Traditional Chinese medicine physique intelligent identification system based on natural language processing and voice recognition technology

Abstract

The invention relates to a traditional Chinese medicine physique intelligent identification system based on natural language processing and voice recognition technology, which is characterized in that a voice acquisition module, a voice recognition module, a natural language processing module, a physique identification module and the like are integrated, key features are extracted from patient voice information and questionnaire answers, and physique categories are judged in combination with a traditional Chinese medicine physique theoretical model. The system comprises a user interaction interface for a patient to fill in questionnaires and check results, a database stores related data, a machine learning algorithm is built in, and the identification precision and efficiency can be continuously optimized by analyzing historical data. The system provides a new tool for personalized prevention and management of rheumatism immune diseases, and also promotes the modern development of the traditional Chinese medicine constitution theory.

Description

Technical Field

[0001] This invention belongs to the field of medical and health information technology, specifically relating to a TCM constitution intelligent identification system that combines natural language processing and speech recognition technology. Background Technology

[0002] In traditional Chinese medicine, constitution refers to the sum of inherent physiological and psychological characteristics of the human body, which determine an individual's adaptability to the external environment and susceptibility to disease. Traditional Chinese medicine constitution theory holds that different people have different constitution types, which influence their susceptibility to various diseases and their responsiveness to treatment. Therefore, accurately identifying a patient's constitution is crucial for disease prevention, diagnosis, and treatment.

[0003] However, traditional methods of constitution identification mainly rely on doctors' clinical experience and patients' subjective responses. This method has some significant drawbacks: First, it is highly dependent on the doctor's professional level and experience, and the accuracy of identification may be affected by individual differences in judgment; second, patients' subjective responses may be affected by factors such as emotions and cognitive biases, leading to inaccurate results; finally, this method is inefficient and difficult to meet the health assessment needs of large-scale populations.

[0004] With the development of artificial intelligence and information technology, natural language processing and speech recognition technologies have been widely applied in the healthcare field, providing an efficient and objective means of automatic identification. By analyzing a patient's speech and language, objective information related to their health status can be extracted, thus overcoming the limitations of traditional methods.

[0005] Therefore, developing a TCM constitution intelligent identification system that combines natural language processing and speech recognition technologies to more accurately and quickly identify patients' constitutions and provide strong technical support for personalized medical services has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] This invention aims to provide a Traditional Chinese Medicine (TCM) constitution intelligent identification system based on natural language processing and speech recognition technologies, addressing the problems of traditional TCM constitution identification relying on physician experience, high subjectivity, and low efficiency. Traditional methods suffer from insufficient accuracy due to differences in individual physician judgment and the influence of patient subjective responses on emotions, making them unsuitable for large-scale population health assessments. This system integrates modules for sound acquisition, speech recognition, natural language processing, constitution identification, user interface, database, and machine learning algorithms to construct a complete intelligent identification process. It extracts key features from patient voice information and questionnaire answers, and automatically determines constitution type by combining TCM constitution theory models and machine learning algorithms. The system possesses self-learning and optimization capabilities, improving identification accuracy and efficiency with data accumulation, providing a scientific tool for personalized prevention and management of rheumatic and immune diseases, and promoting the modernization of TCM constitution theory.

[0007] Technical solution

[0008] The sound acquisition module uses a high-sensitivity microphone as a sensor, with a sampling frequency set at 16kHz and a bit depth of 16 bits. This ensures accurate capture of patient speech information while avoiding data redundancy. The acquired raw speech signal undergoes preprocessing: environmental noise is removed using adaptive filters or noise cancellation techniques to ensure clear speech; standardization processing adjusts the speech signal amplitude to a uniform level, eliminating the impact of differences in pronunciation intensity among different patients on subsequent analysis. The preprocessed speech signal is transmitted to the speech recognition module in real time, laying the foundation for text conversion.

[0009] The core of the speech recognition module is to convert preprocessed speech signals into computer-processable text data. First, feature extraction is performed on the speech signal, decomposing it into time segments (frames). Feature vectors such as Mel-frequency cepstral coefficients (MFCCs) are extracted from each frame to reflect the speech spectrum characteristics. Then, these feature vectors are input into a pre-trained deep learning model (such as a Long Short-Term Memory network (LSTM) or a Convolutional Neural Network (CNN)) to generate a text sequence corresponding to the speech segment. Finally, post-processing using a language model smooths and corrects the text sequence, resolving issues such as homophones and incoherent sentences, achieving an accuracy rate of over 95% before transmitting the data to the natural language processing module.

[0010] After receiving the text data converted by the speech recognition module, the natural language processing module first performs word segmentation to break the text into individual word units, clarifying the sentence structure and semantic boundaries. Then, through syntactic and semantic analysis, it identifies key information such as symptoms and feelings described by the patient (e.g., "chills" and "night sweats"). At the same time, it extracts feature parameters related to constitution through specialized algorithms, such as tone changes, speech rate (unit: words / second), pause frequency and duration, etc., and combines them with sentiment analysis tools to assess the patient's emotional state. Finally, it summarizes the feature parameters to form a feature dataset corresponding to the TCM constitution theory model, and transmits it to the constitution identification module.

[0011] The constitution identification module incorporates multiple TCM constitution theory models, covering common types such as balanced constitution, qi deficiency constitution, yang deficiency constitution, yin deficiency constitution, phlegm-dampness constitution, damp-heat constitution, blood stasis constitution, qi stagnation constitution, and special constitution. Users can select the appropriate model version for different age groups, genders, and geographical locations through the user interface. After receiving the feature dataset from the natural language processing module, it utilizes machine learning algorithms such as Support Vector Machine (SVM) or Random Forest, combined with information such as patient age, gender, and medical history, to automatically determine and output the patient's constitution type by matching constitution patterns and eliminating low-probability options, achieving an accuracy rate of 92%.

[0012] The user interface is designed to be intuitive and user-friendly, with a response time of less than 1 second, and supports touch and voice control. It guides patients to fill out basic information, medical history, and a questionnaire related to their constitution, including necessary prompts and help information. Patients can issue voice commands for data collection, and the progress of constitution identification is displayed in real time (e.g., "Voice conversion in progress," "Feature extraction in progress"). After identification, the results are displayed in graphical (e.g., a chart showing the percentage of different constitution types) or text format, and health advice combining traditional Chinese medicine theory and modern medicine is provided (e.g., dietary adjustments, exercise guidance). A historical record query function is also provided to facilitate patients tracking changes in their constitution. The font size and operation steps are optimized for elderly patients.

[0013] The database employs an SQL database management system to encrypt and store patient questionnaire content, voice samples, physical constitution identification results, and other data. It is categorized and managed according to a preset pattern to ensure structured data and efficient querying. Strict access control and data encryption technologies are implemented to protect privacy, allowing access only to authorized personnel. New data is verified and integrated daily at midnight, redundant data is cleaned monthly, and regular backups, index rebuilds, and performance optimizations are performed, resulting in a 30% improvement in query efficiency. It supports integration with the Hospital Information System (HIS) to synchronize patient medical history and enrich the identification reference dimensions.

[0014] The machine learning algorithm module extracts historical data (including voice samples, questionnaire answers, and constitution identification results) from the database, uses cross-validation technology to train the initial machine learning model, and adjusts parameters to optimize recognition capabilities. It evaluates performance by comparing the model's prediction results with actual constitution types (such as the results of professional TCM doctors' judgments) and corrects parameters for deviations. It regularly introduces new data features (such as adding new constitution-related vocabulary and voice feature parameters) and advanced algorithms such as deep learning, and completes model updates and redeployment every 3 months to continuously improve the accuracy and efficiency of constitution identification.

[0015] The overall process is as follows: After the patient inputs basic information through the user interface, the sound acquisition module collects speech samples at a sampling frequency of 16kHz and a depth of 16 bits, simultaneously performing noise reduction and standardization preprocessing; the speech recognition module extracts MFCCs feature vectors from the preprocessed speech samples, converts them into text sequences using an LSTM or CNN model, and then generates accurate text data through language model post-processing; the natural language processing module performs word segmentation, syntax and semantic analysis on the text data, extracting constitution feature parameters such as intonation, speech rate, and pauses, as well as symptom description keywords, forming a feature dataset; the constitution identification module calls the appropriate TCM constitution theory model, combines SVM or random forest algorithms to analyze the feature dataset, and comprehensively determines the constitution type based on the patient information; the user interface displays the constitution type results in real time, along with a health report containing suggestions on diet and exercise; the system automatically encrypts and stores the speech samples, text data, feature parameters, constitution results, and health report in the database, supporting subsequent queries and machine learning training; the machine learning algorithm module continuously trains the model based on historical data in the database, achieving periodic optimization and improving system performance.

[0016] The above technical solution can bring about the following technical effects: 1. Significantly improves the accuracy and objectivity of TCM constitution identification: Through the high-sensitivity sound pickup and preprocessing technology of the sound acquisition module (16kHz sampling frequency, 16-bit depth combined with noise reduction and standardization), combined with the text conversion accuracy of the speech recognition module (over 95%), and the precise extraction of features such as tone, speech rate, and symptom keywords by the natural language processing module (accuracy over 90%), and then through the multi-model and machine learning algorithms built into the constitution identification module (judgment accuracy of 92%), it effectively avoids the problems of differences in doctor experience and subjective answers from patients in traditional methods, so that the consistency between the identification results and the judgment of professional TCM doctors reaches over 90%, greatly improving the scientificity and reliability of constitution identification.

[0017] 2. Achieving a highly efficient and automated constitution identification process to improve healthcare service efficiency: The system automates the entire process, from patient input and voice acquisition to final output, through the collaborative operation of all modules. Voice acquisition and preprocessing are seamlessly integrated with speech recognition; after text conversion, it immediately enters the natural language processing stage; and after feature extraction, the constitution identification module quickly makes a determination. The entire process is reflected in real-time through a user interface with a response time of less than one second. Furthermore, the structured storage and efficient querying of the database (30% efficiency improvement) support rapid retrieval of historical data. Compared to traditional methods relying on manual inquiry and analysis, this system significantly shortens the time for individual constitution identification, meets the health assessment needs of large-scale populations, and effectively improves the efficiency and coverage of healthcare services.

[0018] 3. Possesses continuous optimization capabilities, adapts to diverse needs, and promotes the modernization of Traditional Chinese Medicine (TCM): The machine learning algorithm module is trained and optimized every three months using cross-validation technology on historical data (voice samples, questionnaire answers, identification results, etc.). Combined with new data features and advanced algorithm iteration models, the system can continuously adapt to the differences in the physical characteristics of different individuals and groups. At the same time, the system transforms TCM constitution theory into a quantifiable and computable model, realizing the digital application of traditional theory through modern information technology. This not only provides data support for personalized prevention (such as targeted health advice) and management of rheumatic and immune diseases, but also promotes the integration of TCM constitution theory with modern technology, providing a practical and feasible technical path for the modernization of TCM. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the structure of a TCM constitution intelligent identification system based on natural language processing and speech recognition technology. Detailed Implementation

[0020] 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.

[0021] The TCM constitution intelligent identification system based on natural language processing and speech recognition technology described in this invention integrates modules such as sound acquisition, speech recognition, natural language processing, constitution identification, user interface, database, and machine learning algorithms to construct a complete automated TCM constitution identification system. The specific implementation methods are described in detail below from aspects such as module configuration, operation process, and technical advantages, ensuring a high degree of conformity with the claims and invention content.

[0022] I. Detailed Configuration and Technical Parameters of Each System Module 1. Sound Acquisition Module This module uses a high-sensitivity electret microphone as its sound sensor, with a pickup range covering 30Hz-16kHz. It can accurately capture subtle features in the patient's voice, including changes in intonation and breathing rhythm. The hardware parameters are set to a sampling frequency of 16kHz and a bit depth of 16 bits, which can fully preserve the detailed information of the speech signal while avoiding data redundancy caused by excessively high sampling rates (the data size of a single 30-second speech is approximately 1.2MB).

[0023] The acquired raw speech signal immediately enters the preprocessing process: Noise reduction: An improved adaptive noise cancellation algorithm is adopted. By comparing environmental noise samples (3 seconds of ambient sound are automatically collected when the system starts), low-frequency noise (such as air conditioner sound) and high-frequency interference (such as electronic device noise) in the background are filtered in real time. After noise reduction, the speech signal-to-noise ratio is improved to more than 35dB. Standardization processing: The amplitude of the speech signal is uniformly adjusted to the range of [-1,1] by using the peak normalization algorithm to eliminate signal intensity fluctuations caused by differences in the vocal intensity of different patients and ensure the consistency of subsequent module analysis.

[0024] The preprocessed audio signal is transmitted to the speech recognition module in real time via TCP / IP protocol, with the transmission delay controlled within 100ms to ensure the continuity of the process.

[0025] 2. Speech Recognition Module The core of the module employs a speech recognition model based on the Transformer architecture. This model has been trained using 100,000 hours of TCM (Traditional Chinese Medicine) consultation speech data, and its recognition ability for TCM terms (such as "phlegm-dampness," "blood stasis," and "qi stagnation") has been specifically optimized. The specific processing steps are as follows: Feature extraction: The preprocessed speech signal is divided into frames of 10ms each. 40-dimensional Mel-frequency cepstral coefficients (MFCCs) and 1-dimensional frame energy are extracted from each frame to form a 41-dimensional feature vector, which comprehensively reflects the spectral characteristics and energy changes of the speech. Model inference: Feature vectors are input into the Transformer encoder, which captures the contextual relationships of the speech sequence through a self-attention mechanism, and the decoder outputs the probability distribution of the corresponding Chinese characters; Post-processing optimization: By combining a language model in the field of traditional Chinese medicine (containing 5,000+ core TCM terms) for beam search decoding, common errors such as misclassifying "yin deficiency" as "yin deficiency" and "night sweats" as "night sweats" are corrected, ultimately stabilizing the text conversion accuracy at over 95.6%.

[0026] The converted text is output in structured JSON format, containing information such as "original speech path", "recognized text", and "confidence score" (0-100 points, those below 70 points are automatically marked as "requires manual review"), and is pushed to the natural language processing module in real time.

[0027] 3. Natural Language Processing Module The module performs text analysis based on the BERT-WWM Chinese pre-trained model, with fine-tuning for the TCM constitution identification scenario. The specific process is as follows: Word segmentation and entity recognition: Using the LTP word segmentation tool from Harbin Institute of Technology, combined with a custom TCM lexicon (containing 2000+ constitution-related terms, such as "fear of cold and cold limbs" and "dry mouth and tongue"), the text is split into lexical units, and entities such as "symptoms", "signs", and "lifestyle habits" are labeled using named entity recognition (NER). For example, in "bitter taste in the mouth every morning", "bitter taste in the mouth" is labeled as "symptom", and "every morning" is labeled as "onset time". Feature parameter extraction: Key features related to physical constitution are extracted using a multi-dimensional algorithm, including: Speech characteristics: speech rate (calculation method: effective number of words / speech duration, unit: words / second, normal range 2-6 words / second), pause characteristics (number of pauses / minute, single pause duration > 0.5 seconds is considered a valid pause), intonation fluctuation (fundamental frequency standard deviation, unit: Hz, reflecting emotional stability); Semantic features: By analyzing sentiment polarity (positive / negative / neutral) and grading symptom severity (e.g., "occasional fatigue" is mild, "persistent fatigue" is severe), textual information is quantified into computable numerical features; Constitution-related features: Based on the theory of constitution in traditional Chinese medicine, a "symptom-constitution" mapping table is constructed (e.g., "fear of cold and cold limbs" corresponds to Yang deficiency constitution, and "dry stools" corresponds to Yin deficiency constitution). The matching degree between the text and each constitution feature is calculated by cosine similarity.

[0028] Ultimately, the module generates a feature dataset with over 200 dimensions, including 150 dimensions of quantitative features and 50 dimensions of semantic labels, achieving an overall extraction accuracy of 91.2%, and transmits it to the body constitution identification module in real time.

[0029] 4. Body Constitution Identification Module The module incorporates nine TCM constitution theory models, strictly adhering to the national standard "Classification and Determination of TCM Constitutions" (GB / T27779-2011). Each model includes 15-20 core characteristics (e.g., the Yang deficiency model includes "aversion to cold," "cold hands and feet," and "preferring hot drinks"). For different population groups (e.g., children, the elderly, and patients with rheumatic immune diseases), the model can automatically adjust the characteristic weights (e.g., increasing the weight of "lower back and knee weakness" in the elderly model).

[0030] The specific determination process is as follows: After receiving the feature dataset from the natural language processing module, the system first performs a preliminary screening using a logistic regression algorithm to exclude body types with a matching degree of less than 30%. For the remaining physical types, support vector machines (SVM) are used for fine classification. The feature data is transformed into a high-dimensional space through kernel function mapping to find the optimal classification hyperplane. Combining the voting mechanism of the random forest algorithm (100 decision trees), the final constitution type and confidence level are output (e.g., "Yang deficiency constitution, confidence level 92.5%)". If the highest confidence level is lower than 60%, it indicates "high probability of mixed constitution, doctor's review is recommended".

[0031] In actual testing, the module achieved an accuracy rate of 92.3% in identifying single constitutions and 88.7% in identifying mixed constitutions, with a consistency rate of 90.1% with the manual judgments of three associate chief physicians of traditional Chinese medicine.

[0032] 5. User Interface The interface adopts a dual-platform design that adapts to both web front-end and mobile app, adhering to the principle of "extremely simple operation" with a response time of ≤1 second. Specific functions are as follows: Guided Questionnaire: A 25-question structured questionnaire covering dimensions such as "cold and heat preference", "food taste", and "sleep quality". Each question provides 3-5 options (e.g., "degree of cold sensitivity: none / slight / obvious / severe"). Voice input is supported (the system's voice recognition module automatically converts the text to text) and includes dynamic illustrations (e.g., "thick and greasy tongue coating" with a high-definition image). Voice acquisition guidance: The voice waveform is displayed in real time, prompting the patient to "describe your physical feelings in the past month naturally, such as 'easily fatigued, sweating a lot', for 20-40 seconds". The process will automatically stop after the timeout and prompt "Acquisition completed"; Results presentation: The results are presented in the form of "body constitution radar chart + text analysis". The radar chart intuitively displays the score (0-100 points) of each body constitution characteristic, and the text analysis includes the body constitution definition, typical manifestations, and susceptible diseases (such as Yang deficiency constitution is prone to rheumatoid arthritis). Health advice: Personalized plans are generated based on body constitution type. For example, for those with Yang deficiency, we recommend "Angelica and Ginger Lamb Soup", "Moxibustion at Guanyuan Acupoint", and "30 minutes of brisk walking per day". The basis for the recommendations is noted (referencing relevant discussions in the "Huangdi Neijing" and modern research literature). Historical records: Display past identification results by timeline, support comparative analysis (e.g., "Yang deficiency characteristic score has decreased by 15 points in the past 6 months"), and data can be exported as PDF or Excel format.

[0033] 6. Database It uses a PostgreSQL 14 database, supports distributed storage, and has a total capacity that can be expanded to 10TB. The specific design is as follows: Data storage: A partitioned table design is adopted, with partitions based on "year + constitution type". The stored content includes: patient basic information (encrypted storage, name and ID number are encrypted using SHA-256), original audio files (compressed storage, WAV format), recognized text, feature dataset, constitution results, physician review records, etc. Security Mechanism: Implement "triple protection" - transmission encryption (HTTPS protocol), storage encryption (AES-256), and access control (based on the RBAC model, patients can only view their own data, and physicians need to obtain approval to access the data). Maintenance and optimization: Tasks are automatically executed daily at 2 AM: data verification (removing invalid questionnaires and muted voice messages), backup (incremental backup + weekly full backup), and index optimization (creating B-tree indexes for the "physical type" and "age group" fields, improving query speed by 50%). Interface integration: Supports integration with hospital HIS systems and electronic health records (EHRs), synchronizing patient medical history data through the standard HL7 protocol to enrich identification reference dimensions.

[0034] 7. Machine Learning Algorithm Module The module employs a hybrid strategy of "incremental learning + active learning" to achieve continuous model optimization. The specific process is as follows: Data screening: 5,000 high-quality data points are extracted from the database each month (must meet the following requirements: clear voice, complete questionnaire, and reviewed by a physician), and divided into training set and validation set in an 8:2 ratio; Model training: Gradient descent algorithm is used to optimize the parameters of SVM and random forest (such as the penalty coefficient C of SVM and the tree depth of random forest), and cross-validation (5-fold) is used to avoid overfitting, thereby improving the accuracy of the validation set by ≥2%; Feature updates: 10-15 new features (such as "tongue description" and "pulse description") are added every quarter. The weights are adjusted through feature importance assessment (based on the Gini coefficient). For example, after adding the feature of "large tongue", the accuracy of identifying phlegm-dampness constitution increased from 89% to 93%. Model deployment: The optimized model is automatically deployed to the production environment through containerization technology (Docker). The old model is retained for 30 days for rollback. The entire iteration cycle is ≤3 months, realizing a closed loop of "data accumulation - model evolution - accuracy improvement".

[0035] II. Complete Operation Process Example Taking a 45-year-old female patient with rheumatoid arthritis as an example, the system's operation process is explained in detail: 1. Patient Login and Information Entry After logging in via a mobile app and completing real-name authentication, the system automatically linked the patient's medical records at the hospital (showing "2-year history of rheumatoid arthritis, currently taking methotrexate"). The patient then filled out a questionnaire as instructed, stating "significant sensitivity to cold," "prefers hot drinks," "easily awakened during sleep," and "joint pain worsens with cold," which took approximately 4 minutes.

[0036] 2. Voice Acquisition and Preprocessing The patient clicked "Start Voice Collection" and described: "I've been feeling cold lately, my hands and feet are always cold, I get tired easily, and my joints are a bit stiff when I wake up in the morning." The system automatically collected 35 seconds of voice recordings, and the sound collection module immediately performed noise reduction (filtering out background TV noise) and standardization processing to generate a clear voice signal.

[0037] 3. Speech recognition and text conversion The preprocessed speech signal is transmitted to the speech recognition module, where MFCCs features are extracted and converted into text using the Transformer model: "I've been feeling cold lately, my hands and feet are always cold, I get tired easily, and my joints are a bit stiff in the morning." After language model correction, the accuracy reaches 98%, and the signal is then pushed to the natural language processing module.

[0038] 4. Natural Language Processing and Feature Extraction After the module segments the text, it identifies symptom entities such as "cold", "cold hands and feet", "tired", and "stiff joints", and extracts feature parameters: speech rate 3.2 words / second, 2 pauses / minute, tone fluctuation 15Hz (stable emotions), and calculates a matching degree of 85% with Yang deficiency constitution and 60% with Qi deficiency constitution.

[0039] 5. Body Constitution Identification and Results Output The constitution identification module combines the patient's medical history (rheumatoid arthritis), calls the "rheumatoid immune disease patient constitution model", analyzes it through SVM and random forest algorithms, outputs "Yang deficiency constitution, confidence level 92%", and marks it as "highly correlated with the onset of rheumatoid arthritis".

[0040] 6. Results Display and Data Storage The interface displays a radar chart (Yang deficiency constitution score 92 points, Qi deficiency constitution score 65 points), with the text explanation "Yang deficiency constitution: insufficient Yang Qi, weakened warming function, prone to feeling cold and joint pain," and health advice "avoid raw and cold foods, moxibustion at the Mingmen acupoint, and moderate practice of the Baduanjin 'Hands Supporting the Sky to Regulate the Triple Burner' posture." All data (voice, text, features, results) is encrypted and stored in the database, simultaneously marked "Pending review by a rheumatologist."

[0041] 7. Model Optimization Feedback After review by the physician, it was confirmed that the patient "met the criteria for Yang deficiency constitution," and the data was included in the training set of the machine learning module. Three months later, the system optimized the model through incremental learning, increasing the accuracy of identifying Yang deficiency constitution in rheumatoid arthritis patients from 92% to 94%.

[0042] III. Core Advantages and Technical Highlights of the Invention 1. Precision surpasses the limitations of traditional methods Compared to traditional "subjective judgment by physicians," the system eliminates differences in human experience and patient description bias by quantifying speech features (such as tone and speed) and semantic features (such as symptom terminology). In practical applications, the identification results of 1,000 samples showed that the system's consistency with 3 chief physicians reached 90.3%, significantly higher than the 78.5% of the traditional method, with a particularly significant improvement in the accuracy of identifying mixed constitutions (from 65% to 88.7%).

[0043] 2. End-to-end automation improves efficiency. The system achieves fully automated processing from "patient input - voice acquisition - recognition and analysis - result output". The time for a single physical condition identification is ≤5 minutes, and it can process more than 10,000 people per day. It meets the needs of large-scale application scenarios such as community health screening and physical examination centers, and solves the pain points of traditional methods that are "dependent on experts and inefficient".

[0044] 3. Deep integration of traditional Chinese medicine theory and modern technology Transforming TCM constitution theory into quantifiable algorithmic models (such as a "symptom-constitution" mapping table), understanding the semantics of TCM terms through natural language processing, and achieving dynamic optimization of the theory through machine learning (such as adding new constitution features) provides a complete path for the modernization of TCM: "theory digitization - technology implementation - clinical validation".

[0045] 4. Highly targeted and suitable for specific groups of people Specialized models are designed for patients with rheumatic and immune diseases, the elderly, and other groups. Combining medical history data improves the accuracy of identification and provides data support for personalized disease prevention (such as recommending warming and dispelling cold treatments for patients with Yang deficiency rheumatoid arthritis) and management, reflecting the traditional Chinese medicine concept of "treating disease before it occurs".

[0046] This specific implementation strictly follows the technical features of the claims. Through modular design, parameter optimization, and process refinement, it fully demonstrates the technical points and advantages of the system and verifies its practicality and innovation in the field of TCM constitution identification.

[0047] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A TCM constitution intelligent identification system, comprising a sound acquisition module, a speech recognition module, a natural language processing module, a constitution identification module, a user interface, and a database, characterized in that: The sound acquisition module uses a high-sensitivity microphone as a sensor, with a sampling frequency set to 16kHz and a bit depth of 16 bits, enabling accurate capture of patient speech information. After acquisition, the original speech signal is preprocessed, using adaptive filters or noise cancellation techniques to remove noise and standardize the speech signal amplitude to a uniform level. The preprocessed speech signal is clear and suitable for subsequent analysis and is transmitted to the speech recognition module in real time. The speech recognition module performs feature extraction on the received preprocessed speech signal, decomposing the signal into time segments (frames) and extracting feature vectors such as Mel-frequency cepstral coefficients (MFCCs). Feature vectors are input into pre-trained deep learning models such as Long Short-Term Memory (LSTM) or Convolutional Neural Networks (CNNs). After decoding, text sequences corresponding to speech segments are generated. These sequences are then smoothed and corrected by a language model post-processing to achieve an accuracy of over 95% in the converted text data. The converted text data is then transmitted to a natural language processing (NLP) module. The NLP module first performs word segmentation on the received text data, breaking it down into individual word units. Then, it identifies key information about the patient's feelings and symptoms through syntactic and semantic analysis. Simultaneously, it extracts constitution-related feature parameters such as intonation changes, speech rate, pause frequency and duration using specialized algorithms, and combines these with sentiment analysis tools for evaluation. The patient's emotional state is analyzed, and the resulting feature parameters are aggregated to form a feature dataset corresponding to the TCM constitution theory model. This dataset is then transmitted to the constitution identification module. The constitution identification module incorporates multiple TCM constitution theory models. After receiving the feature dataset output from the natural language processing module, it uses machine learning algorithms such as Support Vector Machine (SVM) or Random Forest, combined with information such as the patient's age, gender, and medical history, to analyze the feature data. By matching constitution patterns and eliminating low-probability options, it automatically determines and outputs the patient's constitution type. The user interface is designed to be intuitive and user-friendly, providing a clear questionnaire process that guides patients to fill in basic information, medical history, and constitution-related questions. The questionnaire includes necessary prompts and help information. It supports patients issuing voice collection commands, displays the progress of physical fitness identification in real time, and presents identification results and targeted health advice in graphical or textual form. It also provides a historical record query function to facilitate patients tracking health changes. The database adopts an SQL database management system to encrypt and store data such as patient-filled questionnaires, voice samples, and physical fitness identification results. It is classified and managed according to preset modes to ensure structure and query efficiency. Strict access control and data encryption technologies are implemented to protect privacy. Data backup, index reconstruction, and performance optimization are performed regularly. It also supports complex query operations, allowing doctors and researchers to retrieve specific datasets.

2. The TCM constitution intelligent identification system according to claim 1, characterized in that: It also includes a machine learning algorithm module, which extracts historical data (including voice samples, questionnaire answers, and physical fitness identification results) from the database, uses cross-validation technology to train an initial machine learning model to adjust parameters, evaluates performance by comparing the model's prediction results with the actual physical fitness identification results, regularly introduces new data features and advanced algorithms such as deep learning to optimize the model, and redeploys the optimized model to the system to achieve self-optimization every 3 months, continuously improving the accuracy and efficiency of physical fitness identification.

3. The TCM constitution intelligent identification system according to claim 1 or 2, characterized in that: The text analysis technology of the natural language processing module includes multi-level processing logic. Based on word segmentation, syntax and semantic analysis, it uses customized algorithms to quantify and extract parameters such as intonation fluctuation amplitude, speech rate change range (unit: words / second), and pause interval duration. It can also identify specific words related to TCM constitution (such as "fear of cold" and "night sweats"), with feature extraction accuracy exceeding 90%.

4. The TCM constitution intelligent identification system according to any one of claims 1 to 3, characterized in that: The constitution identification module has a built-in TCM constitution theory model that covers common types such as balanced constitution, qi deficiency constitution, yang deficiency constitution, yin deficiency constitution, phlegm-dampness constitution, damp-heat constitution, blood stasis constitution, qi stagnation constitution, and special constitution. Users can select the model version that is suitable for patients of different ages, genders and regions through the user interface. The accuracy rate of constitution type determination is 92%.

5. The TCM constitution intelligent identification system according to any one of claims 1 to 4, characterized in that: The user interface has a response time of less than 1 second and supports both touch and voice control modes. The font size and operation steps are optimized for special groups such as elderly patients. The health advice content combines traditional Chinese medicine constitution theory and modern medical research, including specific solutions such as dietary conditioning, exercise guidance and lifestyle adjustment.

6. The TCM constitution intelligent identification system according to any one of claims 1 to 5, characterized in that: The database has an automatic update mechanism, which verifies and integrates new data every morning at midnight and cleans up redundant historical data every month to ensure data timeliness. It also supports integration with the hospital information system (HIS) to synchronize patients' past medical history and other medical data, further enhancing the reference dimensions for constitution identification.

7. A method for identifying traditional Chinese medicine constitution using the system described in any one of claims 1 to 6, characterized in that... Includes the following steps: s1. After the patient inputs basic information through the user interface, the sound acquisition module collects speech samples at a sampling frequency of 16kHz and a bit depth of 16 bits, and simultaneously completes noise reduction and standardization preprocessing. s2. The speech recognition module extracts MFCCs feature vectors from the preprocessed speech samples, converts them into text sequences through an LSTM or CNN model, and then generates accurate text data through post-processing by a language model. s3. The natural language processing module performs word segmentation, syntactic and semantic analysis on the text data, extracts physical characteristic parameters such as intonation, speech rate, and pauses, as well as symptom description keywords, to form a feature dataset; s4. The constitution identification module calls the adapted TCM constitution theory model, combines SVM or random forest algorithm to analyze the feature dataset, and determines the constitution type by comprehensively considering the patient's basic information; s5. The user interface displays the body type results in real time, along with a health report containing suggestions on diet, exercise, etc. s6. The system automatically encrypts and stores voice samples, text data, feature parameters, physical fitness results, and health reports in the database, supporting subsequent queries and machine learning training.

Images