due to small fluctuations in the product of the three factors, ensuring the stability of the heat therapy temperature. The application of this unit allows the actuator parameters to simultaneously reflect whether the current instruction is worth fully trusting and whether the user is worth responding to aggressively. This gives the system a safety mechanism that distributes evenly when uncertain and executes precisely when certain, ensuring that the heat therapy temperature maintains a gradient distribution from conservative to aggressive, consistent with the quality of the instructions. Attached Figure Description

[0065] Figure 1 This is a structural block diagram of the voice interaction data processing module and the execution control management module of the present invention;

[0066] Figure 2 This is a schematic diagram of the system structure of the present invention;

[0067] Figure 3 This is a flowchart illustrating the operation of the voice interaction data processing module of the present invention. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] Please see Figures 1 to 3 This embodiment provides a voice interaction control system for a massage chair, including a multi-source sensor data acquisition module, a voice signal preprocessing module, a voice interaction data processing module, and an execution control management module;

[0070] Among them: Multi-source sensor data acquisition module:

[0071] Used to obtain credibility correlation data, command persistence correlation data, and massage control correlation data during voice interaction with the massage chair;

[0072] Among them: Speech signal preprocessing module:

[0073] It is used to receive credibility correlation data, instruction duration correlation data, and massage control correlation data, and to preprocess the data;

[0074] Among them: the voice interaction data processing module includes a credibility unit, an instruction duration unit, and a multi-target massage parameter analysis unit;

[0075] Firstly, the credibility unit outputs the command credibility coefficient based on the confidence of the automatic speech recognition engine, the mean square energy of the speech segment, the mean square energy of the background noise, the system clock timestamp of the speech start time detected by the VAD algorithm, the time of the most recent exhalation peak detected by the chair back pressure sensor, and a fixed coefficient in the credibility association data.

[0076] Secondly, the instruction duration unit outputs the effective instruction duration weight based on the signal-to-noise ratio, confidence threshold, actual speech rate of the user's current voice instruction, and standard speech rate of the instruction template in the instruction duration association data, combined with the dynamic energy consumption benchmark value.

[0077] Third, the multi-target massage parameter analysis unit analyzes the interaction history forgetting factor based on the time interval and forgetting half-life of the massage control associated data from the last effective voice command. It calculates the massage intensity coefficient by using the system's preset minimum perceptible intensity, the system's preset maximum safe intensity, and the interaction history forgetting factor. It outputs the region priority probability distribution and the target temperature at the current moment by using the preference weight of the i-th region and the target temperature of the last decision, combined with the command credibility coefficient and the effective command duration weight.

[0078] Among them: Execution Control Management Module:

[0079] It is used to obtain the massage intensity coefficient, the area priority probability distribution and the target temperature at the current moment, and to execute the control management module to control kneading, airbags and hot compress.

[0080] The execution control management module includes a kneading motor control unit, an airbag control unit, and a heat therapy control unit.

[0081] The kneading motor control unit is specifically as follows:

[0082] Controlled objects: One brushless DC kneading motor on each side (driving the kneading ball mechanism to rotate, realizing kneading, tapping and other actions);

[0083] Mapping method: The massage intensity coefficient value is linearly mapped to the motor PWM duty cycle. For example, a massage intensity coefficient S = 0.635 corresponds to a PWM duty cycle of 63.5%.

[0084] Specific control process:

[0085] The timer PWM module of the main control chip outputs a square wave with a specific frequency and duty cycle.

[0086] The square wave is amplified by the motor driver chip and then drives the kneading motor.

[0087] The kneading motor speed sensor feeds back the actual speed, which is compared with the target speed. The controller then adjusts the PWM duty cycle in a closed loop to make the actual speed approach the target value.

[0088] The airbag control unit is specifically as follows:

[0089] Controlled objects: one neck airbag, one shoulder airbag on each side, and one waist airbag, for a total of four independent air paths, each supplied with air by a solenoid valve and an air pump;

[0090] Mapping method: The regional priority probability distribution represents the proportion of working time of each region within a control cycle (e.g., 60 seconds);

[0091] The specific control process is as follows: First, the control cycle is set, and the working time of each area is allocated based on the probability distribution of regional priority. Within one cycle, the solenoid valves of the three areas are opened in turn in a polling manner, and the opening time is allocated proportionally. During the opening of the solenoid valve, the air pump works at a constant power to inflate the airbag of the corresponding area to the set pressure.

[0092] The heat therapy control unit is specifically as follows:

[0093] Controlled objects: Flexible heating pads distributed in the neck, shoulder, and waist areas;

[0094] Mapping method: The target temperature at the current moment is mapped to the target value of the temperature controller;

[0095] Specific control process:

[0096] An NTC thermistor is attached next to each heating element to read the current temperature value;

[0097] The PID controller uses the current target temperature as the target value and the current temperature value as the feedback value to calculate the PWM control quantity.

[0098] The PWM signal drives the switching transistor to control the on-time ratio of the heating element;

[0099] Once the current temperature reaches and stabilizes at the target temperature, the PID controller enters the heat preservation state.

[0100] In this embodiment: the three sets of units form a complete closed loop from signal acquisition and quality assessment to execution mapping, and the output of each step is the input of the next step:

[0101] The credibility unit extracts the command credibility coefficient C from the raw audio and pressure sensor data;

[0102] The instruction duration unit is based on the instruction credibility coefficient C, and combines the signal-to-noise ratio and speech rate information to extract the proportion of effective instruction duration D;

[0103] The multi-target massage parameter analysis unit takes the instruction credibility coefficient C and the effective instruction duration weight D as inputs, and combines them with the interaction history forgetting factor NA to map them into three actuator parameters: massage intensity coefficient S, region priority probability distribution Z, and so on. i and the target temperature T at the current moment n ;

[0104] The overall effect of this progressive structure is that the system no longer executes upon hearing text, but first evaluates the quality of the instruction, and then determines the strength, direction and intensity of execution based on the quality level. The three parameters, instruction credibility coefficient C, effective instruction duration weight D and interaction history forgetting factor NA, describe the quality profile of the current interaction from different dimensions. That is, instruction credibility coefficient C describes the semantic and acoustic credibility of the instruction, effective instruction duration weight D describes the temporal integrity of the instruction, and interaction history forgetting factor NA describes the trust accumulation of the interaction history. The three dimensions together determine the responsiveness of the executor.

[0105] Continuous control of quality grading:

[0106] Traditional massage chair voice control only recognizes a binary state, equivalent to 0 or 1. The combination of three sets of units expands the control space from binary to continuous values. The command confidence coefficient C and the effective command duration weight D both change continuously in the [0,1] interval. The massage intensity coefficient S and the area priority probability distribution Z i It also changes continuously, which means that the system can handle a large number of intermediate states between completely clear and completely ambiguous, such as being interrupted by TV sound halfway through speaking, speaking too fast but the content is clear, and having a noisy background but complete instructions, and give a response that matches the quality of the instructions.

[0107] Safety-first strategy under uncertainty:

[0108] When the instruction quality is low (low instruction reliability coefficient C or low effective instruction duration weight D), the progressive effect of the three sets of formulas causes the system to automatically tend towards conservatism: the massage intensity coefficient S decreases (the intensity decreases), and the area priority probability distribution Z... i The target temperature T at the current moment tends towards uniformity (without gambling on the region). n Setting the temperature to low (no heating) is a necessary conservative approach to uncertainty, which is essential for devices like massage chairs that come into direct contact with the human body. On a massage chair, misoperation (pressing the wrong area or applying too much pressure) can result in immediate physical discomfort or even injury. The three units work in a progressive manner to ensure that the system selects the safest response in any uncertain situation.

[0109] Adaptive accumulation of historical trust:

[0110] The introduction of the interaction history forgetting factor (NA) endows the system with time-dimensional memory capabilities. The system's response strength differs depending on whether the user first powers on the device or after repeated interactions. This adaptive behavior simulates natural patterns in interpersonal interaction, allowing service personnel to better understand and execute commands more boldly for existing customers, while being more cautious with new customers. The system requires no explicit user configuration or learning process; it achieves adaptive adjustment of trust solely through the exponential decay of timestamps.

[0111] Coordinated control of multiple actuators:

[0112] The three actuators share the same set of inputs C×D×NA, but each uses a different mapping structure. This allows the massage intensity, area distribution, and heat application temperature to respond in a coordinated manner to changes in command quality. That is, when the command quality is high, the intensity is high, the area is precise, and the temperature is high; when the command quality is low, the intensity is low, the area is uniform, and the temperature is low. The user perceives that the system seems to know whether what I am saying is reliable or not, rather than each actuator acting independently.

[0113] refer to Figure 1 , Figure 2 and Figure 3 The processing flow for the credibility unit is as follows:

[0114] S11. After performing endpoint detection, acoustic feature extraction and decoding on the speech through the automatic speech recognition engine, the confidence score of the automatic speech recognition engine is output to provide a semantic-level trust basis.

[0115] S12. The mean square energy of the speech segment is obtained by calculating the square mean of the amplitudes of all sampling points within the speech segment, so as to quantify the strength of the user's speech signal.

[0116] S13. The mean square energy of the background noise is obtained by calculating the square mean of the amplitudes of all sampling points within the silent segment before the instruction, so as to quantify the intensity of the current environmental noise.

[0117] S14. The energy ratio is obtained by calculating the ratio of the mean square energy of the speech segment to the sum of the mean square energies of the speech segment and the background noise, so as to assess the degree of pollution of the speech signal by the background noise.

[0118] S15. The chair back pressure sensor detects the periodic changes in chest cavity pressure caused by breathing fluctuations and detects the local minimum point of the pressure waveform to obtain the moment when the chair back pressure sensor detected the most recent expiratory peak.

[0119] S16. The respiratory phase factor is obtained by dividing the difference between the system clock timestamp at the start of speech and the time when the chair back pressure sensor detected the most recent peak of exhalation by a fixed coefficient, in order to assess the impact of the user's speaking phase in the respiratory cycle on vocal cord stability.

[0120] S17. Multiply the confidence level, energy ratio, and breathing phase factor of the automatic speech recognition engine to output the command confidence coefficient.

[0121] The formula for calculating the credibility unit is as follows:

[0122] ;

[0123] in:

[0124] C refers to the instruction credibility coefficient, which represents the overall level of trust the system has in the user's current voice instruction. The value range is [0,1]. The closer the instruction credibility coefficient C is to 1, the more confident the system is that the user is indeed issuing a valid instruction and that the instruction quality is good enough. The closer the instruction credibility coefficient C is to 0, the more the system tends to think that the current audio segment may not be a valid instruction, or that the instruction is unreliable due to environmental interference. The instruction credibility coefficient C is the entry point of the entire recursive system and determines the starting point for the calculation of all subsequent parameters.

[0125] In simple terms, this calculation result is used to determine to what extent the system should trust this voice command under the current acoustic environment and the user's physiological state.

[0126] Casr refers to the confidence score of the automatic speech recognition engine. The microphone array on the massage chair (which can be installed on both sides of the headrest or the front of the armrest) can collect the user's voice signal and transmit it to the speech recognition module. After the automatic speech recognition (ASR) engine performs endpoint detection, acoustic feature extraction and decoding on the speech, in addition to outputting the recognized text, it will also output a confidence score between 0 and 1. This score is the result of the joint evaluation of the acoustic model and language model inside the ASR engine, representing the probability that the recognition result matches the acoustic signal.

[0127] In simple terms, the confidence score (Casr) of this automatic speech recognition engine is used to provide a semantic-level foundation of trust, indicating how certain the ASR engine is that it understands the speech.

[0128] It represents the energy ratio, which provides a correction at the acoustic environment level, that is, to what extent background noise contaminates speech;

[0129] Espe refers to the mean square energy of the speech segment. The unit depends on the reference voltage of the ADC, but it is ultimately used in the form of a ratio. The calculation formula is as follows:

[0130] ;

[0131] In the above formula, NS refers to the total number of sampling points in the speech segment (the number of frames from the start to the end of the speech detected by the VAD algorithm (voice activation detection algorithm) multiplied by the frame length).

[0132] In the above formula, Xspe refers to the amplitude of the nth sampling point in the speech segment (16-bit or 24-bit PCM audio data collected by the microphone).

[0133] Enoi refers to the mean square energy of the background noise, and is calculated using the following formula:

[0134] ;

[0135] In the above formula, NU refers to the number of sampling points corresponding to 200ms (e.g., 3200 points at a sampling rate of 16kHz).

[0136] In the above formula, Xnoi refers to the sampling point amplitude during the 200ms silent period before the instruction (i.e., before the VAD is triggered);

[0137] This refers to the respiratory phase factor, which is a physiological state-level correction. In other words, it refers to which phase of the breathing cycle the user is speaking, affecting the stability of the vocal cords and the clarity of commands.

[0138] Tzl refers to the system clock timestamp of the moment when the VAD algorithm detects the start of the speech, which is recorded by the speech processing thread;

[0139] Thqf refers to the moment when the chair back pressure sensor detected the most recent peak expiratory breath. Specifically:

[0140] The massage chair backrest contains an array of thin-film pressure sensors (for body shape detection and airbag pressure feedback). When a user lies down, the chest rises and falls with breathing, causing periodic changes in pressure at corresponding locations on the chair back.

[0141] During inhalation: The chest cavity expands, and the pressure of the thoracic spine on the chair back increases slightly;

[0142] During exhalation: the chest cavity contracts, and the pressure decreases;

[0143] The pressure sensor continuously acquires pressure values ​​at a sampling rate of 20-50Hz. The system detects the local minimum point of the pressure waveform through a sliding window (e.g., 2 seconds). This minimum point corresponds to the expiratory peak (the moment when the chest cavity retracts the most), and the timestamp of this moment is recorded as Thqf.

[0144] Tref stands for fixed coefficient, which is a fixed constant of 4 seconds, corresponding to the normal resting respiratory cycle of an adult. It is not measured in real time and is preset by the system.

[0145] The calculation formula in this unit uses a three-term multiplication. The significance of multiplying the three terms is that if any one condition is not met, the overall credibility will be reduced. If the ASR understands the message but the background is too noisy, it cannot be fully trusted. If the background is quiet but the ASR confidence is low, it also cannot be fully trusted. If both conditions are good, but the user is speaking at the deepest point of inhalation, the vocal cords may be affected by airflow and the instruction may be unclear. Only when all three conditions are met simultaneously will the instruction credibility coefficient C approach 1.

[0146] In existing technologies, all voice interaction systems assume that users can speak at any time and that the quality of speech is independent of physical condition. However, massage chairs have unique physical conditions, namely, the user lies on the chair, and the pressure sensor built into the backrest can monitor the breathing waveform in real time (the pressure increases during chest expansion and decreases during exhalation). When the user speaks at the end of exhalation, the diaphragm is in a relaxed position, the vocal cords are more stable, and the instructions are clearer. When speaking at the beginning of inhalation, the vocal cords are affected by airflow, and the quality of instructions decreases. Therefore, the breathing phase factor of this unit takes this part into account to improve the practicality of this system.

[0147] In this embodiment: the credibility unit establishes a multi-dimensional instruction quality evaluation entry point for the entire system. In traditional massage chair voice control, the system only has two states: recognition or non-recognition. All instructions output by the ASR engine are treated equally regardless of the acoustic environment. The contribution of the credibility unit is that it evaluates the trustworthiness of an instruction from three independent dimensions: the semantic confidence of the ASR engine, the degree of environmental noise pollution on speech, and the impact of the user's breathing phase on vocal cord stability. This makes the system no longer just do what it hears, but judge whether it is worth doing. This entry evaluation mechanism is the foundation of the entire recursive chain. All subsequent decisions depend on the quality of the instruction credibility coefficient C.

[0148] The purpose of the command credibility coefficient C is to distinguish the credibility level of the command. In the actual use scenario of a massage chair, a user may casually say "press" while watching TV, or clearly say "massage my shoulders" in a quiet bedroom. If the ASR engine recognizes both commands as "massage shoulders", the traditional system will give the exact same response. However, the values ​​of the command credibility coefficient C will be completely different. The former has a lower energy ratio due to background noise, and the breathing phase factor may be lowered when speaking, which may be in the inspiratory phase. As a result, the command credibility coefficient C is significantly lower than that of the latter. Based on this, the system makes differentiated processing in subsequent formulas to achieve a differentiated strategy of high response for high-quality commands and conservative response for low-quality commands. The existence of the command credibility coefficient C enables the system to distinguish the quality of commands, which is not available in traditional massage chair voice control.

[0149] The instruction reliability coefficient C is the entry quality label of the entire recursive system. It does not directly control any actuator, but its value determines the baseline of all subsequent calculation results. Without the instruction reliability coefficient C, the system can only treat all instructions the same and cannot achieve fine-grained control that varies according to the quality of the instructions. The significance of the instruction reliability coefficient C is that it transmits the quality information of the acoustic physical layer to the decision layer, so that the decision layer needs to know the substantive situation of the current instruction before making any judgment.

[0150] refer to Figure 1 , Figure 2 and Figure 3 The processing flow for the instruction duration unit is as follows:

[0151] S21. The signal-to-noise ratio is obtained by calculating the logarithmic form of the ratio of the mean square energy of the speech segment to the mean square energy of the background noise, so as to measure the relative intensity of the effective component and the background noise component in the speech signal.

[0152] S22. By taking the difference between the signal-to-noise ratio and a preset threshold as input, the signal-to-noise ratio factor is obtained to evaluate the degree of influence of the signal-to-noise ratio on the effective proportion of speech segments.

[0153] S23. By using the difference between the instruction credibility coefficient and the confidence threshold as input, the confidence factor is obtained to evaluate the degree of influence of instruction credibility on the effective proportion of speech segments.

[0154] S24. By calculating the ratio of the number of syllables in the text to the total duration of the speech segment, the actual speech rate of the user's current voice command is obtained, so as to quantify the speed at which the user speaks.

[0155] S25. By calculating the ratio of the difference between the user's actual speech rate and the standard speech rate of the instruction template to the standard speech rate, a speech rate imbalance coefficient is obtained to assess the degree of abnormality in the user's pronunciation state.

[0156] S26. Multiply the signal-to-noise ratio factor, confidence factor, and speech rate imbalance coefficient to output the effective instruction duration weight.

[0157] The formula for calculating the instruction duration unit is as follows:

[0158] ;

[0159] in:

[0160] D refers to the valid instruction duration weight, which indicates what proportion of the speech segment detected by VAD can be considered a truly valid instruction signal under the current acoustic conditions and instruction quality. The value range is [0, 1]. The closer D is to 1, the more valid the speech segment is. The closer D is to 0, the more invalid components are mixed in with the speech segment (noise is misjudged as speech, or the instruction itself is too vague and unstable).

[0161] In simple terms, the calculation results of this unit indicate how many of the speech segments detected by the VAD are truly instruction signals that the system should take seriously, after excluding noise interference and pronunciation abnormalities.

[0162] The signal-to-noise ratio (SNR) factor indicates that the speech signal is less contaminated by noise, the start and end boundaries detected by the VAD are reliable, and the effective segment ratio is naturally high. The signal-to-noise ratio is low, which means that speech and noise are difficult to distinguish, and the VAD may misjudge noise segments as speech, thus reducing the effective segment ratio.

[0163] K refers to the preset coefficient 1, which is the system's preset Sigmoid coefficient. In this embodiment, the initial preset value is 0.3, controlling the slope of the curve that controls the influence of SNR on the effective instruction duration weight D.

[0164] SNR stands for Signal-to-Noise Ratio, measured in decibels. It measures the relative intensity of the effective components in a speech signal compared to the background noise components. A higher SNR indicates a stronger relative noise in the speech signal, making the speech clearer and more indistinguishable. A lower SNR indicates a stronger relative noise in the speech signal, making the speech more easily drowned out by noise. The calculation formula is as follows:

[0165] ;

[0166] In the above formula, Espe and Enoi are parameters introduced in the credibility unit, which are the mean square energy of the speech segment and the mean square energy of the background noise, respectively.

[0167] The coefficient 15 in SNR-15 represents a threshold of 15dB, which is the lowest signal-to-noise ratio at which a commercial ASR engine can operate stably.

[0168] The physical meaning of SNR is the degree to which a user's voice is contaminated by background noise in the current acoustic environment;

[0169] This parameter directly determines the intelligibility of the speech signal. The higher the SNR, the stronger the speech energy is than the background noise. The VAD algorithm can accurately detect the start and end boundaries of the speech, and the time domain structure of the speech segment is complete and reliable. The lower the SNR, the closer the noise energy is to or even exceeds the speech energy. VAD may be triggered prematurely (misjudging noise as speech) or terminated late (noise is still judged as speech after the speech ends), resulting in a large number of invalid components in the detected speech segment.

[0170] The confidence factor, or instruction confidence coefficient C, indicates that a high instruction confidence coefficient means the system can trust the VAD's judgment more and accept a more complete speech segment. A low instruction confidence coefficient C means the instruction is suspicious and the system should be more conservative, accepting only a part of the speech even if the VAD detects it.

[0171] NM refers to preset coefficient two, which is the system's preset Sigmoid coefficient. In this embodiment, the initial preset value is 5.

[0172] Cth refers to the confidence threshold, which is preset to 0.5 in this embodiment;

[0173] The speech rate imbalance coefficient indicates that a normal speech rate means the user's pronunciation is natural and fluent, and the instruction's temporal structure is complete. A speech rate that is too fast (phoneme compression) or too slow (dragging) indicates that the user's pronunciation state is abnormal. The speech segment detected by VAD may contain invalid extensions or abnormal pauses.

[0174] Vsc refers to the actual speaking speed of the user's voice command, measured in syllables per second. It measures how fast the user speaks; the higher the value, the faster the speaking, and the lower the value, the slower the speaking. The formula is Vsc = N syllables ÷ T speech segments, where N syllables represents the number of syllables in the text recognized by the ASR engine (each Chinese character corresponds to one syllable, which can be obtained directly from the length of the recognition result string), and T speech segments represents the total duration of the speech segment detected by VAD (in seconds), calculated from the start and end timestamps output by the VAD algorithm.

[0175] Vmb refers to the standard speech rate of this instruction template, also in syllables per second. It is the speech rate of the standard pronunciation pre-recorded by the system, representing the reference speed of the instruction under normal pronunciation conditions. The calculation formula is Vmb = N syllables ÷ T template, where T template represents the standard pronunciation duration preset by the system (in seconds). Each instruction template is pre-recorded and stored. For example, the standard duration for pressing my shoulder is 2 seconds, and the standard duration for applying force is 0.8 seconds.

[0176] The logic of multiplying the three factors in this unit is that the signal must be clean, the command must be reliable, and the pronunciation must be normal. Only when all three conditions are met can the speech segments detected by VAD be considered to be almost entirely valid. If any one of them is problematic, the system should reduce its trust in the integrity of the speech segments.

[0177] In this embodiment: the instruction persistence unit, based on the credibility unit, further evaluates the temporal structural integrity of the instruction. The credibility unit assesses whether the instruction is credible, while the instruction persistence unit assesses how much of the speech segment of the instruction is truly effective. The difference between the two is that an instruction may have a high ASR confidence (e.g., the word "shoulder" is clearly recognized), but if the user's speech rate is abnormally fast, causing phoneme compression, or if background noise causes the speech segment detected by VAD to be longer than actual (including noise tails), then even if the instruction credibility coefficient C is not low, the effective content of this instruction is insufficient. The contribution of the instruction persistence unit is that it adds a layer of temporal quality filtering on top of the instruction credibility coefficient C, enabling the system to not only determine whether the instruction is good or bad, but also how much of the instruction is usable.

[0178] The effective instruction duration weight D controls the system's level of trust in the instruction length. The duration of the speech segment detected by the VAD algorithm is not equal to the effective instruction duration. That is, when the environmental signal-to-noise ratio is low, VAD may misjudge some noise segments as speech, resulting in the detected speech segment being longer than the actual instruction. When the user's speech rate is abnormal, the instruction itself may be stretched or compressed, resulting in a mismatch between the temporal structure of the speech segment and the template. The instruction confidence coefficient C, through a comprehensive evaluation of the instruction confidence coefficient C and speech rate imbalance, gives the proportion of the speech segment detected by VAD that is truly a valid instruction. This proportion directly affects whether the system should execute the instruction completely or with a discount. The existence of the effective instruction duration weight D enables the system to judge the temporal quality of the instruction, avoiding the problem that the system still executes the complete instruction when the text is recognized but the instruction itself is incomplete.

[0179] The effective command duration weight D is a bridging parameter connecting acoustic quality assessment and actuator decision-making. The command credibility coefficient C assesses the "quality" of the command, while the effective command duration weight D assesses the "quantity" of the command. That is, quality and quantity are used together in the multi-target massage parameter analysis unit to jointly determine the degree of actuator response. Without the effective command duration weight D, the system can only make decisions based on whether the command is good or bad (command credibility coefficient C), and cannot consider how many commands are usable. The significance of the effective command duration weight D is that it improves the VAD detection result from binary (with or without speech) to a continuous value (effective proportion), enabling the system to make differentiated execution decisions based on the effective proportion.

[0180] refer to Figure 1 , Figure 2 and Figure 3 The processing flow of the multi-target massage parameter analysis unit is as follows:

[0181] S31. By calculating the exponential decay form of the ratio of the time interval from the last effective voice command to the forgetting half-life, the interaction history forgetting factor is obtained to quantify the degree of memory retention of user habits by the system.

[0182] S32. By introducing the minimum perceptible intensity, maximum safe intensity, and interaction history forgetting factor preset by the system, and combining the linear weighted result of the instruction credibility coefficient and the effective instruction duration weight, the massage intensity coefficient is output.

[0183] S33. By introducing preference weights for each massage area and combining the product of instruction credibility coefficient, effective instruction duration weight, and interaction history forgetting factor as the temperature scaling factor of the normalized exponential function, the region priority probability distribution is output.

[0184] S34. By introducing the product of instruction credibility coefficient, effective instruction duration weight and interaction history forgetting factor, and combining it with the target temperature of the previous decision, the target temperature at the current moment is output.

[0185] The calculation formulas for the multi-target massage parameter analysis unit are as follows:

[0186] ;

[0187] ;

[0188] ;

[0189] ;

[0190] in:

[0191] NA stands for Interaction History Forgetting Factor, which represents the degree of memory retention of user habits by the system. The value range is [0.5, 1]. The closer NA is to 1, the more frequently the user has interacted recently, and the more the system has accumulated the user's preferences, so it can execute instructions more boldly. The closer NA is to 0.5, the longer the user has not interacted, and the system should revert to a conservative strategy and prioritize safety.

[0192] When a user speaks for the first time after powering on the device, the DT (Deep Throat) is high (no history or the last use was several hours ago), and the NA (Near Attack) is close to 0.5. This means the system takes a conservative approach to the first command, and the massage intensity is not immediately maximized. The area distribution tends to be even to prevent accidental operation. As the user issues commands continuously (such as pressing the shoulder → harder → a little more to the left), the DT is continuously reset to a very small value, and the NA gradually approaches 1. The system gradually trusts the user, and the actuator responds more actively and boldly. When the user is silent for a long time, such as 10 minutes, the NA decays back to around 0.5, and the system automatically returns to a conservative state. This mimics the natural behavior of human service personnel: cautious when first taking over, more decisive after becoming familiar with customer needs, and reassessing after a long period of inactivity.

[0193] S refers to the massage intensity coefficient, which is the target output intensity of the kneading motor of the massage chair. Its value range is [Smin, Smax]. It is usually mapped to the motor pulse width modulation (PWM) duty cycle. S directly determines the rotation speed and force of the kneading ball mechanism.

[0194] The massage intensity coefficient S is based on the quality of commands and interaction history, determining how much pressure the massage chair should apply to respond to the user's needs. The quality of the current command is reflected in the degree of trust. The more credible the command and the more complete the voice segment, the more confident the system is in executing the command according to the user's instructions. The interaction history forgetting factor NA, as a weak modulation factor, reflects the long-term accumulation of trust. For users who interact frequently, the system is more willing to maximize the intensity under high-quality commands. For new users or users who have not interacted for a long time, even if the command quality is high, the system will slightly reduce the intensity to avoid discomfort caused by high intensity on the first use.

[0195] Z i The region priority probability distribution refers to the distribution of the massage chair's working time allocation for each massage area according to a probability distribution, where i∈{neck, shoulder, waist}, ∑Z i =1,Z i The closer it is to 1, the more airbag compression time and kneading head dwell time are allocated to that area;

[0196] Priority probability distribution in this region Z iThis indicates the proportion of energy the massage chair should allocate to each body area under the current command quality and trust level. The product of C×D×NA serves as the temperature scaling factor for the normalized exponential function. When C×D×NA is close to 1 (commands are highly reliable, voice is complete, and interaction history is sufficient), the output of the normalized exponential function tends to be an extreme distribution, meaning the area with the highest weight (neck) receives the vast majority of the allocation, while other areas receive very little. This corresponds to the system accurately executing the command when the user explicitly states where to massage. When C×D×NA is close to 0 (commands are ambiguous, voice is incomplete, new user), the output of the normalized exponential function tends to be a uniform distribution, meaning each area is nearly evenly distributed. This corresponds to the system evenly massaging each area when it is unsure where the user wants to massage, avoiding excessive massage in the wrong area that could cause discomfort. This calculation result is the core safety mechanism of the system; when uncertain, it is better to be even than to gamble.

[0197] T n The target temperature at the current moment indicates the target temperature setting of the heat pack, in °C. It is tracked by the controller of the heat pack as the target value. The temperature is divided into three discrete levels: 37°C (low temperature comfort level), 42°C (medium temperature soothing level), and 48°C (high temperature deep heat level).

[0198] The target temperature T at this current moment n Based on the quality of instructions and the accumulation of trust, the massage chair should provide a certain level of heat to match the current massage intensity. Unlike the massage intensity coefficient S, the temperature cannot be continuously and quickly adjusted (heat capacity leads to large inertia). Therefore, discrete levels are used. The higher the C×D×NA, the better the current interaction quality, the clearer the user's understanding, and the more sufficient the trust accumulation. The more willing the system is to provide a higher temperature for heat to enhance the massage effect. The lower the C×D×NA, the more ambiguous the instructions or the insufficient user interaction. The system should maintain a low temperature or room temperature to avoid burns caused by misidentification.

[0199] The temperature of the hot compress is not involved in the continuous calculation of the massage intensity coefficient and the probability distribution of the area priority, but is controlled by an independent state machine. This is for the sake of safety isolation, that is, the temperature control should have its own safety boundary and is not subject to the direct continuous modulation of transient voice signals.

[0200] DT refers to the time interval since the last valid voice command, in seconds. That is, after the system successfully executes a voice command, it records the current system clock timestamp DT1. When the user issues a command again, the difference between the two timestamps is DT.

[0201] Thal refers to the forgetting half-life. This parameter is a preset constant, such as 300 seconds, which means that after 300 seconds, NA decays from 1 to 0.5 + 0.5 × e. −1 ≈0.684;

[0202] Smin refers to the minimum perceptible intensity preset by the system, such as 0.2, which corresponds to a PWM duty cycle of 20% for the motor.

[0203] Smax refers to the maximum safety level preset by the system. For example, 1 corresponds to a PWM duty cycle of 100% for the motor.

[0204] r1 and r2 refer to the weight coefficients of instruction reliability coefficient C and effective instruction duration weight D, respectively. r1 + r2 = 1, such as r1 being 0.5 and r2 being 0.5.

[0205] W i This refers to the preference weight of the i-th region. This parameter is a system-preset constant, such as W. 颈部 =2, W 肩部 =1.5, W 腰部 =1. These weights reflect the default order of importance of the massage chair's areas when it leaves the factory. The neck is usually the most easily fatigued area and the area that users most often request to be massaged, so it has the highest weight.

[0206] M represents the total number of massage areas, which is preset to 3 in this embodiment;

[0207] j refers to the summation index variable used to traverse all massage areas. i is the index of the target area, indicating which area's priority probability is currently being calculated. j is the traversal index in the denominator summation, indicating that all M massage areas are traversed, and the index value is calculated and accumulated for each area. The range of j is exactly the same as i, i.e., j∈{1, 2, ..., M}. When there are three massage areas in the system (neck, shoulder, waist), M=3, and j takes the values ​​1, 2, and 3 respectively, corresponding to the neck, shoulder, and waist.

[0208] T n－1 The target temperature of the previous decision is remembered by the state machine. The C×D×NA combination reflects the quality of the instruction and the trust in the interaction. The three temperature settings of 37℃, 42℃ and 48℃ correspond to normal comfort, gentle soothing and deep heat therapy, respectively.

[0209] No hysteresis is needed when raising the temperature from the low setting of 37°C to the medium setting of 42°C, because 0.25 ≤ C × D × NA ≤ 0.75 covers the intermediate range.

[0210] To increase the temperature from the medium setting to the high setting of 48°C, C×D×NA > 0.75 (the threshold for increasing the temperature).

[0211] To reduce the temperature from high to medium, C×D×NA must be less than 0.25 (the threshold for reduction), rather than returning to 0.75.

[0212] This asymmetric boundary (the rise must be >0.75, and the fall must be <0.25) prevents the temperature from repeatedly jumping between 42℃ and 48℃ when the user's C×D×NA fluctuates between 0.7 and 0.8.

[0213] In this embodiment: the multi-target massage parameter analysis unit is the actuator mapping layer of the entire system. It concretizes the two abstract acoustic evaluation indicators, the instruction credibility coefficient C and the effective instruction duration weight D, as well as the trust accumulation indicator of the time dimension, the interaction history forgetting factor NA, into three physical parameters that can directly drive the actuator: massage intensity coefficient S, regional priority probability distribution Z, and the target temperature T at the current moment.

[0214] The essential role of the massage intensity coefficient S: The massage intensity coefficient S determines the magnitude of the massage force. Its calculation method adopts linear weighting plus weak modulation, which makes the massage intensity coefficient S affected by both the current command quality (command credibility coefficient C and effective command duration weight D) and the interaction history forgetting factor NA. This means that the same force command may have different values ​​for the massage intensity coefficient S when the user first says it after turning on the device and when the user says it after five minutes of continuous interaction. The former has a low interaction history forgetting factor NA, which causes S to be suppressed by weak modulation, while the latter has a high interaction history forgetting factor NA, which makes the massage intensity coefficient S close to the original value of linear weighting. This design enables the system to have an adaptive behavior that is conservative for new users and bold for old users.

[0215] Regional priority probability distribution Z i Its substantial role: Regional priority probability distribution Z i The direction of the massage force is determined by the C×D×NA. When C×D×NA is high, the output tends to be extremely distributed (high-weight areas are concentrated), while when C×D×NA is low, the output tends to be evenly distributed (each area is equally distributed). This allows the system to accurately execute shoulder massage requests when the instructions are clear, and avoids betting on a single area when the instructions are ambiguous. This is because if the massage chair concentrates force on the wrong area, the user's discomfort will be immediate and intense. The area priority probability distribution Z i The design is essentially a safety-oriented probability allocation mechanism that ensures the system chooses the least bad option (uniform allocation) when there is uncertainty, rather than gambling on a potentially wrong option.

[0216] The target temperature T at the current moment n Its essential function: The target temperature Tn at the current moment determines the intensity level of the heat therapy. Discrete levels are used instead of continuous values ​​because the physical characteristics of the heat therapy pad (high heat capacity, high inertia) do not allow for continuous adjustment using instantaneous C×D×NA. The three levels correspond to basic comfort, mild soothing, and deep heat therapy, respectively. The level of C×D×NA determines which level the system selects. Hysteresis prevents temperature jumps near the boundary. The target temperature Tn at the current moment... nThe design decouples the heat therapy control from the massage intensity control, meaning that the temperature will not change frequently with the slight fluctuations in the voice signal, but will remain consistent with the overall command quality level.

[0217] The multi-target massage parameter analysis unit is the final execution layer of the entire system. While the first two formulas are listening and judging, the multi-target massage parameter analysis unit is doing the work. It concretizes the two abstract indicators, the instruction credibility coefficient C and the effective instruction duration weight D, into the motor speed, the valve opening and closing time, and the heating element temperature. These are the things that the user can actually perceive. The three sub-functions of the multi-target massage parameter analysis unit are independent yet interconnected. Together, they define the complete response profile of the system under a given instruction quality: how much force, where to press, and how much heat.

[0218] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A voice interaction control system for a massage chair, characterized in that, It includes a voice interaction data processing module and an execution control management module; The voice interaction data processing module is used to receive preprocessed credibility association data, instruction duration association data, and massage control association data; The voice interaction data processing module: The confidence coefficient of the instruction is output based on the confidence of the automatic speech recognition engine, the mean square energy of the speech segment, the mean square energy of the background noise, the system clock timestamp of the speech start time detected by the VAD algorithm, the time of the most recent exhalation peak detected by the chair back pressure sensor, and a fixed coefficient in the confidence-related data. Based on the signal-to-noise ratio, confidence threshold, actual speech rate of the user's current voice command, and standard speech rate of the command template in the command duration correlation data, and combined with dynamic energy consumption benchmark value, the effective command duration weight is output. Based on the time interval and forgetting half-life from the last effective voice command in the massage control associated data, the interaction history forgetting factor is analyzed. The massage intensity coefficient is calculated by using the system's preset minimum perceptible intensity, the system's preset maximum safe intensity, and the interaction history forgetting factor. The region priority probability distribution and the target temperature at the current moment are output by using the preference weight of the i-th region and the target temperature of the last decision, combined with the command credibility coefficient and the effective command duration weight. The execution control management module: It is used to obtain the massage intensity coefficient, the probability distribution of regional priority and the target temperature at the current moment, and to execute the control management module to control kneading, airbags and heat application.

2. The massage chair voice interaction control system according to claim 1, characterized in that: The voice interaction data processing module includes: Credibility unit, instruction duration unit, and multi-target massage parameter analysis unit; The voice interaction data processing module outputs: The instruction credibility coefficient is used to distinguish the credibility level of instructions and has the ability to distinguish instruction quality. It is the entry quality label of the massage chair voice interaction control system. The effective command duration weight is used to control the system's level of trust in the command length, connecting acoustic quality assessment and execution management decisions; The massage intensity coefficient, the regional priority probability distribution, and the target temperature at the current moment determine the magnitude of the massage force, the direction of the massage force, and the intensity level of the heat application, respectively.

3. The massage chair voice interaction control system according to claim 1, characterized in that: The processing procedure for the credibility unit is as follows: S11. After performing endpoint detection, acoustic feature extraction and decoding on the speech through the automatic speech recognition engine, the confidence score of the automatic speech recognition engine is output to provide a semantic-level trust basis. S12. The mean square energy of the speech segment is obtained by calculating the square mean of the amplitudes of all sampling points within the speech segment, so as to quantify the strength of the user's speech signal. S13. The mean square energy of the background noise is obtained by calculating the square mean of the amplitudes of all sampling points within the silent segment before the instruction, so as to quantify the intensity of the current environmental noise. S14. The energy ratio is obtained by calculating the ratio of the mean square energy of the speech segment to the sum of the mean square energies of the speech segment and the background noise, so as to assess the degree of pollution of the speech signal by the background noise. S15. The chair back pressure sensor detects the periodic changes in chest cavity pressure caused by breathing fluctuations and detects the local minimum point of the pressure waveform to obtain the moment when the chair back pressure sensor detected the most recent expiratory peak. S16. The respiratory phase factor is obtained by dividing the difference between the system clock timestamp at the start of speech and the time when the chair back pressure sensor detected the most recent peak of exhalation by a fixed coefficient, in order to assess the impact of the user's speaking phase in the respiratory cycle on vocal cord stability. S17. Multiply the confidence level of the automatic speech recognition engine, the energy ratio, and the breathing phase factor to output the command confidence coefficient.

4. The massage chair voice interaction control system according to claim 1, characterized in that: The processing procedure for the instruction duration unit is as follows: S21. The signal-to-noise ratio is obtained by calculating the logarithmic form of the ratio of the mean square energy of the speech segment to the mean square energy of the background noise, so as to measure the relative intensity of the effective component and the background noise component in the speech signal. S22. By taking the difference between the signal-to-noise ratio and the preset threshold as input, the signal-to-noise ratio factor is obtained to evaluate the degree of influence of the signal-to-noise ratio on the effective proportion of speech segments. S23. By using the difference between the instruction credibility coefficient and the confidence threshold as input, the confidence factor is obtained to evaluate the degree of influence of instruction credibility on the effective proportion of speech segments. S24. By calculating the ratio of the number of syllables in the text to the total duration of the speech segment, the actual speech rate of the user's current voice command is obtained, so as to quantify the speed at which the user speaks. S25. By calculating the ratio of the difference between the user's actual speech rate and the standard speech rate of the instruction template to the standard speech rate, a speech rate imbalance coefficient is obtained to assess the degree of abnormality in the user's pronunciation state. S26. Multiply the signal-to-noise ratio factor, confidence factor, and speech rate imbalance coefficient to output the effective instruction duration weight.

5. The massage chair voice interaction control system according to claim 1, characterized in that: The processing procedure of the multi-target massage parameter analysis unit is as follows: S31. By calculating the exponential decay form of the ratio of the time interval from the last effective voice command to the forgetting half-life, the interaction history forgetting factor is obtained to quantify the degree of memory retention of user habits by the system. S32. By introducing the minimum perceptible intensity, maximum safe intensity, and interaction history forgetting factor preset by the system, and combining the linear weighted result of the instruction credibility coefficient and the effective instruction duration weight, the massage intensity coefficient is output. S33. By introducing preference weights for each massage area and combining the product of instruction credibility coefficient, effective instruction duration weight, and interaction history forgetting factor as the temperature scaling factor of the normalized exponential function, the region priority probability distribution is output. S34. By introducing the product of the instruction credibility coefficient, the effective instruction duration weight, and the interaction history forgetting factor, and combining it with the target temperature of the previous decision, the target temperature at the current moment is output.

6. The massage chair voice interaction control system according to claim 1, characterized in that: The execution control management module includes a kneading motor control unit, an airbag control unit, and a heat therapy control unit.

7. The massage chair voice interaction control system according to claim 1, characterized in that: The kneading motor control unit is specifically: Controlled objects: one kneading DC brushless motor on each side; Mapping method: The massage intensity coefficient value is linearly mapped to the motor PWM duty cycle; Specific control process: The timer PWM module of the main control chip outputs a square wave with a specific frequency and duty cycle. The square wave is amplified by the motor driver chip and then drives the kneading motor. The kneading motor speed sensor provides feedback on the actual speed, which is then compared with the target speed. The controller then uses a closed-loop adjustment to regulate the PWM duty cycle, bringing the actual speed closer to the target value.

8. The massage chair voice interaction control system according to claim 1, characterized in that: The airbag control unit is specifically: Controlled objects: one neck airbag, one shoulder airbag on each side, and one waist airbag, for a total of four independent air paths, each supplied with air by a solenoid valve and an air pump; Mapping method: The regional priority probability distribution represents the proportion of working time of each region within a control cycle; The specific control process is as follows: First, the control cycle is set, and the working time of each area is allocated based on the probability distribution of regional priority. Within one cycle, the solenoid valves of the three areas are opened in turn in a polling manner, and the opening time is allocated proportionally. During the opening of the solenoid valve, the air pump works at a constant power to inflate the airbag of the corresponding area to the set pressure. The heat therapy control unit is specifically: Controlled objects: Flexible heating pads distributed in the neck, shoulder, and waist areas; Mapping method: The target temperature at the current moment is mapped to the target value of the temperature controller; Specific control process: An NTC thermistor is attached next to each heating element to read the current temperature value; The PID controller uses the current target temperature as the target value and the current temperature value as the feedback value to calculate the PWM control quantity. The PWM signal drives the switching transistor to control the on-time ratio of the heating element; Once the current temperature reaches the target temperature and stabilizes, the PID controller enters the heat preservation state.

Images