An artificial intelligence acceptance system for emergency calls

Abstract

The present invention discloses an artificial intelligence acceptance system for emergency calls, which relates to the technical field of speech recognition and analysis, and includes: an acceptance center, which is used to connect emergency calls and establish an acceptance platform based on artificial intelligence. The acceptance center sets corresponding transfer call nodes according to different departments; when an emergency call is incoming to the acceptance center, the acceptance center processes the voice of the caller, obtains the emotion of the caller, and determines the identity range of the caller, the disease condition characteristics of the caller, and the environmental characteristics according to the emotion of the caller, and conducts patient quality analysis according to the disease condition characteristics and environmental characteristics of the caller. When the present invention accepts emergency calls through artificial intelligence, it first analyzes the emotion in the voice of the caller, preliminarily judges the identity of the caller according to the emotion in the voice of the caller, so as to carry out the corresponding process, and determines the patient quality of the caller according to the description of the disease condition by the caller.

Description

Technical Field

[0001] The present invention relates to the technical field of speech recognition and analysis, and specifically to an artificial intelligence acceptance system for emergency calls. Background Art

[0002] Artificial intelligence acceptance of calls generally refers to a system that uses artificial intelligence technology to handle telephone services. These systems can automatically make calls, answer questions, provide information, etc., and are commonly used in multiple fields such as customer service, marketing promotion, and financial collection. They can improve work efficiency, reduce costs, and provide services 24 hours a day.

[0003] The artificial intelligence acceptance system for calls, also known as an AI call robot, has the following disadvantages:

[0004] Limitations in complex interactions: There are limitations in dealing with complex emotions and needs, and it is difficult to fully understand the diversity and complexity of humans. Lack of humanized communication: Robots lack true compassion and emotions, and it is difficult to provide emotional support and humanized services. Technical dependence: The operation depends on advanced technology, and technical failures may lead to service interruptions. Privacy and security issues: Handling a large amount of personal data may lead to privacy leaks and data security problems. Customer acceptance: Some customers may be reluctant to communicate with robots and prefer to communicate with real people.

[0005] The limitations brought by artificial intelligence acceptance of calls are relatively large. By identifying keywords in the caller's voice, questions are answered based on the keywords. When accepting emergency calls through artificial intelligence, simply identifying keywords in the caller's voice to handle the emergency not only fails to improve the rescue efficiency but also causes the caller to make incorrect operations, affecting the treatment.

[0006] Publication No. CN116682436A discloses an emergency police situation acceptance information recognition method, including: obtaining truncation information for truncating the feature sequence of the voice signal based on the input voice signal; truncating the feature sequence into multiple subsequences based on the truncation information, and splitting the multiple subsequences to obtain independent word units; inputting the multiple word units into a trained recognition model to obtain the recognized text, and combining the text into the corresponding event.

[0007] Since the identities of the callers of emergency calls are different, the degree of understanding of the patient's condition by the callers is also inconsistent. Programmed acceptance based on artificial intelligence cannot minimize the treatment time for the patient, and due to the different symptoms of the patient, it is impossible to analyze whether an ambulance should be dispatched for the treatment of the patient. Summary of the Invention

[0008] One of the objectives of the present invention is to provide an artificial intelligence acceptance system for emergency calls. By analyzing the emotional state of the caller's voice, the relationship between the caller and the patient is narrowed down. According to the emotional state of the caller's voice, corresponding procedures are executed, and the quality of the patient is analyzed to determine whether the patient needs to dispatch an ambulance.

[0009] To achieve the above objectives, the present invention is realized through the following technical solutions: An artificial intelligence acceptance system for emergency calls, comprising:

[0010] An acceptance center, which is used to connect emergency calls and establish an acceptance platform based on artificial intelligence. The acceptance center sets corresponding transfer call nodes according to different departments, and the acceptance platform accepts and diverts emergency calls.

[0011] When an emergency call is made to the acceptance center, the acceptance center processes the caller's voice, obtains the caller's emotion, determines the identity range of the caller, the patient's condition characteristics, and environmental characteristics according to the caller's emotion. Based on the patient's condition characteristics and environmental characteristics, the patient quality is analyzed, and the transfer call of the transfer call node is carried out according to the patient quality and condition characteristics.

[0012] The acceptance center includes:

[0013] A voice acquisition unit, which acquires the caller's voice when the acceptance center accepts an emergency call and preprocesses the voice.

[0014] A voice emotion recognition unit, which is used to recognize the emotion characteristics contained in the caller's voice and determine the identity range of the caller according to the caller's emotion characteristics, including: a first identity range, a second identity range, and a third identity range.

[0015] A voice feature extraction unit, including a disease condition feature extraction module and an environmental feature extraction module, which extracts the disease condition characteristics and environmental characteristics in the caller's voice when the acceptance center accepts the caller.

[0016] A quality analysis unit, which analyzes the patient quality based on the disease condition characteristics and environmental characteristics and feeds back the corresponding patient quality to the acceptance platform.

[0017] In one or more embodiments of the present invention, the environmental feature extraction module includes a weather module and a time module for obtaining the weather in the area and the acceptance time of the emergency call. The time module records and calibrates the acceptance time of the emergency call, and the weather module is used to obtain the weather environment in the acceptance area at the acceptance time.

[0018] In one or more embodiments of the present invention, the voice acquisition module includes:

[0019] An A / D conversion module, which converts the voice signal into a digital signal.

[0020] A digital filter that emphasizes the high-frequency part of speech, removes the influence of lip radiation, and increases the high-frequency resolution of speech;

[0021] A frame segmentation and windowing module that divides the caller's speech signal into short-time frames for processing, applies a window function to each short-time frame to emphasize the signal at the frame center and weaken the signals at both ends;

[0022] An endpoint detection module that determines the start and end points of the speech signal, removes non-speech parts, and extracts effective speech content.

[0023] In one or more embodiments of the present invention, the sound acquisition module inputs the speech content and non-speech parts to the sound emotion recognition unit for sound recognition. The sound emotion recognition unit recognizes the emotional characteristics of the caller and determines the range of the caller's identity. The sound emotion recognition unit includes:

[0024] An emotion feature extraction module that extracts emotion-related features in the speech content, including pitch, volume, and speech rate;

[0025] An emotion recognition module that accesses machine learning algorithms to analyze the emotion features and recognize the emotional state in the speech content;

[0026] A result output module that outputs the recognized emotional state to the acceptance platform.

[0027] In one or more embodiments of the present invention, the steps for determining the range of the caller's identity are as follows:

[0028] Determine the distribution of these features in a sentence of the emotion features through statistical analysis, including the mean and variance, and determine the intensity and range of the emotion;

[0029] Map the recognized emotional state to the emotion space, and determine the caller's identity according to the emotion intensity and range;

[0030] Among them, statistical analysis is carried out by collecting speech data containing expressions for model training, calculating the frequency distribution of different emotion categories in the dataset, and verifying the accuracy and generalization of the model through cross-validation;

[0031] The variance is a statistic that measures the degree of dispersion of data distribution, calculates the average value of the sum of the squares of the differences between each data point and the mean, and the calculation formula for the population variance is

[0032] where μ is the mean and N is the total number of data points;

[0033] The calculation formula for the sample variance is

[0034] where χ is the sample mean and n is the sample size.

[0035] In one or more embodiments of the present invention, after determining the range of the caller's identity through the voice content, a disease condition feature and an environmental feature are extracted by a voice feature extraction unit, and the voice feature extraction unit further includes:

[0036] A voice recognition module, configured to convert the voice content into text, establish a language library, and complete unclear voices through the matching of the language library;

[0037] A disease condition feature extraction module obtains the text information and extracts the disease condition features included in the text;

[0038] A summarization module summarizes the disease condition features and the environmental features to generate a feature set.

[0039] In one or more embodiments of the present invention, an environmental feature database is established, environmental sounds are collected, and a machine learning model is established to identify the environmental sounds, analyze the approximate location of the caller, and add the location of the caller as an additional feature to the feature set.

[0040] In one or more embodiments of the present invention, the quality analysis unit includes:

[0041] A disease condition matching module forms a communication with the summarization module and analyzes the disease condition features in the feature set, and matches the corresponding diseases through the disease condition features;

[0042] A disease classification module classifies the diseases matched with the disease condition features into different departments;

[0043] A quality analysis module classifies the patients into three levels: ordinary, urgent, and dangerous according to the diseases corresponding to the patients and the disease condition features. Among them, the diseases and disease condition features that do not require the dispatch of an ambulance are of ordinary quality.

[0044] In one or more embodiments of the present invention, the disease condition matching module establishes a disease model, and the specific steps are as follows:

[0045] Data collection: Collect a data set containing disease condition features, and the data comes from clinical records, medical records, and laboratory test results;

[0046] Feature selection: Select features related to disease diagnosis from the data set, including symptoms, signs, and laboratory indicators;

[0047] Model construction: Use machine learning methods to construct a model;

[0048] Performance evaluation: Evaluate the accuracy, sensitivity, and specificity of the model;

[0049] Model verification: Verify the generalization ability of the model on different data sets to ensure the applicability of the model;

[0050] Among them, the disease condition features match the disease model, and the disease range corresponding to the disease condition features is determined through the disease model.

[0051] In one or more embodiments of the present invention, when the receiving platform is unable to determine the specific disease of the patient, a transfer call is made according to the patient quality analysis result and the department corresponding to the disease condition features, and the call is transferred to the department corresponding to the disease condition, and the disease condition features are summarized and pushed to the corresponding transfer node of the department while transferring to the corresponding department.

[0052] Through the above technical solutions, the present invention has the following beneficial effects:

[0053] 1. When the present invention receives an emergency call through artificial intelligence, it first analyzes the emotion in the caller's voice, preliminarily judges the identity of the caller according to the emotion in the caller's voice, and then proceeds with the corresponding process. It also determines the patient quality of the caller according to the description of the patient's condition by the caller, and makes corresponding operations according to the call quality.

[0054] 2. When analyzing the emotional changes in the voice, emotional-related features such as pitch, volume, and speech rate after preprocessing the extracted voice are used to determine the emotional state of the voice. According to the emotional state of the voice, the identity range of the caller is determined, and the corresponding acceptance procedure is optimized according to the different identity ranges of the caller and the corresponding situations.

[0055] 3. When the caller describes the patient's condition, corresponding screening is carried out according to the different descriptions of the patient's condition by the caller. When a complex situation occurs, the disease range of the patient is determined to transfer the corresponding medical staff. When transferring the medical staff, the key situation features of the patient are refined, so as to shorten the corresponding treatment time.

[0056] 4. When analyzing the call quality of the caller, the environmental information of the patient and the description of the patient's condition are extracted, and the patient quality analysis is carried out according to the weather environment conditions of the corresponding region. Corresponding screening is carried out according to the corresponding disease condition, so as to avoid ordinary patients affecting the treatment of high-risk patients and improve the accuracy of dispatched ambulances. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is the system flow chart of the present invention;

[0058] Figure 2 is the schematic diagram of the voice feature extraction unit of the present invention;

[0059] Figure 3 is the schematic diagram of the quality analysis unit of the present invention;

[0060] Figure 4 is the schematic diagram of the voice acquisition unit of the present invention;

[0061] Figure 5 Schematic diagram of the voice emotion unit of the present invention. Specific implementation manners

[0062] The following will disclose multiple implementation manners of the present invention with the accompanying drawings. For the sake of clear description, many practical details will be described together in the following narration. However, it should be understood that these practical details should not be used to limit the present invention. That is to say, in some implementation manners of the present invention, these practical details are not necessary. And if possible in implementation, the features of different embodiments can be applied interactively.

[0063] Unless otherwise defined, all the terms (including technical and scientific terms) used herein have their ordinary meanings, and their meanings can be understood by those skilled in this field. Further, the definitions of the above terms in commonly used dictionaries should be interpreted as having the same meanings consistent with the relevant fields of the present invention in the content of this specification. Unless specifically defined otherwise, these terms will not be interpreted as idealized or overly formal meanings.

[0064] Please refer to Figures 1-5 , the present invention provides an artificial intelligence acceptance system for emergency calls, which accepts emergency calls and judges the quality of the accepted emergency calls, so as to accurately carry out corresponding treatments.

[0065] In one embodiment, the acceptance and quality analysis system includes:

[0066] An acceptance center, which is used to connect emergency calls and establish an acceptance platform based on artificial intelligence. The acceptance center sets corresponding transfer call nodes according to different departments, and the acceptance platform conducts the acceptance and diversion of emergency calls.

[0067] When an emergency call is incoming to the acceptance center, the acceptance center processes the voice of the caller, obtains the emotion of the caller, determines the identity range of the caller, the disease characteristics of the caller and the environmental characteristics according to the emotion of the caller, conducts patient quality analysis according to the disease characteristics and environmental characteristics of the caller, and transfers the call node according to the patient quality and disease characteristics.

[0068] The acceptance center identifies and analyzes the voice of the caller, obtains the reason for calling for help, the pick-up address and the contact phone number of the caller. The acceptance and diversion of the call are conducted according to the reason for calling for help. For major emergencies or situations where the number of incoming calls to the 120 emergency center surges during an epidemic outbreak, automatic diversion is carried out according to the urgency of the call, which is of great significance for the emergency center to better respond to urban 120 emergencies in case of emergencies.

[0069] The acceptance center includes:

[0070] A voice collection unit collects the voice of the caller when the acceptance center accepts an emergency call and preprocesses the voice.

[0071] A voice emotion recognition unit is used to recognize the emotion features contained in the caller's voice and determine the identity range of the caller according to the caller's emotion features, including: a first identity range, a second identity range, and a third identity range.

[0072] A voice feature extraction unit includes a disease condition feature extraction module and an environmental feature extraction module, which extracts the disease condition features and environmental features in the caller's voice when the acceptance center accepts the caller.

[0073] A quality analysis unit conducts a quality analysis of the patient based on the disease condition features and environmental features and feeds back the corresponding patient quality to the acceptance platform.

[0074] In this embodiment, the acceptance center processes the emergency call, processes the voice of the caller, obtains the disease condition features, environmental features, and the caller's emotion features of the patient, and conducts corresponding processing for the disease condition features, environmental features, and emotion features in the voice, so as to improve the accuracy of processing for different callers. Since the identities of the callers of the emergency call are different, therefore, corresponding acceptance adjustments are made for callers with different identities to shorten the processing time for the emergency.

[0075] Among them, recognizing the emotion features in the caller's voice can obtain the state of the caller, that is, analyze the relationship between the caller and the patient, and divide the identity range into a first identity range, a second identity range, and a third identity range. The first identity range is the patient himself / herself, the second identity range is the accompanying person, and the third identity range is the passerby.

[0076] In one embodiment, the environmental feature extraction module includes a weather module and a time module for obtaining the weather in this area and the acceptance time of the emergency call. The time module records and calibrates the acceptance time of the emergency call, and the weather module is used to obtain the weather environment in the acceptance area at the acceptance time moment.

[0077] In this embodiment, the sudden symptoms corresponding to different seasons are different. Therefore, by obtaining the time and weather, the corresponding disease symptoms of the disease condition in the corresponding environment can be obtained more accurately.

[0078] Exemplarily, the diseases that are prevalent in the corresponding seasons:

[0079] Summer: Heatstroke: High temperature weather is likely to cause heatstroke. Attention should be paid to replenishing water and avoiding long-term activities under the scorching sun.

[0080] Water activity accidents: The risk of water activity accidents such as swimming and rowing increases. Life jackets should be worn and safety rules should be followed.

[0081] Winter: Cardiovascular and cerebrovascular diseases: In winter, the temperature is low, and the incidence of cardiovascular and cerebrovascular diseases increases. Attention should be paid to keeping warm and avoiding strenuous exercise. Carbon monoxide poisoning: When using a coal stove or gas for heating, attention should be paid to ventilation to prevent carbon monoxide poisoning.

[0082] Heatstroke and carbon monoxide poisoning can also cause people to coma, but based on the time and weather, it is possible to quickly distinguish between heatstroke and carbon monoxide poisoning.

[0083] In one embodiment, the sound acquisition module includes:

[0084] An A / D conversion module that converts the sound signal into a digital signal;

[0085] A digital filter that emphasizes the high-frequency part of the speech, removes the influence of lip radiation, and increases the high-frequency resolution of the speech;

[0086] A framing and windowing module that divides the caller's speech signal into short-time frames for processing, and applies a window function, such as a Hamming window or a Hanning window, to each short-time frame to emphasize the signal at the center of the frame and weaken the signals at both ends, avoiding edge effects during Fourier transform;

[0087] An endpoint detection module that determines the start point and end point of the speech signal, removes non-speech parts, such as silence or noise, to extract effective speech content.

[0088] In this embodiment, effective speech content is obtained through the sound acquisition module, thereby completing the preprocessing of the caller's speech content, intercepting useful speech content, reducing the influence of lip radiation, edge effects, and silence noise on feature extraction in the speech, and further shortening the processing time for the caller's speech content, and further improving the response time of the acceptance platform.

[0089] Optionally, the endpoint detection module further splits the non-speech part into breathing sounds and ambient sounds, and transmits the ambient sounds to the ambient feature extraction module and the breathing sounds to the disease condition feature extraction module for corresponding feature extraction to assist in the recognition and processing of disease condition features and ambient features.

[0090] In one embodiment, the sound acquisition module inputs the speech content and non-speech part into the sound emotion recognition unit for sound recognition. The sound emotion recognition unit recognizes the emotional characteristics of the caller and determines the identity range of the caller. The sound emotion recognition unit includes:

[0091] An emotion feature extraction module that extracts emotion-related features in the speech content, including pitch, volume, and speech rate. Exemplarily, it is extracted through Mel Frequency Cepstral Coefficients (MFCC);

[0092] The emotion recognition module accesses machine learning algorithms to analyze emotion features and identify the emotional state in the speech content;

[0093] The result output module outputs the recognized emotional state to the acceptance platform.

[0094] In this embodiment, the machine learning algorithm is a convolutional neural network (CNN). It recognizes the speech content of the caller to obtain the situation of the caller, determines the identity range of the caller, and thus judges the degree of the caller's understanding of the illness. Then, different acceptance strategies are adopted based on artificial intelligence, so as to be able to understand the information most beneficial to making decisions for the patient in the shortest time and shorten the treatment time for the patient.

[0095] Exemplarily, when the caller is in the third identity range, first briefly understand the patient's status, and then determine the patient's location;

[0096] When the caller is in the second identity range, determine the disease state and the state before the patient developed the disease, and then determine the patient's location;

[0097] When the caller is in the first identity range, determine the patient's detailed feelings about the disease and the patient's location.

[0098] The present invention only provides an acceptance strategy when the caller is in the first, second, and third identity ranges, and does not limit the acceptance strategy. Those skilled in the art can adjust it according to the usage environment to better optimize the acceptance process.

[0099] In one embodiment, the steps to determine the caller's identity range are as follows:

[0100] Determine the distribution of these features of the emotion features in a sentence through statistical analysis, including the mean and variance, and determine the intensity and range of the emotion;

[0101] Map the recognized emotional state into the emotion space, and determine the caller's identity according to the emotion intensity and range;

[0102] Among them, statistical analysis trains the model by collecting speech data containing expressions, calculates the frequency distribution of different emotion categories in the dataset, and verifies the accuracy and generalization of the model through cross-validation.

[0103] In this embodiment, the statistical analysis specifically includes the following steps:

[0104] Data collection: First, a large amount of speech data containing emotional expressions needs to be collected.

[0105] Feature extraction: Extract emotion-related features from the speech data, such as pitch, volume, speech rate, etc. These features can be extracted using methods such as Mel Frequency Cepstral Coefficients (MFCC).

[0106] Annotated data: Manually annotate the collected voice data to determine the emotional category of each sample, such as happy, sad, angry, etc.

[0107] Model training: Use machine learning algorithms, such as Naive Bayes, Support Vector Machine (SVM), deep learning models, etc., to train the features and annotated data.

[0108] Statistical analysis:

[0109] Frequency distribution: Calculate the frequency distribution of different emotional categories in the dataset.

[0110] Mean and variance: Calculate the mean and variance of the features of each emotional category, and analyze the central tendency and dispersion degree of the emotional features.

[0111] Correlation analysis: Analyze the correlation between different emotional features to determine which features have a greater impact on emotional classification.

[0112] Hypothesis testing: Conduct statistical hypothesis testing, such as t-test or ANOVA, to determine whether the differences between different emotional categories are significant.

[0113] Model evaluation: Use methods such as cross-validation to evaluate the accuracy and generalization ability of the model.

[0114] Result interpretation: Interpret different emotional categories based on the results of statistical analysis.

[0115] Exemplarily, frequency distribution calculation:

[0116] Determine the maximum and minimum values of the dataset.

[0117] Divide the data into several groups, and the number of groups can be adjusted according to the characteristics and needs of the data.

[0118] Calculate the group interval, that is, the width of each group, which can be obtained by dividing the difference between the maximum and minimum values (range) by the number of groups.

[0119] Calculate the boundaries of each group to ensure that the boundaries of each group are clear for counting the frequency of each group.

[0120] Mean calculation:

[0121] For a set of data, the mean (average value) is the sum of all data values divided by the number of data.

[0122] Variance calculation:

[0123] Variance is a statistic that measures the dispersion degree of data distribution, and calculates the average of the sum of the squares of the differences between each data point and the mean.

[0124] The calculation formula for the population variance is

[0125] Where μ is the mean and N is the total number of data points.

[0126] The calculation formula for the sample variance is

[0127] Where χ is the sample mean and n is the sample size.

[0128] Obtain the mean values of the emotional features related to emergencies such as anxiety, sadness, and panic. When the emotional feature of the caller is less than the mean value, it is determined as the third identity range, and when it is equal to or greater than the mean value, it is determined as the second identity range.

[0129] The first identity range matches the emotional feature through the patient's condition. After the relevant emotional feature matching the patient's condition is met, the caller is determined as the first identity range.

[0130] In one embodiment, after determining the identity range of the caller through the voice content, a disease condition feature and an environment feature are extracted by a voice feature extraction unit. The voice feature extraction unit further includes:

[0131] A speech recognition module, configured to convert the voice content into text, establish a language library, and complete the unclear voice through the matching of the language library;

[0132] The disease condition feature extraction module obtains the text information and extracts the disease condition features included in the text;

[0133] A summarization module, which summarizes the disease condition features and the environment features to generate a feature set.

[0134] Optionally, in another embodiment, the emotion recognition unit includes an emotion feature extraction module, which extracts emotion-related features in the voice content, including pitch, volume, and speech rate; an emotion recognition module, which accesses a machine learning algorithm to analyze the emotion features and identify the emotion state in the voice content;

[0135] An automatic acceptance module: accepts and diverts emergency calls and non-emergency calls according to the recognition result. For non-emergency consultation calls, intelligent automatic acceptance can be performed, and corresponding intelligent interactive acceptance can be performed according to the content of the knowledge base. For emergency calls, they can be transferred to the manual seat for processing.

[0136] In this embodiment, since the voice emotion recognition unit has already recognized the emotion of the caller, when extracting the disease condition features, only the content spoken by the caller's voice needs to be obtained. Therefore, converting the voice content into text can enable faster recognition.

[0137] Moreover, due to the differences in local dialects, the language library is established based on the local dialect. For the voice of relevant disease characteristics, a language library is established. After converting the voice into text, the language library is used to replace some content in the text to reduce the misrecognition problem of the voice recognition module and further ensure the accuracy of the recognized text.

[0138] In one embodiment, an environmental feature database is established, environmental sounds are collected, and a machine learning model is established to recognize the environmental sounds, analyze the approximate location of the caller, and add the location of the caller as an additional feature to the feature set.

[0139] In this embodiment, the additional feature, as an additional feature for determining the patient's characteristics, can assist in determining the patient's condition. This can narrow down the scope of the patient's condition and further accurately determine the patient's condition status.

[0140] Exemplarily, when the patient is in a coma, it is determined that the patient is outdoors by collecting environmental sounds containing vehicle driving noise through the environmental sound. The additional feature can eliminate the disease matching such as carbon monoxide poisoning in an enclosed space.

[0141] Among them, the environmental sound, as an additional feature, only functions when there are multiple corresponding diseases in the case of disease feature matching. After the disease feature accurately matches the disease, the additional feature has no effect.

[0142] In one embodiment, the quality analysis unit includes:

[0143] A disease matching module that communicates with the summary module and analyzes the disease characteristics in the feature set to match the corresponding diseases through the disease characteristics;

[0144] A disease classification module that classifies the diseases matching the disease characteristics into different departments;

[0145] A quality analysis module that classifies the patient into three grades: ordinary, urgent, and dangerous according to the disease and disease characteristics corresponding to the patient. Among them, the diseases and disease characteristics that do not require the dispatch of an ambulance are of ordinary quality.

[0146] In this embodiment, performing quality analysis on the patient's disease and disease characteristics can better handle different quality patients separately, thereby ensuring the best treatment time and treatment method. Among them, urgent corresponds to the acute and non-acute grades in the emergency level, and dangerous corresponds to the endangered and critical grades in the emergency level.

[0147] Exemplarily, for the general population, coma is divided into multiple situations. Among them, heatstroke, as the main symptom of coma in summer, can be relieved by carrying the patient to a ventilated and shady place and pouring cold water to lower the patient's body temperature. Subsequently, the body temperature change should be continuously monitored, which can better relieve the heatstroke situation. Therefore, heatstroke coma with relatively mild fever as the disease characteristic is of ordinary quality, and when the body temperature of the heatstroke coma is around 40° and cannot be relieved, the quality of the patient is urgent.

[0148] In one embodiment, the disease condition matching module establishes a disease condition model, and the specific steps are as follows:

[0149] Data collection: Collect a dataset containing disease condition characteristics, and the data comes from clinical records, medical records, and laboratory test results.

[0150] Feature selection: Select features related to disease diagnosis from the dataset, including symptoms, signs, and laboratory indicators.

[0151] Model construction: Use machine learning methods to construct a model. Common methods include logistic regression, decision tree, random forest, support vector machine, etc.

[0152] Performance evaluation: Evaluate the accuracy, sensitivity, and specificity of the model. This is usually done through cross-validation or using an independent test dataset.

[0153] Model validation: Validate the generalization ability of the model on different datasets to ensure the applicability of the model.

[0154] Among them, the disease condition characteristics match the disease condition model, and the disease condition range corresponding to the disease condition characteristics is determined through the disease condition model.

[0155] In this embodiment, by matching the disease condition model through the disease condition characteristics, one or more symptoms corresponding to the patient's characteristic set can be determined, thereby narrowing the disease condition range corresponding to the patient and ensuring the accuracy of disease condition determination, so as to accurately diagnose the patient.

[0156] The disease condition model adjusts the model parameters according to the results of performance evaluation to improve the ability of the model to determine the disease condition. With the acquisition of new features, the model is updated regularly to maintain its accuracy and relevance.

[0157] In one embodiment, when the acceptance platform cannot determine the specific disease condition of the patient, a transfer call is made according to the patient quality analysis result and the department corresponding to the disease condition characteristics, and the patient is transferred to the department corresponding to the disease condition. The disease condition characteristics are summarized and pushed to the corresponding transfer node of the department at the same time as the transfer call to the corresponding department.

[0158] In this embodiment, by summarizing the disease characteristics, the medical staff in the corresponding department can understand the situation of the patient more quickly. After transferring the call to the corresponding medical staff, the situation of repeatedly asking certain questions can be avoided, further reducing the response time of the emergency call, thereby improving the treatment efficiency of the patient.

[0159] In summary, the technical solutions disclosed in the above embodiments of the present invention have at least the following advantages:

[0160] 1. When the present invention accepts an emergency call through artificial intelligence, it first analyzes the emotion in the caller's voice, preliminarily judges the identity of the caller based on the emotion in the caller's voice, and then proceeds with the corresponding process. It determines the patient quality of the caller based on the description of the patient's condition by the caller, and makes corresponding operations according to the call quality.

[0161] 2. When analyzing the emotional changes in the voice, extract the emotional-related features such as pitch, volume, speech rate, etc. after preprocessing the voice, determine the emotional state of the voice, determine the identity range of the caller based on the emotional state of the voice, and optimize the corresponding acceptance procedure according to the different identity ranges of the caller and the corresponding situations.

[0162] 3. When the caller describes the patient's situation, make corresponding selections according to the different descriptions of the patient's situation by the caller. When a complex situation occurs, determine the range of the patient's condition and transfer the call to the corresponding medical staff accordingly. When transferring the call to the medical staff, extract the key situation characteristics of the patient, so as to shorten the corresponding treatment time.

[0163] 4. When analyzing the call quality of the caller, extract the environmental information of the patient and the description of the patient's condition, analyze the patient quality according to the weather environment conditions in the corresponding area, and make corresponding selections according to the corresponding condition, so as to avoid ordinary patients affecting the treatment of high-risk patients and improve the accuracy of dispatching ambulances.

[0164] Although the present invention is disclosed in combination with the above embodiments, it is not intended to limit the present invention. Any person skilled in the art can make various modifications and refinements without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention should be defined by the appended claims.

Claims

1. An artificial intelligence emergency call handling system, characterized in that: include: The acceptance center is used to answer emergency calls and establish an acceptance platform based on artificial intelligence. The acceptance center sets up corresponding transfer nodes for different departments, and the acceptance platform accepts and diverts emergency calls; Emergency calls are received by the reception center, which processes the caller's voice and determines the caller's identity range, patient's condition characteristics and environmental characteristics based on the caller's voice. The center also performs patient quality analysis based on the patient's condition characteristics and environmental characteristics, and transfers the call to the transfer node based on the patient's quality and condition characteristics. The acceptance centers include: The sound collection unit collects the caller's voice when the receiving center receives the emergency call and pre-processes the voice; A voice emotion recognition unit is used to recognize the emotion features contained in the caller's voice, and determine the caller's identity range according to the caller's emotion features, including: a first identity range, a second identity range, and a third identity range. The first identity range is the patient himself, the second identity range is the companions, and the third identity range is the passers-by; The voice feature extraction unit includes a disease feature extraction module and an environmental feature extraction module, and the reception center extracts the disease feature and environmental feature from the caller's voice when receiving the caller; The quality analysis unit analyzes the patient quality based on the disease characteristics and environmental characteristics, and feeds back the corresponding patient quality to the acceptance platform; The environmental feature extraction module includes a weather module and a time module for obtaining the weather in the area and the time of emergency call acceptance. The time module records and verifies the time of emergency call acceptance, and the weather module is used to obtain the weather environment of the acceptance area at the time of acceptance. The sound collection unit includes: A / D conversion module, converting sound signals into digital signals; Digital filter, which emphasizes the high-frequency part of speech, removes the influence of lip radiation and increases the high-frequency resolution of speech; The frame-splitting and windowing module divides the caller's voice signal into short-time frames for processing, applies a window function to each short-time frame, emphasizes the signal in the center of the frame and weakens the signals at both ends; Endpoint detection module determines the starting and ending points of the speech signal, removes the non-speech part, and extracts the effective speech content; The voice collection unit inputs the voice content and non-voice part to the voice emotion recognition unit for voice recognition. The voice emotion recognition unit recognizes the emotional characteristics of the caller and determines the identity range of the caller. The voice emotion recognition unit includes: Emotional feature extraction module, which extracts emotion-related features from speech content, including pitch, volume, and speech speed; The emotion recognition module uses machine learning algorithms to analyze emotional features and identify the emotional state in speech content; The result output module outputs the identified emotional state to the acceptance platform.

2. The artificial intelligence emergency call handling system according to claim 1, characterized in that: After the caller's identity range is determined by the voice content, the condition characteristics and environmental characteristics are extracted by the sound feature extraction unit, and the sound feature extraction unit also includes: The speech recognition module is used to convert speech content into text and establish a language library to complete unclear speech by matching the language library; The disease characteristic extraction module obtains text information and extracts the disease characteristics contained in the text; The summary module summarizes the disease characteristics and environmental characteristics to generate a feature set.

3. The artificial intelligence emergency call handling system according to claim 2, characterized in that: Establish an environmental feature database, collect environmental sounds, and build a machine learning model to identify environmental sounds, analyze the approximate location of the caller, and add the caller's location as an additional feature to the feature set.

4. The artificial intelligence emergency call handling system according to claim 3, characterized in that: The quality analysis unit includes: The disease matching module communicates with the aggregation module and analyzes the disease features in the feature set, matching the corresponding disease through the disease features; The disease classification module divides the diseases matching the disease characteristics into different departments; The quality analysis module divides patients into three levels: ordinary, emergency, and critical according to their corresponding symptoms and disease characteristics. Among them, the symptoms and disease characteristics that do not require the dispatch of an ambulance are of ordinary quality.

5. The artificial intelligence emergency call handling system according to claim 4, characterized in that: The disease matching module establishes a disease model. The specific steps are as follows: Data collection: Data sets containing disease characteristics were collected from clinical records, medical records, and laboratory test results; Feature selection: Select features related to disease diagnosis in the data set, including symptoms, signs, and laboratory indicators; Model building: Building models using machine learning methods; Performance evaluation: Evaluate the accuracy, sensitivity, and specificity of the model; Model validation: Verify the generalization ability of the model on different data sets to ensure the applicability of the model; Among them, the disease characteristics are matched with the disease model, and the disease range corresponding to the disease characteristics is determined through the disease model.

Images