Intelligent muscle strength rehabilitation training method based on voice interaction

Abstract

This invention relates to the field of speech analysis technology, specifically disclosing an intelligent muscle strength rehabilitation training method based on voice interaction. The technical solution of this invention determines the target training intensity and mode by extracting acoustic features, recognizing speech, and semantically parsing the patient's voice commands. It then collects and fuses electromyographic and force signals in real time to obtain a muscle strength assessment value. The muscle strength assessment value is compared with a safety threshold to calculate an adaptive adjustment amount. The target training intensity is then superimposed and combined with the training mode to generate training parameters that drive the device. This solves the problems of existing rehabilitation training methods using fixed preset schemes, leading to a mismatch between training intensity and the patient's actual ability, and the inconvenience and easy interruption of training rhythm for patients with limited limb movement due to touchscreen or button interaction. It achieves real-time adaptive adjustment and convenient, natural interaction of the rehabilitation training program, improving training safety and rehabilitation effectiveness.

Description

Technical Field

[0001] This invention relates to the field of speech analysis technology, and in particular to an intelligent muscle strength rehabilitation training method based on speech interaction. Background Technology

[0002] Muscle strength rehabilitation training is a crucial component of motor function reconstruction in patients with sports injuries, postoperative recovery, and neurological diseases. A scientifically sound and reasonable training program plays a key role in restoring muscle strength and joint range of motion. Existing rehabilitation training systems mostly employ preset, fixed programs, requiring therapists to manually set training parameters based on experience. This makes it difficult to dynamically adjust the training based on the patient's real-time muscle strength status, leading to a mismatch between training intensity and the patient's actual ability, thus limiting rehabilitation effectiveness. Furthermore, current intelligent rehabilitation devices primarily rely on touchscreens or buttons for human-computer interaction, which is inconvenient for patients with limited limb mobility, resulting in low interaction efficiency and frequent manual operations that can easily disrupt the training rhythm. Therefore, how to achieve real-time adaptive adjustment of rehabilitation training programs and provide a convenient and natural interaction method for patients with limited limb mobility has become a pressing technical problem to be solved in the field of muscle strength rehabilitation training. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides an intelligent muscle strength rehabilitation training method based on voice interaction. The technical solution of this intelligent muscle strength rehabilitation training method based on voice interaction is as follows: Acoustic feature extraction and speech recognition are performed on the patient's voice commands to obtain the text content corresponding to the voice commands, and semantic parsing is performed on the text content to extract the target training intensity value and target training mode; During the execution of the current training program on the rehabilitation training equipment, the patient's electromyographic signals and force signals are collected in real time, and the electromyographic signals and force signals are fused to calculate the real-time muscle strength assessment value. The real-time muscle strength assessment value is compared with a preset muscle strength safety threshold range. When the real-time muscle strength assessment value exceeds the muscle strength safety threshold range, an adaptive adjustment amount is calculated based on the difference between the real-time muscle strength assessment value and the boundary value of the muscle strength safety threshold range; otherwise, the adaptive adjustment amount is set to zero. The target training intensity value is superimposed with the adaptive adjustment amount, and combined with the target training mode to generate comprehensive training parameters. Based on the comprehensive training parameters, drive commands for controlling the rehabilitation training equipment are generated.

[0004] The technical solution of this invention determines the target training intensity and mode by extracting acoustic features, recognizing speech, and parsing semantics from the patient's voice commands. It collects electromyographic signals and force signals in real time and fuses them to obtain muscle strength assessment values. The muscle strength assessment values ​​are compared with safety thresholds to calculate adaptive adjustment amounts. The target training intensity is superimposed and combined with the training mode to generate training parameters to drive the device. This solves the problems of existing rehabilitation training methods using fixed preset schemes, which lead to a mismatch between training intensity and the patient's actual ability, and the inconvenience and easy interruption of training rhythm for patients with limited limb movement due to touch screen or button interaction. It realizes real-time adaptive adjustment and convenient and natural interaction of rehabilitation training programs, improving training safety and rehabilitation effect.

[0005] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0006] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0007] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart illustrating an embodiment of an intelligent muscle strength rehabilitation training method based on voice interaction according to the present invention. Detailed Implementation

[0008] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0009] Figure 1 The diagram illustrates a flowchart of an embodiment of an intelligent muscle strength rehabilitation training method based on voice interaction provided by the present invention, executed by a control terminal. Figure 1 As shown, this intelligent muscle strength rehabilitation training method based on voice interaction includes the following steps: S1. Perform acoustic feature extraction and speech recognition on the patient's voice commands to obtain the text content corresponding to the voice commands, and perform semantic parsing on the text content to extract the target training intensity value and target training mode.

[0010] Here, "patient" refers to an individual receiving muscle strength rehabilitation training; for example, Patient A is a knee joint surgery rehabilitation patient who interacts with the rehabilitation training equipment via voice during the training process. A voice command refers to a voice signal issued by the patient to express their intention to adjust the training intensity; for example, Patient A saying "Adjust the training intensity to 60" during training constitutes a voice command.

[0011] Acoustic feature extraction refers to the process of extracting parameter vectors that characterize the acoustic properties of speech from the raw signal of a speech command. For example, the speech signal of patient A saying "Adjust the training intensity to 60" is processed by frame segmentation, and acoustic parameters such as the spectral envelope and fundamental frequency are extracted from each frame. Speech recognition refers to the process of converting the acoustic feature sequence of a speech signal into a corresponding text sequence. For example, based on the acoustic features extracted from patient A's speech signal, the speech signal "Adjust the training intensity to 60" is converted into the text sequence "Adjust the training intensity to sixty" through decoding processing using an acoustic model and a language model.

[0012] Here, text content refers to the written expression corresponding to the voice command obtained after speech recognition processing; for example, patient A's voice command, after speech recognition, yields the written expression "adjust the training intensity to sixty," which is the text content. Semantic parsing refers to the process of extracting parameter information with training control significance from the text content; for example, word segmentation and keyword matching are performed on the text content "adjust the training intensity to sixty" to extract parameter information representing the intention to adjust the intensity.

[0013] The target training intensity value refers to the quantitative indicator of the training load that the patient expects to achieve through voice commands. For example, if patient A expects to adjust the training intensity to 60 through voice commands, this value of 60 is the target training intensity value. The target training mode refers to the type of training action or training method that the patient expects to use through voice commands. For example, if patient A expects to use the knee flexion and extension training mode through voice commands, this training mode is the target training mode.

[0014] S2. During the execution of the current training program on the rehabilitation training equipment, the patient's electromyographic signals and force signals are collected in real time, and the electromyographic signals and force signals are fused and processed to calculate the real-time muscle strength assessment value.

[0015] The term "rehabilitation training equipment" refers to mechanical devices used to perform muscle strength rehabilitation training and apply training loads to patients. For example, an isokinetic muscle strength training device with a motor drive function and capable of providing knee flexion-extension training loads is a rehabilitation training device. The "current training program" refers to a complete set of training parameters that the rehabilitation training equipment is executing at the moment it receives a voice command. For example, at the moment patient A issues a voice command, the rehabilitation training equipment is running with a parameter combination of intensity 55, knee flexion-extension mode, and 3 sets of 12 repetitions each; this parameter combination is the current training program.

[0016] Electromyography (EMG) signals refer to the bioelectrical signals generated during muscle contraction. For example, when patient A performs knee flexion and extension exercises, the electrophysiological signals generated by the contraction of the quadriceps femoris muscle are collected in real time via surface electrodes; this electrophysiological signal is the EMG signal. Force signals refer to the mechanical feedback applied by the patient to the rehabilitation training equipment. For example, when patient A performs knee flexion and extension exercises, the force generated by the contraction of the leg muscles is transmitted to the force sensor via the foot pedal. The voltage signal output by the force sensor is calibrated and converted into a force value; this force value data is the force signal.

[0017] The real-time muscle strength assessment value refers to the comprehensive quantitative index reflecting the patient's current muscle strength level obtained by fusing electromyographic signals and force signals collected at the same time. For example, the effective value of the electromyographic signal and the measured value of the force signal collected by patient A at a certain moment during training are substituted into the fusion calculation formula, and the comprehensive value of 0.72 is the real-time muscle strength assessment value at that moment.

[0018] S3. Compare the real-time muscle strength assessment value with a preset muscle strength safety threshold range. When the real-time muscle strength assessment value exceeds the muscle strength safety threshold range, calculate the adaptive adjustment amount based on the difference between the real-time muscle strength assessment value and the boundary value of the muscle strength safety threshold range; otherwise, set the adaptive adjustment amount to zero.

[0019] The muscle strength safety threshold range refers to the safe fluctuation range of muscle strength assessment values ​​pre-set based on the patient's personal historical muscle strength data and the current training stage. For example, based on the muscle strength data from the patient A's first three training sessions and the current fourth week post-surgery training stage, the muscle strength safety threshold range is set to 0.50 to 0.85. Any muscle strength assessment value exceeding this range is considered to pose a training risk. The boundary value refers to the upper or lower limit of the safety threshold range used to define the safe range. For example, the muscle strength safety threshold range for patient A is 0.50 to 0.85, where 0.85 is the upper boundary value and 0.50 is the lower boundary value. When the real-time muscle strength assessment value is 0.90, the boundary value used is 0.85; when the real-time muscle strength assessment value is 0.40, the boundary value used is 0.50.

[0020] The adaptive adjustment amount refers to the adjustment range used to correct the target training intensity value, calculated based on the deviation between the real-time muscle strength assessment value and the boundary value. For example, if patient A's real-time muscle strength assessment value of 0.90 exceeds the upper limit boundary value of 0.85, the adjustment range calculated based on the deviation of 0.05 is +3, and this adjustment range is the adaptive adjustment amount.

[0021] S4. The target training intensity value is superimposed with the adaptive adjustment amount, and combined with the target training mode to generate comprehensive training parameters. Based on the comprehensive training parameters, drive commands for controlling the rehabilitation training equipment are generated.

[0022] The comprehensive training parameters refer to the complete set of control parameters used to drive the rehabilitation training equipment, generated by superimposing the target training intensity value and the adaptive adjustment amount, and then combining them with the target training mode. For example, the target training intensity value of patient A (60) is superimposed with the adaptive adjustment amount +3 to obtain the final training intensity value (63). This is then combined with the knee flexion-extension target training mode to generate a comprehensive set of control parameters including intensity 63, flexion-extension mode, and speed parameters. This set is the comprehensive training parameters. The drive command refers to the control signal generated based on the comprehensive training parameters to control the rehabilitation training equipment to perform corresponding actions. For example, a pulse width modulation (PWM) control signal is generated based on the comprehensive training parameters and sent to the motor driver of the rehabilitation training equipment; this PWM control signal is the drive command.

[0023] The technical solution of this embodiment determines the target training intensity and mode by extracting acoustic features, recognizing speech, and parsing semantics from the patient's voice commands. It collects electromyographic signals and force signals in real time and fuses them to obtain muscle strength assessment values. The muscle strength assessment values ​​are compared with safety thresholds to calculate adaptive adjustment amounts. The target training intensity is superimposed and combined with the training mode to generate training parameters to drive the device. This solves the problems of existing rehabilitation training methods using fixed preset schemes, which lead to a mismatch between training intensity and the patient's actual ability, and the inconvenience and easy interruption of training rhythm for patients with limited limb movement due to touch screen or button interaction. It realizes real-time adaptive adjustment and convenient and natural interaction of rehabilitation training programs, improving training safety and rehabilitation effect.

[0024] In the aforementioned technical solution, voice interaction and adaptive adjustment are not independent additional functions, but rather synergistically enhanced through an intent correction and superposition mechanism. The voice channel allows patients to conveniently express their training adjustment needs based on their subjective feelings, compensating for the lack of patient subjective participation in purely physiological signal feedback. The physiological feedback channel, through muscle strength assessment and threshold comparison, corrects for potentially excessively high or low demands in the patient's voice intent in real time, preventing unsafe training intensity due to perceptual bias or speech recognition errors. The two channels work together to ensure that the training program simultaneously considers the patient's subjective will and objective physiological safety at the execution level, achieving precise rehabilitation training control under human-machine collaborative decision-making.

[0025] In one alternative approach, when extracting acoustic features from the voice command, the Mel frequency cepstral coefficients of the voice command are extracted as acoustic features, and speech recognition is performed using a hidden Markov model based on the Mel frequency cepstral coefficients to obtain the text content.

[0026] Mel-frequency cepstral coefficients refer to the feature parameters extracted on the Mel-frequency scale to characterize the spectral envelope of a speech signal. For example, performing a Fast Fourier Transform on each frame of the speech signal of patient A saying "Adjust the training intensity to 60," mapping the transformed spectral energy onto the Mel-frequency scale, and then performing a Discrete Cosine Transform yields a set of 12-dimensional feature coefficients, which are the Mel-frequency cepstral coefficients. Acoustic features refer to the set of parameters extracted from a speech signal to characterize its acoustic properties. For example, the 12-dimensional Mel-frequency cepstral coefficients extracted from patient A's speech signal, along with the fundamental frequency and short-time energy parameters, constitute the acoustic features of this speech signal.

[0027] Hidden Markov Model (HMM) refers to a probabilistic model used for statistical modeling and state decoding of the temporal features of speech signals. For example, the Mel frequency cepstral coefficient sequence of patient A's speech signal is used as the observation sequence and input into a pre-trained HMM. This model decodes the most likely state sequence using the Viterbi algorithm and maps this state sequence to the text output "adjust the training intensity to sixty".

[0028] Among the above-mentioned optional methods, further extraction of Mel frequency cepstral coefficients and combination with a hidden Markov model for speech recognition can accurately capture frequency characteristics from speech signals and convert them into text content, providing a reliable input basis for subsequent semantic parsing and ensuring the recognition accuracy of speech interaction.

[0029] In one alternative approach, when performing semantic parsing on the text content, the text content is matched with a preset training intensity keyword library and a training mode keyword library to obtain the target training intensity value and the target training mode from the text content.

[0030] The training intensity keyword database refers to a vocabulary set that stores words related to training intensity and their corresponding numerical mappings. For example, the training intensity keyword database includes "intensity" corresponding to training intensity categories, "adjust to" corresponding to adjustment actions, and the numerical value "60" corresponding to the target intensity value. By matching the words in the text content "adjust the training intensity to sixty" with this vocabulary, the target training intensity value of 60 is extracted. The training mode keyword database refers to a vocabulary set that stores words related to training modes and their corresponding mode identifiers. For example, the training mode keyword database includes "flexion and extension" corresponding to knee flexion and extension modes, "abduction" corresponding to hip abduction modes, and "adduction" corresponding to hip adduction modes. By matching the words in the text content with this vocabulary, the knee flexion and extension training mode corresponding to "flexion and extension" is identified, and this mode identifier is the target training mode.

[0031] In the above-mentioned optional methods, by further matching the text content with a preset keyword library, the training intensity and pattern information in the voice command can be quickly identified, simplifying the semantic parsing process, making the conversion of voice commands into training parameters more direct and clear, and facilitating the rapid generation of training schemes.

[0032] In one alternative approach, when acquiring the patient's electromyographic signals in real time, the electromyographic signals are acquired through surface electrodes attached to the skin surface of the patient's target muscle group; when acquiring the force signals in real time, the force signals are acquired through a force sensor disposed on the rehabilitation training device.

[0033] The target muscle group skin surface refers to the body surface area corresponding to the main force-generating muscle group targeted by the target training program. For example, when patient A performs knee flexion and extension training, the main force-generating muscle group is the quadriceps femoris, and the body surface area on the front of the thigh corresponding to the quadriceps femoris is the target muscle group skin surface. Surface electrodes refer to sensor devices attached to the skin surface to pick up bioelectrical signals generated by muscle activity. For example, two Ag / AgCl surface electrodes are attached to the skin surface at the location of the quadriceps femoris muscle belly on the front of patient A's thigh. These surface electrodes pick up the electromyographic signals generated when the quadriceps femoris contracts in real time and transmit them to the signal acquisition circuit.

[0034] Among them, a force sensor refers to a transducer device installed on a rehabilitation training device to measure the force applied by the patient. For example, a strain gauge force sensor is installed under the foot pedal of the rehabilitation training device used by patient A. When patient A performs knee flexion and extension training, the force applied by the leg to the foot pedal causes the strain gauge inside the sensor to deform. The deformation is converted into a voltage signal output. This sensor is a force sensor.

[0035] In the above-mentioned optional methods, electromyographic signals and force signals can be collected by surface electrodes and force sensors respectively. Real-time data can be obtained from two dimensions: the patient's muscle groups and rehabilitation training equipment. This provides multi-source data support for muscle strength assessment and makes muscle strength monitoring more comprehensive and accurate.

[0036] In one alternative approach, the preprocessing step prior to the fusion of the electromyographic signal and the force signal includes: The electromyographic signal is subjected to bandpass filtering and amplification, and the force signal is subjected to analog-to-digital conversion.

[0037] In the above-mentioned optional methods, further bandpass filtering and amplification of the electromyographic signal and analog-to-digital conversion of the force signal can remove signal noise and enhance signal quality, making the original data before fusion processing more stable and reliable, laying the foundation for subsequent muscle strength assessment value calculation.

[0038] In one alternative approach, the expression for calculating the real-time muscle strength assessment value is: ; Where E represents the real-time muscle strength assessment value, S EMG S represents the effective value of the electromyographic signal. EMG_max S represents the preset maximum value of the electromyographic signal. Force S represents the current measured value of the force signal. Force_maxThe preset maximum value of the force signal is represented by α, the fusion weighting coefficient of the electromyographic signal is represented by p, the normalization exponent of the electromyographic signal is represented by q, the normalization exponent of the force signal is represented by β, and the interaction coupling coefficient between the electromyographic signal and the force signal is represented by β.

[0039] It should be noted that the above expression is constructed based on the following principle: muscle strength assessment needs to consider two dimensions simultaneously: the level of neural activation of the muscle and the actual mechanical output. A weighted nonlinear normalization framework is used to normalize the electromyographic (EMG) and force signals separately using a power-law method before weighted summation. The power-law parameter can independently adjust the sensitivity difference between the two signals in the low-value and high-value regions. Simultaneously, a multiplicative coupling factor reflecting the interaction effect of the two signals is multiplied on the weighted sum. This factor generates additional synergistic gain when the EMG and force signals are synchronously enhanced, capturing the nonlinear synergistic relationship between muscle activation and external force output. By introducing an interaction coupling coefficient, the above expression extends single-signal assessment to dual-signal synergistic assessment, enabling the calculated real-time muscle strength assessment value to more comprehensively reflect the patient's actual muscle exertion state, thus providing a more accurate input basis for subsequent safety threshold comparison and adaptive adjustment.

[0040] Among the above-mentioned optional methods, by further introducing a weighted fusion formula of electromyographic signals and force signals and setting an interactive coupling coefficient, it is possible to comprehensively reflect the degree of muscle activation and the level of external force output, so that the real-time muscle strength assessment value is closer to the patient's actual muscle strength state, providing an accurate basis for adaptive adjustment.

[0041] In one alternative approach, before comparing the real-time muscle strength assessment value with the muscle strength safety threshold range, the muscle strength safety threshold range is preset based on the patient's personal historical muscle strength data and the training phase in the current training program. The muscle strength safety threshold range includes an upper limit boundary value and a lower limit boundary value.

[0042] Personal historical muscle strength data refers to the sequence of muscle strength assessment values ​​recorded during multiple training sessions in the current rehabilitation cycle. For example, patient A has completed 5 rehabilitation training sessions before this training, with average muscle strength assessment values ​​recorded in each session being 0.62, 0.65, 0.68, 0.70, and 0.73, respectively. These 5 muscle strength assessment values ​​constitute personal historical muscle strength data. Training phase refers to rehabilitation periods with different training goals, divided according to the patient's postoperative recovery time. For example, patient A is in the recovery period from week 4 to week 6 after knee surgery; this period is classified as the muscle strength recovery training phase, and this phase is the training phase.

[0043] The upper limit boundary value refers to the larger endpoint value within the muscle strength safety threshold range; for example, patient A's muscle strength safety threshold range is 0.50 to 0.85, where 0.85 is the upper limit boundary value. The lower limit boundary value refers to the smaller endpoint value within the muscle strength safety threshold range; for example, patient A's muscle strength safety threshold range is 0.50 to 0.85, where 0.50 is the lower limit boundary value.

[0044] Among the above-mentioned optional methods, by further combining personal historical muscle strength data and preset safety threshold ranges for training stages, safety boundaries can be dynamically set according to individual patient differences and rehabilitation progress, making the warning range of muscle strength monitoring more in line with actual rehabilitation needs and avoiding the application deviation caused by uniform thresholds.

[0045] In one alternative approach, when the real-time muscle strength assessment value exceeds the muscle strength safety threshold range, if the real-time muscle strength assessment value is greater than the upper limit boundary value, then the upper limit boundary value is used as the boundary value; if the real-time muscle strength assessment value is less than the lower limit boundary value, then the lower limit boundary value is used as the boundary value, and the difference between the real-time muscle strength assessment value and the boundary value is calculated.

[0046] In the above-mentioned optional methods, by further distinguishing between cases exceeding the upper or lower limit boundary values ​​and calculating the difference separately, it is possible to clarify the specific direction and degree of the muscle strength assessment value deviating from the safe range, making the calculation basis of the adaptive adjustment amount clearer and facilitating subsequent targeted intensity adjustments.

[0047] In one alternative approach, the expression for calculating the adaptive adjustment amount is: ; Wherein, ΔI represents the adaptive adjustment amount, ΔE represents the difference between the real-time muscle strength assessment value and the boundary value, and E range I represents the difference between the upper and lower boundary values ​​of the muscle strength safety threshold range. scale This represents the base intensity adjustment coefficient in the current training scheme. and All are nonlinear adjustment exponents, where η represents the nonlinear adjustment amplitude coefficient, and sgn(ΔE) represents the sign of ΔE.

[0048] It should be noted that the above expression is constructed based on the following principle: the adaptive adjustment amount needs to be calculated differently according to the direction and degree of muscle strength deviation from the safe range. The ratio of the deviation value to the range width is used as the normalized deviation variable. A sign function is used to control the positive and negative directions of the adjustment amount to be consistent with the deviation direction. The nonlinear exponent of the normalized deviation variable is used as the basic scaling factor for the adjustment amplitude. At the same time, an additional nonlinear increment term is superimposed on the basic intensity adjustment coefficient. This increment term has a weak effect when the deviation is small, but plays a dominant role when the deviation is large, thus forming a two-layer nonlinear architecture that combines basic adjustment and enhancement adjustment. The above expression generates a gradual adjustment amount when the deviation is small to avoid over-correction and disrupting training continuity, and generates an accelerated adjustment amount when the deviation is large to quickly pull the training intensity back to the safe range. This achieves a differentiated control strategy of fine adjustment for small deviations and rapid adjustment for large deviations, thereby accurately matching the adjustment amplitude of training intensity with the degree of muscle strength abnormality.

[0049] In the above-mentioned optional methods, an adaptive adjustment calculation model is further constructed by introducing a nonlinear adjustment index and an amplitude coefficient. This model can perform differentiated intensity adjustment according to the degree of muscle strength deviation, so that the adjustment range of training parameters matches the degree of muscle strength abnormality, avoiding over- or under-adjustment.

[0050] In one alternative approach, when the target training intensity value is superimposed with the adaptive adjustment amount, the algebraic sum of the target training intensity value and the adaptive adjustment amount is calculated, and the algebraic sum is used as the final training intensity value. The final training intensity value is combined with the target training mode to generate the comprehensive training parameters. A pulse width modulation drive command is generated based on the comprehensive training parameters. The pulse width modulation drive command is used to control the motor output torque of the rehabilitation training device.

[0051] The final training intensity value refers to the quantitative indicator of training load used for actual execution, obtained by superimposing the target training intensity value and the adaptive adjustment amount. For example, if patient A's target training intensity value is 60 and the adaptive adjustment amount is +3, the final training intensity value obtained by superimposing the two is 63. This value of 63 is the actual training intensity executed by the rehabilitation training equipment. The pulse width modulation drive command refers to the control signal that controls the output torque of the motor by adjusting the duty cycle of the pulse signal. For example, a pulse width modulation signal with a frequency of 20 kHz is generated based on the comprehensive training parameters, and the duty cycle of this signal is set to 65%. This pulse width modulation signal is the pulse width modulation drive command.

[0052] Among them, the motor output torque refers to the rotational torque output by the drive motor of the rehabilitation training equipment under the control of the pulse width modulation drive command; for example, after the motor of the rehabilitation training equipment receives the pulse width modulation drive command with a duty cycle of 65%, it outputs a rotational torque of 15 N·m. This rotational torque acts on the knee joint of patient A to provide training load, and 15 N·m is the motor output torque.

[0053] In the above-mentioned optional methods, by further superimposing the target training intensity and the adaptive adjustment amount to generate comprehensive training parameters, and converting them into pulse width modulation drive commands, the output torque of the motor of the rehabilitation training equipment can be directly controlled, so that the adjustment results of the training plan can be quickly implemented at the equipment execution level.

[0054] In one optional approach, after obtaining the target training intensity value, the current training intensity value in the current training scheme is obtained, and the difference between the target training intensity value and the current training intensity value is calculated. When the difference is greater than the preset maximum allowable adjustment step size, the target training intensity value is corrected to the sum of the current training intensity value and the maximum allowable adjustment step size.

[0055] The current training intensity value refers to the quantitative indicator of the training load actually used by the rehabilitation training equipment when executing the current training program. For example, at the moment patient A issues a voice command, the rehabilitation training equipment is running with an intensity of 55, and this value of 55 is the current training intensity value. The maximum allowable adjustment step size refers to the maximum adjustment range of the target training intensity value relative to the current training intensity value allowed by a single voice command. For example, if the maximum allowable adjustment step size is set to 5, when patient A adjusts the training intensity from 55 to 60 via voice command, the difference of 5 does not exceed the maximum allowable adjustment step size of 5, and the corrected target training intensity value remains at 60. When patient A adjusts the training intensity from 55 to 70, the difference of 15 exceeds the maximum allowable adjustment step size of 5, and the corrected target training intensity value is limited to 60.

[0056] In the above-mentioned optional methods, by further comparing the target training intensity value obtained from the parsing of the voice command with the current training intensity value in the current training scheme, and by limiting and correcting the voice intent that exceeds the maximum allowable adjustment step size, it is possible to prevent sudden changes in training intensity caused by voice recognition errors or patient misoperation. This provides a safe buffer mechanism for the training intensity adjustment process under the voice interaction method, and avoids patients from bearing inappropriate training loads due to deviations in a single voice command.

[0057] In one alternative approach, the calculation expression for correcting the target training intensity value to the sum of the current training intensity value and the maximum allowed adjustment step size is as follows: ; in, I represents the corrected target training intensity value. current I represents the current training intensity value in the current training scheme. target D represents the target training intensity value. max This indicates the maximum allowed adjustment step size, clip(x,-D) max D max ) indicates that the variable x is restricted to -D max To D max Within the closed interval.

[0058] It should be noted that the above expression is constructed based on the following principle: the speech recognition process may produce recognition deviations due to environmental noise, unclear patient pronunciation, or model decoding errors. Patients may also issue intensity adjustment commands that exceed the safe range due to inaccurate judgment of their own tolerance. Therefore, after the speech command is parsed and before the target training intensity value enters the adaptive adjustment link, an amplitude limiting constraint is added. By comparing the difference between the target training intensity value corresponding to the patient's speech intent and the training intensity value currently executed by the rehabilitation training device, and using the amplitude limiting function to forcibly constrain the difference within the preset maximum allowable adjustment step size range, the intensity change range that a single speech command can actually cause is strictly limited within the safe boundary.

[0059] The above expression, by embedding a safety limiting mechanism in the voice interaction link, filters out abnormal intensity adjustment intentions caused by voice recognition errors or patient misoperation before entering subsequent adaptive adjustment. Without weakening the convenience of voice interaction, it effectively prevents the safety risks to patients caused by sudden changes in training intensity, and provides a safety protection barrier for rehabilitation training under voice interaction mode that is independent of physiological feedback adjustment.

[0060] In the above-mentioned optional methods, by further adding an amplitude limiting constraint step between the voice command parsing stage and the intensity superposition stage, the patient's voice intention is first checked for safety boundaries before entering the adaptive adjustment link. This can add a layer of security protection without weakening the convenience of voice interaction and ensure the safety of the training intensity adjustment process.

[0061] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A voice-interactive intelligent muscle strength rehabilitation training method, characterized in that, include: Acoustic feature extraction and speech recognition are performed on the patient's voice commands to obtain the text content corresponding to the voice commands, and semantic parsing is performed on the text content to extract the target training intensity value and target training mode; During the execution of the current training program on the rehabilitation training equipment, the patient's electromyographic signals and force signals are collected in real time, and the electromyographic signals and force signals are fused to calculate the real-time muscle strength assessment value. The real-time muscle strength assessment value is compared with a preset muscle strength safety threshold range. When the real-time muscle strength assessment value exceeds the muscle strength safety threshold range, an adaptive adjustment amount is calculated based on the difference between the real-time muscle strength assessment value and the boundary value of the muscle strength safety threshold range; otherwise, the adaptive adjustment amount is set to zero. The target training intensity value is superimposed with the adaptive adjustment amount, and combined with the target training mode to generate comprehensive training parameters. Based on the comprehensive training parameters, drive commands for controlling the rehabilitation training equipment are generated.

2. The intelligent muscle strength rehabilitation training method based on voice interaction according to claim 1, characterized in that, When extracting acoustic features from the voice command, the Mel frequency cepstral coefficients of the voice command are extracted as acoustic features, and speech recognition is performed using a hidden Markov model based on the Mel frequency cepstral coefficients to obtain the text content.

3. The intelligent muscle strength rehabilitation training method based on voice interaction according to claim 2, characterized in that, When performing semantic parsing on the text content, the text content is matched with a preset training intensity keyword library and a training mode keyword library to obtain the target training intensity value and the target training mode from the text content.

4. The intelligent muscle strength rehabilitation training method based on voice interaction according to claim 1, characterized in that, When collecting the patient's electromyographic signals in real time, the electromyographic signals are collected by surface electrodes attached to the skin surface of the target muscle group of the patient. When collecting the force signals in real time, the force signals are collected by force sensors set on the rehabilitation training equipment.

5. The intelligent muscle strength rehabilitation training method based on voice interaction according to claim 4, characterized in that, The preprocessing steps before fusing the electromyographic signals and the force signals include: The electromyographic signal is subjected to bandpass filtering and amplification, and the force signal is subjected to analog-to-digital conversion.

6. The intelligent muscle strength rehabilitation training method based on voice interaction according to claim 1, characterized in that, The expression for calculating the real-time muscle strength assessment value is as follows: ； Where E represents the real-time muscle strength assessment value, S EMG S represents the effective value of the electromyographic signal. EMG_max S represents the preset maximum value of the electromyographic signal. Force S represents the current measured value of the force signal. Force_max The preset maximum value of the force signal is represented by α, the fusion weighting coefficient of the electromyographic signal is represented by p, the normalization exponent of the electromyographic signal is represented by q, the normalization exponent of the force signal is represented by β, and the interaction coupling coefficient between the electromyographic signal and the force signal is represented by β.

7. The intelligent muscle strength rehabilitation training method based on voice interaction according to claim 1, characterized in that, Before comparing the real-time muscle strength assessment value with the muscle strength safety threshold range, the muscle strength safety threshold range is preset based on the patient's personal historical muscle strength data and the training stage in the current training program. The muscle strength safety threshold range includes an upper limit boundary value and a lower limit boundary value.

8. The intelligent muscle strength rehabilitation training method based on voice interaction according to claim 7, characterized in that, When the real-time muscle strength assessment value exceeds the muscle strength safety threshold range, if the real-time muscle strength assessment value is greater than the upper limit boundary value, then the upper limit boundary value is used as the boundary value; if the real-time muscle strength assessment value is less than the lower limit boundary value, then the lower limit boundary value is used as the boundary value, and the difference between the real-time muscle strength assessment value and the boundary value is calculated.

9. The intelligent muscle strength rehabilitation training method based on voice interaction according to claim 8, characterized in that, The expression for calculating the adaptive adjustment amount is: ; Wherein, ΔI represents the adaptive adjustment amount, ΔE represents the difference between the real-time muscle strength assessment value and the boundary value, and E range I represents the difference between the upper and lower boundary values ​​of the muscle strength safety threshold range. scale This represents the base intensity adjustment coefficient in the current training scheme. and All are nonlinear adjustment exponents, where η represents the nonlinear adjustment amplitude coefficient, and sgn(ΔE) represents the sign of ΔE.

10. The intelligent muscle strength rehabilitation training method based on voice interaction according to claim 1, characterized in that, When the target training intensity value is superimposed with the adaptive adjustment amount, the algebraic sum of the target training intensity value and the adaptive adjustment amount is calculated, and the algebraic sum is used as the final training intensity value. The final training intensity value is combined with the target training mode to generate the comprehensive training parameters. A pulse width modulation drive command is generated according to the comprehensive training parameters. The pulse width modulation drive command is used to control the motor output torque of the rehabilitation training device.

Images