Voice feature analysis and schizophrenia auxiliary diagnosis system

Abstract

The invention provides a voice feature analysis and schizophrenia auxiliary diagnosis system, and belongs to the technical field of medical auxiliary diagnosis, the voice feature analysis and schizophrenia auxiliary diagnosis system comprises a voice acquisition system, a feature extraction system, a data analysis system and a result output system, a voice signal is collected and converted into a digital signal, a feature extraction system extracts feature parameters from multiple dimensions of a frequency domain, a time domain and a voice range, each unit can be adjusted and optimized according to reality, a data analysis system processes the feature parameters, a proper algorithm is selected for classification and recognition, and accuracy and stability are improved through optimization. According to the schizophrenia diagnosis system and the schizophrenia diagnosis method, through multi-dimensional speech feature analysis and in combination with a machine learning algorithm, the problems that an existing speech analysis method is single in index, low in accuracy and lack of systematicness are solved, and the schizophrenia diagnosis accuracy and objectivity are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical auxiliary diagnosis, and in particular relates to a speech feature analysis and schizophrenia auxiliary diagnosis system. Background Art

[0002] Schizophrenia is a serious mental disorder that has a great impact on the patient's quality of life and social function. At present, the diagnosis of schizophrenia mainly relies on the subjective judgment and experience of clinicians, and lacks objective diagnostic indicators. Speech feature analysis, as a non-invasive detection method, has the advantages of simple operation and low cost. In recent years, it has received widespread attention in the auxiliary diagnosis of schizophrenia. However, the existing speech feature analysis methods have problems such as single analysis indicators, low accuracy, and lack of systematicity, which can hardly meet the needs of clinical diagnosis.

[0003] In order to solve the problems existing in the prior art, the present invention provides a speech feature analysis and auxiliary diagnosis system for schizophrenia. The system performs multi-dimensional analysis of the patient's speech features and combines machine learning algorithms to achieve auxiliary diagnosis of schizophrenia, improve the accuracy and objectivity of diagnosis, and provide a new and effective means for the clinical diagnosis of schizophrenia. Summary of the invention

[0004] The purpose of the present invention is to provide a speech feature analysis and schizophrenia auxiliary diagnosis system, aiming to solve the problems that the existing speech feature analysis methods in the prior art have single analysis indicators, low accuracy, lack of systematicity, etc., and are difficult to meet the needs of clinical diagnosis.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A speech feature analysis and schizophrenia auxiliary diagnosis system, including a speech acquisition system, a speech feature extraction system, a data analysis system and a diagnosis result output system;

[0007] The voice acquisition system is used to collect the patient's voice signal, including a voice acquisition unit, which uses a microphone and other devices to collect voice and convert the collected voice signal into a digital signal; the specific operation steps are as follows:

[0008] Choose a quiet environment and avoid external noise interference;

[0009] Adjust the position and parameters of the microphone to ensure that the patient's voice can be clearly captured;

[0010] Start the speech acquisition device and ask the patient to express speech according to the instructions, such as reading a specific text or having a free conversation;

[0011] The collected voice signal is converted into a digital signal through an analog-to-digital conversion module for subsequent processing;

[0012] The speech feature extraction system is used to extract feature parameters from speech signals, and includes a frequency domain feature extraction unit, a time domain feature extraction unit, and a prosody feature extraction unit; wherein the frequency domain feature extraction unit uses a Mel Frequency Cepstral Coefficient (MFCC) algorithm to extract frequency domain feature parameters of speech signals, the time domain feature extraction unit uses a Linear Prediction Coding Coefficient (LPC) algorithm to extract time domain feature parameters of speech signals, and the prosody feature extraction unit uses a prosody analysis algorithm to extract prosody feature parameters of speech signals, such as fundamental frequency, duration, intensity, and other parameter indicators; the specific operation steps are as follows:

[0013] Frequency domain feature extraction:

[0014] The digital voice signal is divided into frames, and the length of each frame is usually between 20-40 milliseconds;

[0015] Perform fast Fourier transform (FFT) on each frame signal to obtain its spectrum;

[0016] The spectrum is filtered through a Mel filter bank to obtain a Mel spectrum;

[0017] Perform logarithmic operation and discrete cosine transform (DCT) on the Mel spectrum to obtain the Mel frequency cepstral coefficient (MFCC);

[0018] Time domain feature extraction:

[0019] Perform autocorrelation analysis on digital speech signals and calculate the linear prediction coefficient (LPC);

[0020] Other time domain feature parameters, such as linear prediction residual, can be further calculated based on the LPC coefficient;

[0021] Prosodic feature extraction:

[0022] To analyze the fundamental frequency of speech signals, methods such as autocorrelation method and average amplitude difference function method can be used;

[0023] Calculate the duration of the speech signal, that is, the duration of each speech segment;

[0024] The strength of the speech signal can be measured by calculating the root mean square value of the signal.

[0025] The data analysis system is used to analyze and process the extracted speech feature parameters, and includes a data input unit, a classification and recognition unit, and an algorithm optimization unit; the data input unit inputs the extracted speech feature parameters into a machine learning algorithm, the classification and recognition unit uses machine learning algorithms such as support vector machine (SVM) and random forest (RF) to classify and recognize the speech feature parameters to determine whether the patient suffers from schizophrenia, and the algorithm optimization unit uses methods such as cross-validation to optimize the machine learning algorithm to improve the accuracy and stability of diagnosis, wherein the cross-validation parameter can be set to k-fold cross-validation, and the value range of k is [5,10]; the specific operation steps are as follows:

[0026] Data input: The extracted speech feature parameters are organized into a suitable data format and input into the machine learning algorithm;

[0027] Classification and recognition: Select an appropriate machine learning algorithm, such as support vector machine (SVM) or random forest (RF);

[0028] Train the algorithm, using speech feature data with known diagnosis results as a training set;

[0029] Input the voice feature data to be diagnosed into the trained algorithm for classification and recognition to determine whether the patient suffers from schizophrenia;

[0030] Algorithm optimization: Use the k-fold cross-validation method to divide the data set into k parts, take k-1 parts as training sets and 1 part as test sets in turn, and perform multiple training and testing;

[0031] According to the results of cross-validation, adjust the algorithm parameters, such as the kernel function parameters in SVM and the number of trees in random forest, to improve the accuracy and stability of diagnosis;

[0032] The diagnosis result output system is used to output the diagnosis result, including a result presentation unit and a report generation unit; the result presentation unit outputs the diagnosis result in the form of text, chart, etc., and the report generation unit provides a diagnosis report, including the patient's basic information, voice feature analysis results, diagnosis conclusion, etc.; the specific operation steps are as follows:

[0033] Results presentation: The diagnosis results are displayed on the screen in text form, clearly stating whether the patient suffers from schizophrenia and the reliability of the diagnosis;

[0034] You can draw charts, such as bar charts or line charts, to show the distribution of different speech feature parameters and the comparison with normal people;

[0035] Report generation: Collect basic information of patients, such as name, age, gender, etc.;

[0036] Arrange the speech feature analysis results, including the values ​​of various extracted feature parameters;

[0037] Based on the diagnosis results, write a diagnosis conclusion and provide further suggestions and references;

[0038] The above information is integrated to generate a diagnostic report, which can be output in electronic or paper form.

[0039] As a preferred solution of the present invention, when collecting voice signals, the voice collection unit ensures that the collection environment is quiet and avoids external interference.

[0040] As a preferred solution of the present invention, the feature parameters extracted by the frequency domain feature extraction unit, the time domain feature extraction unit and the rhythm feature extraction unit can be adjusted and optimized according to actual conditions.

[0041] As a preferred solution of the present invention, the machine learning algorithm in the classification and recognition unit can be selected and adjusted according to different data sets and application scenarios.

[0042] As a preferred solution of the present invention, the result presentation form and report content of the diagnosis result output system can be customized according to user needs.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. In this solution, through the voice acquisition system, a quiet environment is selected, the microphone position and parameters are adjusted, the patient's voice signal is collected and converted into a digital signal; a professional soundproof room or a specific time period can be used for acquisition, and noise monitoring equipment can be placed. The microphone can also be adjusted according to the patient's condition, music can be played or prompts can be set to help the patient relax, and the collected signal can be pre-processed;

[0045] Multi-dimensional feature extraction: Frequency domain feature extraction: Using the MFCC algorithm, the frame processing parameters and Mel filter group parameters, as well as the MFCC order can be adjusted according to actual conditions;

[0046] Time domain feature extraction: Using the LPC algorithm, you can try different calculation methods and further analyze the linear prediction residual to extract more parameters;

[0047] Rhythmic feature extraction: Extract fundamental frequency through a variety of methods, precisely measure duration and intensity, and conduct comprehensive analysis in combination with other rhythmic features.

[0048] 2. In this scheme, data processing: standardization, screening and dimension reduction of the extracted speech feature parameters;

[0049] Algorithm selection and optimization: Select appropriate machine learning algorithms (such as SVM, RF, ANN, etc.) according to the characteristics of the data set and the application scenario, and optimize the algorithm parameters through various methods, and use incremental learning, data enhancement and other methods to improve algorithm performance;

[0050] The results can be presented in various forms: the diagnosis results can be presented in the form of text, charts, voice broadcasts, animation demonstrations, etc., and the results can also be explained;

[0051] Customized report content: The report can include basic patient information, voice feature analysis results, diagnostic conclusions, suggestions and references, etc. The format and level of detail can be customized according to user needs. DETAILED DESCRIPTION

[0052] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] Example 1

[0054] The present invention provides the following technical solutions:

[0055] A speech feature analysis and schizophrenia auxiliary diagnosis system, including a speech acquisition system, a speech feature extraction system, a data analysis system and a diagnosis result output system;

[0056] The voice acquisition system is used to collect the patient's voice signal, including a voice acquisition unit, which uses a microphone and other devices to collect voice and convert the collected voice signal into a digital signal; the specific operation steps are as follows:

[0057] Choose a quiet environment and avoid external noise interference;

[0058] Adjust the position and parameters of the microphone to ensure that the patient's voice can be clearly captured;

[0059] Start the speech acquisition device and ask the patient to express speech according to the instructions, such as reading a specific text or having a free conversation;

[0060] The collected voice signal is converted into a digital signal through an analog-to-digital conversion module for subsequent processing;

[0061] The speech feature extraction system is used to extract feature parameters from speech signals, and includes a frequency domain feature extraction unit, a time domain feature extraction unit, and a prosody feature extraction unit; wherein the frequency domain feature extraction unit uses a Mel Frequency Cepstral Coefficient (MFCC) algorithm to extract frequency domain feature parameters of speech signals, the time domain feature extraction unit uses a Linear Prediction Coding Coefficient (LPC) algorithm to extract time domain feature parameters of speech signals, and the prosody feature extraction unit uses a prosody analysis algorithm to extract prosody feature parameters of speech signals, such as fundamental frequency, duration, intensity, and other parameter indicators; the specific operation steps are as follows:

[0062] Frequency domain feature extraction:

[0063] The digital voice signal is divided into frames, and the length of each frame is usually between 20-40 milliseconds;

[0064] Perform fast Fourier transform (FFT) on each frame signal to obtain its spectrum;

[0065] The spectrum is filtered through a Mel filter bank to obtain a Mel spectrum;

[0066] Perform logarithmic operation and discrete cosine transform (DCT) on the Mel spectrum to obtain the Mel frequency cepstral coefficient (MFCC);

[0067] Time domain feature extraction:

[0068] Perform autocorrelation analysis on digital speech signals and calculate the linear prediction coefficient (LPC);

[0069] Other time domain feature parameters, such as linear prediction residual, can be further calculated based on the LPC coefficient;

[0070] Prosodic feature extraction:

[0071] To analyze the fundamental frequency of speech signals, methods such as autocorrelation method and average amplitude difference function method can be used;

[0072] Calculate the duration of the speech signal, that is, the duration of each speech segment;

[0073] The strength of the speech signal can be measured by calculating the root mean square value of the signal.

[0074] The data analysis system is used to analyze and process the extracted speech feature parameters, and includes a data input unit, a classification and recognition unit, and an algorithm optimization unit; the data input unit inputs the extracted speech feature parameters into a machine learning algorithm, the classification and recognition unit uses machine learning algorithms such as support vector machine (SVM) and random forest (RF) to classify and recognize the speech feature parameters to determine whether the patient suffers from schizophrenia, and the algorithm optimization unit uses methods such as cross-validation to optimize the machine learning algorithm to improve the accuracy and stability of diagnosis, wherein the cross-validation parameter can be set to k-fold cross-validation, and the value range of k is [5,10]; the specific operation steps are as follows:

[0075] Data input: The extracted speech feature parameters are organized into a suitable data format and input into the machine learning algorithm;

[0076] Classification and recognition: Select an appropriate machine learning algorithm, such as support vector machine (SVM) or random forest (RF);

[0077] Train the algorithm, using speech feature data with known diagnosis results as a training set;

[0078] Input the voice feature data to be diagnosed into the trained algorithm for classification and recognition to determine whether the patient suffers from schizophrenia;

[0079] Algorithm optimization: Use the k-fold cross-validation method to divide the data set into k parts, take k-1 parts as training sets and 1 part as test sets in turn, and perform multiple training and testing;

[0080] According to the results of cross-validation, adjust the algorithm parameters, such as the kernel function parameters in SVM and the number of trees in random forest, to improve the accuracy and stability of diagnosis;

[0081] The diagnosis result output system is used to output the diagnosis result, including a result presentation unit and a report generation unit; the result presentation unit outputs the diagnosis result in the form of text, chart, etc., and the report generation unit provides a diagnosis report, including the patient's basic information, voice feature analysis results, diagnosis conclusion, etc.; the specific operation steps are as follows:

[0082] Results presentation: The diagnosis results are displayed on the screen in text form, clearly stating whether the patient suffers from schizophrenia and the reliability of the diagnosis;

[0083] You can draw charts, such as bar charts or line charts, to show the distribution of different speech feature parameters and the comparison with normal people;

[0084] Report generation: Collect basic information of patients, such as name, age, gender, etc.;

[0085] Arrange the speech feature analysis results, including the values ​​of various extracted feature parameters;

[0086] Based on the diagnosis results, write a diagnosis conclusion and provide further suggestions and references;

[0087] The above information is integrated to generate a diagnostic report, which can be output in electronic or paper form.

[0088] In a specific embodiment of the present invention, 1. Voice collection system: When selecting a quiet environment, you can consider using a professional soundproof room or collecting voices in a relatively quiet time period, such as early morning or late at night. At the same time, you can place noise monitoring equipment at the collection site to monitor the environmental noise level in real time to ensure that it is below a certain threshold to ensure the quality of the collected voice signal.

[0089] When adjusting the position and parameters of the microphone, you can make personalized adjustments based on the patient's height, body shape, and voice characteristics. For example, for patients with soft voices, you can appropriately increase the sensitivity of the microphone; for patients with high-pitched voices, you can adjust the frequency response range of the microphone to better capture their voice characteristics.

[0090] After starting the voice collection device, you can help patients relax by playing some relaxing music or having simple conversations to reduce the impact of tension on voice expression. At the same time, you can set some prompts to guide patients to express themselves in the correct way, such as pronouncing clearly and keeping a steady speed.

[0091] Before the analog-to-digital conversion, the collected voice signal can be preprocessed, such as filtering, noise reduction, etc., to remove interference and noise in the signal. At the same time, the signal can be amplitude normalized to make the signal amplitude within a certain range for subsequent processing and analysis.

[0092] 2. Speech feature extraction system: In the process of frequency domain feature extraction, the frame length and frame shift of the frame processing can be adjusted according to the actual situation. For example, for speech signals with a faster speaking speed, the frame length can be appropriately reduced and the frame shift can be increased to improve the accuracy of feature extraction. At the same time, the parameters of the Mel filter bank can be optimized to better adapt to different speech signal characteristics.

[0093] In the process of time domain feature extraction, multiple methods can be used to calculate the linear prediction coefficient (LPC) and compare the effects of different methods. For example, the autocorrelation method, covariance method and other methods can be used for calculation, and the method with the best effect can be selected for subsequent processing. In addition, the linear prediction residual can be further analyzed to extract more time domain feature parameters, such as residual energy, residual autocorrelation coefficient, etc.

[0094] In the process of prosodic feature extraction, multiple methods can be combined to analyze the fundamental frequency of the speech signal to improve the accuracy of fundamental frequency extraction. For example, the autocorrelation method and the average amplitude difference function method can be used simultaneously to extract the fundamental frequency, and the results of the two methods can be weighted averaged. At the same time, the duration and intensity of the speech signal can be measured more finely, for example, the speech signal can be divided into smaller segments, and the duration and intensity of each segment can be calculated separately to better reflect the prosodic changes of the speech signal.

[0095] 3. Data analysis system: During the data input process, the extracted speech feature parameters can be standardized so that the values ​​of different feature parameters are at the same order of magnitude, so that the machine learning algorithm can better process them. At the same time, the feature parameters can be screened and reduced in dimension to remove some redundant and irrelevant feature parameters, thereby improving the efficiency and accuracy of the algorithm.

[0096] When selecting a machine learning algorithm, you can compare and evaluate multiple algorithms and select the most suitable algorithm for this system. For example, you can compare the performance of algorithms such as support vector machine (SVM), random forest (RF), artificial neural network (ANN), and select the best performing algorithm based on actual conditions. In addition, you can further optimize the parameters of the algorithm, such as adjusting the kernel function type and parameters in SVM, the number and depth of trees in random forest, etc.

[0097] During the algorithm training process, incremental learning methods can be used to continuously update the training set to improve the adaptability and accuracy of the algorithm. At the same time, data enhancement processing can be performed on the training set, such as adding noise, changing the speech speed and pitch of the speech signal, etc., to increase the diversity of the training set and improve the generalization ability of the algorithm.

[0098] In the process of algorithm optimization, a variety of optimization methods can be used to automatically optimize the parameters of the algorithm, such as grid search, random search, genetic algorithm, etc. At the same time, the results of cross-validation can be deeply analyzed to find out the key factors affecting the performance of the algorithm, and make targeted adjustments and optimizations.

[0099] 4. Diagnosis result output system: In the process of result presentation, the diagnosis results can be presented in a variety of forms to meet the needs of different users. For example, in addition to text and chart forms, the diagnosis results can also be presented in voice broadcast, animation demonstration, etc. At the same time, the diagnosis results can be explained and illustrated to help users better understand the meaning of the diagnosis results.

[0100] During the report generation process, the content and format of the report can be customized according to the user's needs. For example, some professional medical terminology explanations, treatment recommendations, etc. can be added to make the report more detailed and comprehensive. At the same time, the report generation process can be automated to improve work efficiency. In addition, the report can be encrypted and signed to ensure the security and authenticity of the report.

[0101] Specifically in this solution, when collecting voice signals, the voice collection unit ensures that the collection environment is quiet to avoid external interference.

[0102] In this embodiment, in order to ensure a quiet collection environment and avoid external interference, the following measures can be taken:

[0103] Choose a special soundproof room as the voice collection place, and decorate the room with sound-absorbing materials, such as sound-absorbing cotton, sound-absorbing panels, etc., to reduce sound reflection and reverberation.

[0104] Before collecting data, conduct noise detection at the collection site to ensure that the ambient noise level is below a certain threshold. You can use professional noise measurement instruments, such as sound level meters.

[0105] Inform the patient to keep quiet during the collection process and avoid making unnecessary sounds, such as coughing, clearing the throat, etc. At the same time, ask the patient to turn off devices that may generate noise, such as mobile phones.

[0106] During the collection process, some soft background music can be played to help patients relax and reduce the impact of tension on voice expression, while also covering up some slight environmental noise.

[0107] For some special cases, such as when the patient cannot collect data in a quiet environment, you can consider using portable soundproofing equipment, such as a soundproof cover, to wrap the patient and the microphone in it to reduce interference from external noise.

[0108] Specifically in this solution, the feature parameters extracted by the frequency domain feature extraction unit, the time domain feature extraction unit and the rhythm feature extraction unit can be adjusted and optimized according to actual conditions.

[0109] In this embodiment: for the frequency domain feature extraction unit, different Mel filter bank parameters, such as the number of filters, frequency range, etc., can be tried to find the most suitable parameter combination for the current data set. At the same time, the order of Mel frequency cepstral coefficients (MFCC) can be adjusted to observe the influence of different orders on the diagnosis results.

[0110] For the time domain feature extraction unit, different linear prediction coding coefficient (LPC) calculation methods can be tried, such as autocorrelation method, covariance method, etc., and the accuracy of feature parameters extracted by different methods on the diagnosis results can be compared. In addition, the linear prediction residual can be further analyzed to extract more time domain feature parameters, such as residual energy, residual autocorrelation coefficient, etc., and observe the contribution of these parameters to the diagnosis results.

[0111] For the prosodic feature extraction unit, different fundamental frequency extraction methods can be tried, such as the autocorrelation method, the average amplitude difference function method, the harmonic product spectrum method, etc., and the accuracy and stability of different methods can be compared. At the same time, the duration and intensity of the speech signal can be measured more finely. For example, the speech signal can be divided into smaller segments, and the duration and intensity of each segment can be calculated separately to better reflect the prosodic changes of the speech signal. In addition, other prosodic features, such as intonation and stress, can be combined for comprehensive analysis to improve the accuracy of the diagnosis results.

[0112] Specifically in this solution, the machine learning algorithm in the classification and recognition unit can be selected and adjusted according to different data sets and application scenarios.

[0113] In this embodiment: Data set characteristic analysis: First, the characteristics of different data sets are analyzed, including data size, data distribution, feature dimensions, etc. For example, if the data set is small, a support vector machine (SVM) that performs well on small sample data can be selected; if the data distribution is more complex, an artificial neural network (ANN) with strong nonlinear fitting capabilities can be considered.

[0114] Analyze the noise in the data set. For data sets with large noise, you can choose algorithms such as random forest (RF) that have certain noise resistance capabilities.

[0115] Application scenario considerations: If the application scenario has high real-time requirements, you can choose an algorithm with faster computing speed, such as linear SVM or decision tree.

[0116] If higher accuracy is required, you can try ensemble learning algorithms such as random forests, gradient boosting trees, etc., which improve the overall performance by combining multiple weak learners.

[0117] Depending on the professional level of the application scenario, for scenarios used by clinical professionals, algorithms with more intuitive and easy-to-interpret output results, such as decision trees or rule induction algorithms, can be selected to help doctors better understand the basis for diagnosis.

[0118] Algorithm adjustment: For the selected machine learning algorithm, its parameters can be adjusted according to the specific situation. For example, for SVM, the kernel function type, regularization parameter, etc. can be adjusted; for random forest, the number of trees, maximum depth and other parameters can be adjusted.

[0119] Cross-validation, grid search and other methods are used to find the optimal parameter combination to improve the performance and generalization ability of the algorithm.

[0120] Specifically in this solution, the result presentation format and report content of the diagnosis result output system can be customized according to user needs.

[0121] In this embodiment: Customization of result presentation: Text presentation: The format and content of the text can be adjusted according to user needs. For example, for professional doctors, detailed technical parameters and analysis process can be provided; for patients' families, concise and easy-to-understand diagnosis conclusions and suggestions can be provided.

[0122] Chart presentation: You can choose different types of charts, such as bar charts, line charts, pie charts, etc., according to your preferences to display the distribution of different speech feature parameters and the diagnosis results. At the same time, you can adjust the color, font size and other appearance attributes of the chart to improve readability.

[0123] Multimedia presentation: For users with special needs, we can provide multimedia presentation of results such as voice broadcast and animation demonstration to help them understand the diagnosis results more intuitively.

[0124] Customization of report content: Basic information: In addition to the patient's name, age, gender and other basic information, other relevant information such as family medical history, past medical history, etc. can be added according to user needs.

[0125] Speech feature analysis results: Speech feature analysis results of different levels of detail can be provided according to the user's professional level and needs. For professional doctors, specific feature parameter values, statistical analysis results, etc. can be provided; for patients' families, concise feature descriptions and comparisons with normal ranges can be provided.

[0126] Diagnostic conclusion: The way the diagnostic conclusion is expressed can be adjusted according to user needs to make it clearer and easier to understand. At the same time, different levels of confidence assessment can be provided so that users can understand the reliability of the diagnostic results.

[0127] Recommendations and references: Based on the diagnosis results, provide users with personalized recommendations and references, such as further examination suggestions, treatment plan recommendations, etc. Recommendations can be provided at different levels of detail based on the user's needs and background.

[0128] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A speech feature analysis and schizophrenia auxiliary diagnosis system, characterized in that: It includes voice collection system, voice feature extraction system, data analysis system and diagnosis result output system; The voice acquisition system is used to collect the patient's voice signal, including a voice acquisition unit, which uses a microphone and other devices to collect voice and convert the collected voice signal into a digital signal; the specific operation steps are as follows: Choose a quiet environment and avoid external noise interference; Adjust the position and parameters of the microphone to ensure that the patient's voice can be clearly captured; Start the speech acquisition device and ask the patient to express speech according to the instructions, such as reading a specific text or having a free conversation; The collected voice signal is converted into a digital signal through an analog-to-digital conversion module for subsequent processing; The speech feature extraction system is used to extract feature parameters from speech signals, and includes a frequency domain feature extraction unit, a time domain feature extraction unit, and a prosody feature extraction unit; wherein the frequency domain feature extraction unit uses a Mel Frequency Cepstral Coefficient (MFCC) algorithm to extract frequency domain feature parameters of speech signals, the time domain feature extraction unit uses a Linear Prediction Coding Coefficient (LPC) algorithm to extract time domain feature parameters of speech signals, and the prosody feature extraction unit uses a prosody analysis algorithm to extract prosody feature parameters of speech signals, such as fundamental frequency, duration, intensity, and other parameter indicators; the specific operation steps are as follows: Frequency domain feature extraction: The digital voice signal is divided into frames, and the length of each frame is usually between 20-40 milliseconds; Perform fast Fourier transform (FFT) on each frame signal to obtain its spectrum; The spectrum is filtered through a Mel filter bank to obtain a Mel spectrum; Perform logarithmic operation and discrete cosine transform (DCT) on the Mel spectrum to obtain the Mel frequency cepstral coefficient (MFCC); Time domain feature extraction: Perform autocorrelation analysis on digital speech signals and calculate the linear prediction coefficient (LPC); Other time domain feature parameters, such as linear prediction residual, can be further calculated based on the LPC coefficient; Prosodic feature extraction: To analyze the fundamental frequency of speech signals, methods such as autocorrelation method and average amplitude difference function method can be used; Calculate the duration of the speech signal, that is, the duration of each speech segment; The strength of the speech signal can be measured by calculating the root mean square value of the signal. The data analysis system is used to analyze and process the extracted speech feature parameters, and includes a data input unit, a classification and recognition unit, and an algorithm optimization unit; the data input unit inputs the extracted speech feature parameters into a machine learning algorithm, the classification and recognition unit uses machine learning algorithms such as support vector machine (SVM) and random forest (RF) to classify and recognize the speech feature parameters to determine whether the patient suffers from schizophrenia, and the algorithm optimization unit uses methods such as cross-validation to optimize the machine learning algorithm to improve the accuracy and stability of diagnosis, wherein the cross-validation parameter can be set to k-fold cross-validation, and the value range of k is [5,10]; the specific operation steps are as follows: Data input: The extracted speech feature parameters are organized into a suitable data format and input into the machine learning algorithm; Classification and recognition: Select an appropriate machine learning algorithm, such as support vector machine (SVM) or random forest (RF); Train the algorithm, using speech feature data with known diagnosis results as a training set; Input the voice feature data to be diagnosed into the trained algorithm for classification and recognition to determine whether the patient suffers from schizophrenia; Algorithm optimization: Use the k-fold cross-validation method to divide the data set into k parts, take k-1 parts as training sets and 1 part as test sets in turn, and perform multiple training and testing; According to the results of cross-validation, adjust the parameters of the algorithm, such as the kernel function parameters in SVM and the number of trees in random forest, to improve the accuracy and stability of diagnosis; The diagnosis result output system is used to output the diagnosis result, including a result presentation unit and a report generation unit; the result presentation unit outputs the diagnosis result in the form of text, chart, etc., and the report generation unit provides a diagnosis report, including the patient's basic information, voice feature analysis results, diagnosis conclusion, etc.; the specific operation steps are as follows: Results presentation: The diagnosis results are displayed on the screen in text form, clearly stating whether the patient suffers from schizophrenia and the reliability of the diagnosis; You can draw charts, such as bar charts or line charts, to show the distribution of different speech feature parameters and the comparison with normal people; Report generation: Collect basic information of patients, such as name, age, gender, etc.; Arrange the speech feature analysis results, including the values ​​of various extracted feature parameters; Based on the diagnosis results, write a diagnosis conclusion and provide further suggestions and references; The above information is integrated to generate a diagnostic report, which can be output in electronic or paper form.

2. The speech feature analysis and schizophrenia auxiliary diagnosis system according to claim 1, characterized in that: When collecting voice signals, the voice collection unit ensures that the collection environment is quiet and avoids external interference.

3. The speech feature analysis and schizophrenia auxiliary diagnosis system according to claim 2, characterized in that: The feature parameters extracted by the frequency domain feature extraction unit, the time domain feature extraction unit and the rhythm feature extraction unit can be adjusted and optimized according to actual conditions.

4. The speech feature analysis and schizophrenia auxiliary diagnosis system according to claim 3, characterized in that: The machine learning algorithm in the classification and recognition unit can be selected and adjusted according to different data sets and application scenarios.

5. The speech feature analysis and schizophrenia auxiliary diagnosis system according to claim 4, characterized in that: The result presentation format and report content of the diagnosis result output system can be customized according to user needs.