WO2019242155A1 - Procédé et appareil de gestion de la santé basés sur la reconnaissance vocale et dispositif informatique - Google Patents

Procédé et appareil de gestion de la santé basés sur la reconnaissance vocale et dispositif informatique Download PDF

Info

Publication number
WO2019242155A1
WO2019242155A1 PCT/CN2018/108388 CN2018108388W WO2019242155A1 WO 2019242155 A1 WO2019242155 A1 WO 2019242155A1 CN 2018108388 W CN2018108388 W CN 2018108388W WO 2019242155 A1 WO2019242155 A1 WO 2019242155A1
Authority
WO
WIPO (PCT)
Prior art keywords
health
sound
data
target
database
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2018/108388
Other languages
English (en)
Chinese (zh)
Inventor
王健宗
黄章成
蔡元哲
肖京
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2019242155A1 publication Critical patent/WO2019242155A1/fr
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/48—Other medical applications
    • A61B5/4803—Speech analysis specially adapted for diagnostic purposes
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L17/00—Speaker identification or verification techniques
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L17/00—Speaker identification or verification techniques
    • G10L17/26—Recognition of special voice characteristics, e.g. for use in lie detectors; Recognition of animal voices

Definitions

  • the present application relates to the technical field of health management, and in particular, to a health management method, device, computer device, and storage medium based on voice recognition.
  • the main purpose of this application is to provide a sound recognition-based health management method, device, computer equipment and storage medium with low cost and wide promotion.
  • the present application proposes a health management method based on voice recognition, including: acquiring voice information;
  • the health database includes multiple sound features and health status data corresponding to each sound feature
  • the health status data corresponding to the sound characteristics in the health database is a prediction result obtained by predicting the sound characteristics through a health status model;
  • the health status model is a model trained on sound characteristics of multiple known health status data;
  • a matching result is obtained, and the health state of the target vocal creature corresponding to the sound information is output according to the matching result.
  • This application also provides a health management device based on voice recognition, including:
  • An extraction unit configured to extract a target sound feature of the sound information
  • a matching unit configured to input the target sound feature into a preset health database for matching, wherein the health database includes multiple sound features and health status data corresponding to each sound feature;
  • the health state data corresponding to the sound feature is a prediction result obtained by predicting the sound feature through a health state model;
  • the health state model is a model obtained by training sound features of multiple known health state data;
  • An output unit is configured to obtain a matching result, and output the health state of the target vocal creature corresponding to the sound information according to the matching result.
  • the present application further provides a computer device including a memory and a processor, where the memory stores computer-readable instructions, and is characterized in that, when the processor executes the computer-readable instructions, implements the steps of the foregoing method.
  • the present application also provides a computer non-volatile readable storage medium on which computer-readable instructions are stored, and the computer-readable instructions implement the steps of the foregoing method when executed by a processor.
  • FIG. 1 is a schematic diagram of steps in a health management method based on voice recognition in an embodiment of the present application
  • FIG. 2 is a schematic diagram of steps of a health management method based on voice recognition in another embodiment of the present application
  • FIG. 3 is a schematic block diagram of a structure of a health management device based on voice recognition in an embodiment of the present application
  • FIG. 4 is a schematic block diagram of a structure of a health management device based on voice recognition in another embodiment of the present application.
  • FIG. 5 is a schematic block diagram of a structure of a health management device based on voice recognition in another embodiment of the present application.
  • FIG. 6 is a schematic block diagram of a structure of a health management device based on voice recognition in another embodiment of the present application.
  • FIG. 7 is a schematic block diagram of a structure of a matching unit in an embodiment of the present application.
  • FIG. 8 is a schematic block diagram of a structure of a computer device according to an embodiment of the present application.
  • a voice recognition-based health management method in this embodiment includes:
  • Step S1 Acquire sound information
  • Step S2 extract a target sound feature of the sound information
  • Step S3 input the target sound feature into a preset health database for matching, wherein the health database includes multiple sound features and health status data corresponding to each sound feature; the sound features in the health database
  • the corresponding state of health data is a prediction result obtained by predicting the sound characteristics through a state of health model; the state of health model is a model trained by the sound characteristics of multiple known state of health data;
  • Step S4 Obtain a matching result, and output the health status of the target vocal creature corresponding to the sound information according to the matching result.
  • step S1 the health management method based on sound recognition in this embodiment needs to obtain the sound information to be measured.
  • the sound can be collected by a sound collector.
  • the above sound can be collected. Multiple sound collectors are placed in the active area of the owner, and multiple sound collectors are placed in different positions within the active area. In this way, the sound can be obtained without wearing it on the subject like a traditional data acquisition instrument, which avoids discomfort to the sound owner and widens the promotion surface.
  • the above sound collector includes a microphone array. After the sound is obtained through the sound collector, noise reduction processing is performed on the sound to obtain sound information that can extract features.
  • step S2 before detecting and matching the sound, it is necessary to perform sound feature extraction on the sound information, and the extracted feature is recorded as the target sound feature.
  • the target sound feature includes time domain feature parameters and frequency domain feature parameters. Among them, the time domain The characteristic parameters include short-term average energy, short-term average amplitude, short-term average zero-crossing rate, formant, pitch frequency, etc.
  • the frequency-domain characteristic parameters include linear prediction coefficients, linear prediction cepstrum coefficients, and Mel frequency cepstrum coefficients.
  • the above formant reflects the characteristics of channel response
  • the pitch frequency reflects the characteristics of glottic excitation
  • the linear prediction coefficient and linear prediction cepstrum coefficient simultaneously reflect the characteristics of glottal excitation and channel response
  • the Mel frequency cepstrum coefficient simulates human ear hearing Characteristics
  • the target sound feature is obtained by extracting the sound information to be measured, and then inputted into a preset health database for matching.
  • the health database includes multiple sound features and the health corresponding to each sound feature. State data.
  • the corresponding health state data can be matched.
  • the health state data in the health database is a prediction result obtained by predicting the health feature model from the sound feature.
  • the health state model is a model trained by sound characteristics of multiple known health state data.
  • the health state model is obtained by training a specified sample set through a Hidden Markov Model (HMM).
  • the sample set includes sound characteristics of known health status data and health status data corresponding to one-to-one sound characteristics.
  • the health status data includes complete health data, a variety of different disease data, and a variety of different sub-health data.
  • step S4 the target sound feature is input into the health database to obtain a matching result, and according to the matching result, the health state of the target vocal creature corresponding to the sound information can be output.
  • health status includes complete health status, many different sub-health statuses, and many different disease states. For example, when a target sound feature corresponding to a patient with heart disease is input to the health database for matching, the target sound feature will be matched with the heart disease data of the health database, and the matching result will be output to output the sound information corresponding Of target vocal creatures have heart disease.
  • the method for health management based on voice recognition after step S1 of acquiring the voice information, includes:
  • Step S10 extracting a voiceprint feature of the sound information
  • Step S11 inputting the voiceprint feature into a voiceprint recognition model to obtain the identity of a target voiced creature corresponding to the voice information;
  • Step S12 combining the identity of the target vocal creature and the health state of the target vocal creature, and outputting a combination result.
  • step S10 before analyzing the identity of the vocalizer corresponding to the sound information according to the characteristics of the voiceprint, the obtained sound is first subjected to noise reduction processing to obtain sound information, the sound information is divided into frames, and then the sound information divided into frames is divided into frames.
  • Perform MFCC Mel-Frequency, Cepstral, Coefficients
  • the specific extraction process is to first convert the sound information segmented into frames into Mel frequencies, then perform cepstrum analysis, and finally extract the voiceprint features.
  • step S11 the voiceprint feature and the identity of the target vocal creature corresponding to the voiceprint sign are first trained to obtain a voiceprint recognition model.
  • the voiceprint feature can be input into the voiceprint recognition model for matching, and the identity of the target vocal creature of the voiceprint feature can be matched.
  • the identification of the target vocal creature is because the target vocal creature of the above-mentioned sound information needs to be determined. Since the voiceprint characteristics of each creature after MFCC feature extraction are different and unique, it can avoid obtaining the sound The result of a false match between a creature and its health.
  • the training voiceprint recognition model is as follows: a large number of voiceprint features and the identities of the vocal creatures corresponding to the voiceprint features are taken as samples.
  • the voiceprint features are obtained by extracting MFCC features from the noise-reduced sound, and then The features of the voiceprint and the target voiced biological identity input corresponding to the feature are preset and trained based on the voiceprint recognition neural network to obtain a voiceprint recognition model.
  • step S21 after identifying the identity of the target vocal creature and obtaining the health status of the target vocal creature, the identity and health status of the target vocal creature can be combined, and then a combination result is output, such as outputting Zhang San's complete health Li Si suffers from heart disease.
  • the above method can be applied to humans or animals.
  • the insured is a pet dog
  • the sound of the pet dog is collected before the insurance is used to train the voiceprint recognition model. If the owner of the pet dog applies If the pet dog is not sick, you can identify the voiceprint feature of the pet dog and enter the corresponding target voice feature into the health database to match the health status of the pet dog to ensure that the pet dog is insured.
  • the object is consistent with the sounding creature corresponding to the health state obtained by using the above-mentioned sound information, thereby preventing the owner of the pet dog from deceiving insurance through various means to indicate that the pet dog is ill; similarly, when the person being insured is insured The voice of the insured was previously collected and used to train the voiceprint recognition model.
  • the voiceprint characteristics of the insured can be identified.
  • matching the health status of the insured by entering the corresponding target sound characteristics into the health database can ensure the insured Corresponding to the use of the health state of the sound information obtained by the sound of people the same, so as to prevent the insured by all means show Pibao Ren success illness potentially fraudulent.
  • A needs to monitor his physical health status in real time in order to receive timely treatment or recuperation when he is ill.
  • a sound collector such as a hall or bedroom can be placed in the activity area. The sound collector can collect A The voice of daily life does not need to speak to the sound collector deliberately.
  • the sound collector collects the sound of A, the sound is transmitted to the system, and then the feature is extracted to obtain the voiceprint feature, and the voiceprint feature Enter the voiceprint recognition model to perform matching to obtain A's identity, and enter A's target voice characteristics into the health database for matching to obtain A's physical health status.
  • combine A's identity and health status, and combine the above The combined result is sent to A's mobile phone, so that A can know his health status in real time. Allows the person under test to monitor their own health while avoiding discomfort such as traditional data collection by wearing devices.
  • the preset health database includes a plurality of health databases, each of which corresponds to a biological species one by one, and the target sound characteristics are input to Before the matching step S3 in the preset health database, the method includes:
  • Step S30 ' determine the biological type of the target vocal creature corresponding to the sound information according to the voiceprint characteristics
  • Step S31 ' calling the health check database corresponding to the biological species corresponding to the sound information.
  • step S30 ' both humans and different kinds of animals can make sounds, and they may suffer from diseases due to physiological reasons. Therefore, the above-mentioned sound recognition-based health management method can be applied to humans and animals.
  • different detection systems are used to implement the above methods.
  • the data used to train the health status models of different species are different, and their predicted health databases are also different. Therefore, the same detection system cannot be used between different species, so according to the type of vocal creatures, The above sound information is assigned to the corresponding detection system so as to proceed to the next step.
  • each health database corresponds one-to-one with the above detection system, that is, one-to-one correspondence with biological species.
  • the biological type of the target vocal creature corresponding to the sound information may be determined according to the voiceprint feature, and the species database may be used for matching.
  • the voiceprint feature is the voiceprint feature of an animal dog
  • the voiceprint feature is compared with that in the species database.
  • the data of animal dogs can be matched to determine that the vocal creature of the voiceprint feature is an animal dog.
  • the target sound features extracted above can be assigned to a dog detection system.
  • the species database can be generated by training on a neural network model, where The training data includes the voiceprint characteristics of different creatures and the types of creatures corresponding to the voiceprint characteristics.
  • step S31 ' when determining the biological type of the target vocal creature corresponding to the sound information based on the voiceprint characteristics, assign the sound characteristic to the corresponding detection system, and then call the health database corresponding to the biological type corresponding to the sound information, Because the data used to train the health status models of different species is different, the predicted health database is also different, that is, each biological species corresponds to a corresponding health database. For example, when judging that the starting sound creature is an animal dog, The sound feature is assigned to the dog detection system, and the animal dog health database is called. The animal dog health database is obtained by training the animal's sound feature and the health state data corresponding to the sound feature into a hidden Markov model.
  • the above method can be applied to a farm, such as placing sound collectors at different spatial positions in the farm, collecting sounds of various animals in the farm through the sound collector, and processing it to pass
  • the above method results in the health status of different animals: healthy or suffering from a certain disease. This greatly reduces the cost of breeding, timely understands the health status of the animals on the farm, and prevents the occurrence of plague.
  • the health state data includes disease data, sub-health data, and complete health data; and the step of inputting the target voice characteristics into a preset health database for matching S3, including:
  • Step S30 determine whether the target sound feature matches the disease data of the health database
  • Step S31 if the target sound feature matches the disease data, determine that the target vocal creature corresponding to the sound information is a disease state; if the sound target feature does not match the disease data, determine the target Whether the target sound characteristics match the sub-health data of the health database;
  • Step S32 if the target sound feature matches the sub-health data, determine that the target vocal creature corresponding to the sound information is a sub-health state; if the target sound feature does not match the sub-health data of the health database , It is determined whether the target sound characteristics match the complete health data of the health database;
  • Step S33 If the target sound characteristics match the complete health data, determine that the target vocal creature corresponding to the sound information is in a completely healthy state.
  • the health database includes a plurality of sound characteristics and health status data corresponding to each of the sound characteristics, and the health status data includes disease data and health data, where due to the diversity of diseases, data for each disease Are different, so the disease data includes many different disease data; because health includes sub-health and complete health, that is, the above health data includes a variety of different sub-health data and complete health data, and a variety of different sub-health data and The complete health data has commonality, that is, the sub-health data corresponding to the sound characteristics and the complete health data have some of the same data. For convenience of expression, the above-mentioned part of the same data is called common data.
  • the target sound characteristics are compared with various disease data of the health database, and then it is determined whether they match.
  • step S31 if the target sound feature matches one of the disease data in the health database, it can be determined that the target vocal creature corresponding to the target sound feature is a disease state and has the disease, and the voiceprint can be combined with the above
  • the output result of the target vocal biometric identified by the recognition model such as the name of the owner of the sound and the type of disease it has. If it is determined that the target sound feature does not match all the disease data of the health database, it is further determined whether the target sound feature matches the sub-health data of the health database.
  • step S32 if the target sound feature matches one of the sub-health data of the health detection model, the target vocal creature corresponding to the sound information may be determined to be in a sub-health state.
  • Voice feature extraction If the health state data corresponding to the target voice feature is sub-health data with sleep disturbances, when matching is performed in the health database, it will match the sleep disorder data in the health state of the health database. The result is that B has a subhealthy sleep disorder. When the above sound characteristics do not match the sub-health data of the health database, it is determined whether the target sound characteristics match the complete health data of the health database.
  • step S33 when the target sound feature matches the complete health data of the health database, it indicates that the health status data corresponding to the sound feature is complete health data, and it is determined that the target vocal creature of the sound feature is completely healthy and free of disease. There is no sub-health in the disease. At this time, the health state of the target vocal creature corresponding to the sound information is output, that is, the state of complete health and no disease.
  • step S32 of determining whether the target sound feature matches the complete health data of the health database the method includes:
  • Step S34 If the target sound feature does not match the complete health data, then input the target sound feature into the health state model to obtain health state data corresponding to the target sound feature, and pass the target sound Features and corresponding health status data are added to the health database.
  • the target sound feature does not match the complete health data
  • the health state of the vocal creature corresponding to the target sound feature in reality is obtained, and the health state data corresponds to the target sound feature, and the target sound feature is obtained.
  • the corresponding health state data is input into the health state model, thereby predicting the health state data corresponding to the target sound feature that can be added to the health database, and finally adding the target sound feature and corresponding health state data to the health database.
  • health data includes a variety of different sub-health data and complete health data, and a variety of different sub-health data and complete health data have common data, and disease data has uncertainty, full health data and sub-health
  • the common data of health data is deterministic. Due to the variety of diseases and the great difference, each disease data is different, so the disease data is uncertain; when everyone is not sick, and when they are completely healthy, At this time, the body shows only one state, so the complete health data is deterministic, and sub-health also belongs to the category of physical health.
  • the sub-health data has the same data as the full health data part, and the same data in this part That is, the above common data must also be deterministic, and since each person's health is determined by the state of the body, the common data can be used to determine whether the health state data corresponding to the target sound characteristics belong to the health data. Therefore, if the above target sound characteristics do not match all the disease data and sub-health data of the health database, and at this time the above target sound characteristics do not match the safety and health data of the health database, then it is necessary to determine the health status corresponding to the target sound characteristics. Whether part of the data matches the above common data.
  • the health status data is a new type of sub-health data That is, the body of the vocal creature of the target sound characteristic is in a healthy state without disease and in a new sub-health state. Due to the limited number of samples for training the health state model, there is no such data in the health database. At this time, the above-mentioned sound features are transmitted to the background system, the health state model is trained, and the predicted sound features and the corresponding new sub-health The data is added to the above health database to avoid misjudgment caused by the previous health database not matching all the physical conditions.
  • the common data is determined, that is, the sound health state of the vocal creature is determined. If it does not match, it indicates that The health state data corresponding to the target sound characteristics is new type of disease data. Because the samples of the training health state model are limited, there is no such data in the health database. At this time, the sound characteristics are transmitted to the background system for training.
  • the above-mentioned health state model adds the obtained voice characteristics and corresponding new disease data to the above-mentioned health database, so as to avoid misjudgment caused by the previous health database not matching the new disease.
  • the health management method based on voice recognition in this embodiment after step S4 of outputting the health state of the target vocal creature corresponding to the voice information according to the matching result, includes:
  • Step S5 Score the physical health level of the target vocal creature according to the health state, and match a recuperation suggestion corresponding to the physical health level.
  • step S5 the system sets a table of recuperation suggestions and scores that are paired with each of the sub-health types and disease types in the health database.
  • the health status of the target vocal creature corresponding to the sound information is determined according to the matching result, it is different.
  • the health status indicates that the vocal creatures have different physical health levels.
  • the system looks for corresponding recuperation suggestions and scores in the table according to the above physical health levels. For example, when it is judged to be completely healthy, it is recommended to maintain the status quo, with a score of 100; The result is sub-health of fatigue and excessive sleep disturbance.
  • the corresponding recuperation recommendations are reasonable diet, moderate exercise, and regular work and rest.
  • the score is 85 points.
  • the system outputs the target vocal creature identity and the above-mentioned sub-healthy type, corresponding recuperation suggestions and scores to a designated place, such as being transmitted to the mailbox of the owner of the above-mentioned voice information, and the score makes the user more intuitively know the physical health status
  • users can recuperate their bodies in a targeted manner by implementing recuperation recommendations to make the body healthier, and the method is convenient, intuitive and convenient.
  • the method for health management based on voice recognition further includes the following steps: obtaining target voice characteristics, and The target sound feature is input to the disease database for matching. If the target sound feature matches a certain disease data in the disease database, the target vocal creature of the target sound feature may be determined to have a certain type according to the matching result.
  • the target sound feature if the target sound feature does not match the disease data in the disease database, the target sound feature is entered into the health database to match, and if the target sound feature matches the common data in the health database, then the matching can be based on the match As a result, it was determined that the target vocal creature of the sound characteristic was in a healthy state not to be ill. Further, if the target sound feature does not match the disease data of the disease database or the common data of the health database, it is determined that the health state data corresponding to the target sound feature is new disease data. The new disease data and corresponding sound features are used to train the disease detection model again, and the sound features and corresponding new disease data are added to the disease database. When the above target sound features match the common data in the health database, they are not in complete health.
  • the vocal creature of the sound feature is judged to be sub-healthy. Further, when the above target sound feature matches the common data in the health database, but does not match the complete health data and all the sub-health data , It is determined that the health state data corresponding to the target sound feature is a new type of sub-health data, and the health detection model can be trained again through the new type of sub-health data and the corresponding sound feature, and the sound feature and the corresponding new sub-health data are added to the health database.
  • the above disease database and health database can be obtained by referring to the above-mentioned health database generation process. Specifically, sound characteristics and disease data corresponding to the sound characteristics can be input into a preset HMM model for training to obtain a disease detection model to predict Obtain a disease database; input sound features and health data corresponding to the sound features into a preset HMM model for training, obtain a health detection model, and predict a health database.
  • the method further includes:
  • sample data Acquiring a specified amount of sample data, and dividing the sample data into a training set and a test set, wherein the sample data includes extracted sound features and health status data corresponding to the sound features;
  • the result training model is recorded as the health state model.
  • the health database used to match the corresponding physical health state can be predicted only after the training is completed.
  • a large amount of sample data needs to be obtained, and the sample data is divided into a training set and a test set, where the sample data includes the extracted sound features and the corresponding ones of the extracted sound features.
  • Health data The sample data of the training set is input into a preset hidden Markov model for training, and a training model is obtained as a result of performing health detection.
  • the voice recognition-based health management device in this embodiment includes:
  • An obtaining unit 100 configured to obtain sound information
  • An extraction unit 200 configured to extract a target sound feature of the sound information
  • the matching unit 300 is configured to input the target sound feature into a preset health database for matching, wherein the health database includes multiple sound features and health status data corresponding to each sound feature respectively; in the health database
  • the health state data corresponding to the sound feature is a prediction result obtained by predicting the sound feature by a health state model; the health state model is a model trained by sound features of multiple known health state data;
  • the output unit 400 is configured to obtain a matching result, and output the health state of the target vocal creature corresponding to the sound information according to the matching result.
  • the obtaining unit 100 needs to obtain the sound information to be measured.
  • the sound can be collected by a sound collector.
  • the owner of the sound can be collected.
  • Multiple sound collectors are placed in the active area of, and multiple sound collectors are placed in different positions within the active area. In this way, the sound can be obtained without wearing it on the subject like a traditional data acquisition instrument, which avoids discomfort to the sound owner and widens the promotion surface.
  • the above sound collector includes a microphone array. After obtaining the sound through the sound collector, the obtaining unit 100 performs noise reduction processing on the sound to obtain sound information that can extract features.
  • the extraction unit 200 Before detecting and matching the sound, the extraction unit 200 performs sound feature extraction on the sound information, and the extracted features are recorded as target sound features.
  • the target sound features include time domain feature parameters and frequency domain feature parameters. Among them, the time domain feature parameters. Including short-term average energy, short-term average amplitude, short-term average zero-crossing rate, formant, pitch frequency, etc., the frequency domain characteristic parameters include linear prediction coefficient, linear prediction cepstrum coefficient, Mel frequency cepstrum coefficient and so on.
  • the above formant reflects the characteristics of channel response
  • the pitch frequency reflects the characteristics of glottic excitation
  • the linear prediction coefficient and linear prediction cepstrum coefficient simultaneously reflect the characteristics of glottal excitation and channel response
  • the Mel frequency cepstrum coefficient simulates human ear hearing Characteristics
  • the matching unit 300 obtains the target sound feature after extracting the sound information to be measured, and then inputs it into a preset health database for matching.
  • the health database includes multiple sound features and health status data corresponding to each sound feature.
  • the corresponding health status data can be matched.
  • the health status data of the health health database is a prediction result obtained by predicting the health feature model from the above sound feature, in which health
  • the state model is a model trained by sound characteristics of multiple known health state data.
  • the above health state model is obtained by training a specified sample set through a Hidden Markov Model (HMM).
  • HMM Hidden Markov Model
  • the specified sample set It includes sound characteristics of known health status data, and health status data corresponding to the sound characteristics.
  • the health status data includes complete health data, a variety of different disease data, and a variety of different sub-health data.
  • the target sound feature is input into the health database after successful training to obtain a matching result, and the output unit 400 can output the health state of the target vocal creature corresponding to the sound information according to the matching result.
  • health status includes complete health status, many different sub-health statuses, and many different disease states. For example, when a target sound feature corresponding to a patient with heart disease is input to the health database for matching, the target sound feature will be matched with the heart disease data of the health database, and the matching result will be output to output the sound information corresponding Of vocal creatures have heart disease.
  • a voice recognition-based health management device further includes:
  • a voiceprint unit 500 a voiceprint feature for extracting the voice information
  • An identity unit 510 configured to input the voiceprint feature into a voiceprint recognition model to obtain the identity of a target voiced creature corresponding to the voice information;
  • the combining unit 520 is configured to combine the identity of the target vocal creature and the health state of the target vocal creature, and output a combination result.
  • the obtained sound is first subjected to noise reduction processing to obtain sound information, the above sound information is divided into frames, and then the sound information divided into frames is subjected to MFCC (Mel- Frequency (Cepstral Coefficients) feature extraction.
  • MFCC Mel- Frequency (Cepstral Coefficients) feature extraction.
  • the specific extraction process is to first convert the sound information segmented into frames into Mel frequencies, then perform cepstrum analysis, and finally extract the voiceprint features.
  • the voiceprint feature and the identity of the target vocal creature corresponding to the voiceprint feature are used for training to obtain a voiceprint recognition model.
  • the voiceprint feature can be input into the voiceprint recognition model for matching, and the identity of the vocalizing creature of the voiceprint feature can be matched through the voiceprint recognition library.
  • the matching unit 300 inputs a target sound feature into a health database to perform matching to obtain a matching result, finally, the above-mentioned vocal creature identity is combined, and a final result is output.
  • the identification of the target vocal creature is because the target vocal creature of the above-mentioned sound information needs to be determined. Since the voiceprint characteristics of each creature after MFCC feature extraction are different and unique, it can avoid obtaining the sound The result of a false match between a creature and its health.
  • the training voiceprint recognition model is specifically as follows: a large number of voiceprint features and the identities of the vocal creatures corresponding to the voiceprint features are used as samples, and the voiceprint features are obtained by extracting MFCC features of the denoised sound, and then a large number of The above MFCC feature and the voiceprint recognition neural network preset based on the vocal biometric input corresponding to the feature are trained to obtain a voiceprint recognition model.
  • the combining unit 510 may combine the identity and health status of the target vocal creature, and then output a combination result, such as outputting Zhang San ’s complete body Health, Li Si has heart disease, etc.
  • the above device can be applied to humans or animals.
  • the sound of the pet dog is collected before the insurance is used to train the voiceprint recognition model. If the owner of the pet dog applies If the pet dog is not sick, you can identify the pet dog's voiceprint characteristics and match the health status of the pet dog by entering the corresponding target voice characteristics into the health database to ensure that The insured object is consistent with the sounding creature corresponding to the health state obtained by using the above-mentioned sound information, thereby preventing the owner of the pet dog from successfully deceiving insurance through various means to indicate that the pet dog is sick; similarly, when the insured is a human, Collect the voice of the insured before training, and use it to train the voiceprint recognition model.
  • the identity of the insured's voiceprint can be identified. Recognition, and matching the health status of the insured by entering the corresponding target sound characteristics into the health database to ensure that the insured is insured The object is consistent with the sound person corresponding to the health state obtained by using the above-mentioned sound information, thereby preventing the insured person from successfully deceiving insurance by indicating that the insured person is sick through various means.
  • A needs to monitor his physical health status in real time in order to receive timely treatment or recuperation when he is ill.
  • a sound collector such as a hall or bedroom can be placed in the activity area. The sound collector can collect A The voice of daily life does not need to speak to the sound collector deliberately.
  • the sound collector collects the sound of A, the sound is transmitted to the system, and then the feature is extracted to obtain the voiceprint feature, and the voiceprint feature Enter the voiceprint recognition model to perform matching to obtain A's identity, and enter A's target voice characteristics into the health database for matching to obtain A's physical health status.
  • combine A's identity and health status, and combine the above The combined result is sent to A's mobile phone, so that A can know his health status in real time. Allows the person under test to monitor their own health while avoiding discomfort such as traditional data collection by wearing devices.
  • the preset health database includes a plurality of health databases, each health database corresponds to a biological species one to one, and further includes:
  • a classification unit 600 configured to determine a biological type of a target vocal creature corresponding to the sound information according to the voiceprint characteristics
  • the invoking unit 700 is configured to invoke a health database corresponding to the biological species corresponding to the sound information.
  • the above-mentioned sound recognition-based health management device can be applied to humans and animals.
  • different detection systems are used to implement the above device. It should be pointed out that the data used to train the health status models of different species are different, and their predicted health databases are also different. Therefore, the same detection system cannot be used between different species, so the above-mentioned sound information is allocated according to the type of sounding organisms Go to the corresponding detection system for the next step.
  • the classification unit 600 determines the biological type of the vocal creature corresponding to the sound information according to the voiceprint feature, and may use a species database to match.
  • the voiceprint feature is the voiceprint feature of an animal dog
  • the voiceprint feature and the species The animal dog data in the database can be matched to determine the vocal creature of the voiceprint feature as an animal dog.
  • the target sound feature extracted above can be assigned to a dog detection system, and the species database can be generated through neural network model training.
  • the training data includes the voiceprint characteristics of different creatures and the types of creatures corresponding to the voiceprint characteristics.
  • the classification unit 600 determines the biological type of the target vocal creature corresponding to the sound information according to the voiceprint characteristics, assign the sound characteristic to the corresponding detection system, and then call the unit 700 to call the health database corresponding to the biological type corresponding to the sound information.
  • the predicted health database is also different, that is, each biological species corresponds to a corresponding health database. For example, when judging that the starting sound creature is an animal dog, The sound feature is assigned to the dog detection system, thereby calling the animal dog's health database, wherein the animal dog's health database is obtained by training the animal dog's sound feature and the health state data corresponding to the sound feature into a hidden Markov model.
  • the above device can be applied to a farm, such as placing sound collectors at different spatial positions in the farm, and collecting sounds of various animals in the farm through the sound collector.
  • the above method results in the health status of different animals: healthy or suffering from a certain disease. This greatly reduces the cost of breeding, timely understands the health status of the animals on the farm, and prevents the occurrence of plague.
  • the health status data includes disease data, sub-health data, and complete health data;
  • the matching unit 300 includes:
  • a first determining subunit 310 configured to determine whether the target sound feature matches the disease data of the health database
  • the second judging subunit 320 is configured to determine that the target vocal creature corresponding to the sound information is a disease state when the target sound feature matches the disease data; the target sound feature is not in accordance with the disease data When matching, determining whether the target sound feature matches the sub-health data of the health database;
  • the third judging subunit 330 is configured to determine that the target vocal creature corresponding to the sound information is in a sub-health state when the target sound feature matches the sub-health data; the sound feature and the sub-health database When the health data does not match, it is determined whether the target sound feature matches the complete health data of the health database;
  • An output subunit 340 is configured to determine that the target sound feature matches the complete health data, and then determine that the target vocal creature corresponding to the sound information is in a completely healthy state.
  • the health database includes a plurality of sound characteristics and health condition data corresponding to each of the sound characteristics.
  • the health status data includes disease data and health data. Because of the diversity of diseases, the data for each disease is different, so Disease data includes a variety of different disease data; because health includes sub-health and complete health, that is, the above health data includes a variety of different sub-health data and complete health data, and a variety of different sub-health data and complete health data have common characteristics That is, the sub-health data and the complete health data corresponding to the sound characteristics have partially the same data, and for convenience of expression, the above-mentioned partially the same data are called common data.
  • the extracted sound features are input into the health database, and the first determination sub-unit 310 compares the sound target features with various disease data of the health database, and then determines whether they match.
  • the target sound feature is consistent with one of the disease data in the health database, it can be determined that the target vocal creature corresponding to the target sound feature is a disease state and has the disease, which can be combined with the above identified by the voiceprint recognition model
  • the output result of the target's vocal identity such as the name of the owner of the sound and the type of disease it has. If it is determined that the target sound feature does not match all the disease data of the health database, the second determination sub-unit 320 further determines whether the target sound feature matches the sub-health data of the health database.
  • the target sound feature matches one of the sub-health data of the health detection model, it can be determined that the target vocal creature corresponding to the sound information is in a sub-health state. For example, the sound of B is collected, and the feature of B's sound is extracted. If the health state data corresponding to the extracted target sound feature is sub-healthy data with sleep disturbance, when the target sound feature is matched in the health database, it is matched with the sleep disorder data of the health state in the health database. At this time, the result that B has a subhealthy sleep disorder can be output. When the target sound feature does not match the sub-health data of the health database, the third determining sub-unit 330 determines whether the target sound feature matches the safety and health data of the health database.
  • the output subunit 340 outputs the health state of the target vocal creature corresponding to the sound information, that is, the state of complete health and no disease.
  • the matching unit 300 further includes:
  • Adding a sub-unit 350 for determining that the target sound feature does not match the complete health data then inputting the target sound feature into the health state model to obtain health state data corresponding to the target sound feature, and Adding the target sound feature and corresponding health status data to the health database.
  • the health state of the vocal creature corresponding to the target sound feature in reality is obtained, and the health state data corresponds to the target sound feature, and the target sound feature is obtained.
  • the health state data is input into the health state model, thereby predicting the health state data corresponding to the target sound feature that can be added to the health database, and finally adding the target sound feature and the corresponding health state data to the health database.
  • health data includes a variety of different sub-health data and complete health data, and a variety of different sub-health data and complete health data have common data, and disease data has uncertainty, full health data and sub-health
  • the common data of health data is deterministic. Due to the variety of diseases and the great difference, each disease data is different, so the disease data is uncertain; when everyone is not sick, and when they are completely healthy, At this time, the body shows only one state, so the complete health data is deterministic, and sub-health also belongs to the category of physical health.
  • the sub-health data has the same data as the full health data part, and the same data in this part That is, the above common data must also be deterministic, and since each person's health is determined by the state of the body, the common data can be used to determine whether the health state data corresponding to the target sound characteristics belong to the health data. Therefore, if the above target sound characteristics do not match all the disease data and sub-health data of the health database, and at this time the above target sound characteristics do not match the safety and health data of the health database, then it is necessary to determine the health status corresponding to the target sound characteristics. Whether some of the data matches the above common data.
  • the adding sub-unit 350 sends the sound features to the background system, trains the health state model, and predicts the sound features and Corresponding new sub-health data is added to the above health database to avoid misjudgment caused by the previous health database not matching all the physical conditions.
  • the common data is determined, that is, the sound health status of the vocal creature is determined.
  • the health status data corresponding to the target sound feature is new-type disease data. Because the sample of the training health detection model is limited, there is no such data in the health database.
  • the adding subunit 350 sends the sound feature to the background system.
  • the above health state model is trained, and the predicted sound characteristics and corresponding new disease data are added to the above health database, so as to avoid misjudgment caused by the previous health database not matching the new disease.
  • the voice recognition-based health management device in this embodiment further includes:
  • the scoring unit 800 is configured to score the physical health level of the target vocal creature according to the health state, and match a recuperation suggestion corresponding to the physical health level.
  • the system is provided with a table of recuperation suggestions and scores that are paired with each sub-health type and disease type in the health database.
  • the scoring unit 800 searches the table for corresponding recuperation recommendations and scores based on the above-mentioned physical health. For the judgment of complete health, it is recommended to maintain the status quo, with a score of 100; for judgment, fatigue
  • the corresponding recuperation recommendations are reasonable diet, moderate exercise, and regular work and rest, and its score is 85 points.
  • the system outputs the vocal biological identity of the target sound feature and the above-mentioned sub-healthy type, corresponding recuperation suggestions and scores to a designated place, such as being transmitted to the mailbox of the owner of the above-mentioned sound information, and the score makes the user more intuitively know the physical The health level, and at the same time, the user can recuperate the body in a targeted manner by implementing recuperation recommendations to make the body healthier, and the method is convenient, intuitive and convenient.
  • the method for health management based on voice recognition further includes the following steps: obtaining the target voice characteristics, first The target sound feature is input into the disease database for matching. If the target sound feature matches a certain disease data in the disease database, the target vocal creature of the target sound feature may be determined to have a certain disease according to the matching result; If the target sound feature does not match the disease data in the disease database, the target sound feature is entered into the health database for matching. If the target sound feature matches the common data in the health database, then the matching result can be determined The target vocal creature of the acoustic feature is in a healthy state that is not diseased.
  • the target sound feature does not match the disease data of the disease database or the common data of the health database, it is determined that the health state data corresponding to the target sound feature is new disease data.
  • the new disease data and corresponding sound features are used to train the disease detection model again, and the target sound feature and corresponding new disease data are added to the disease database.
  • the above target sound feature matches the common data in the health database, it does not match the complete data.
  • the vocal creature of the sound feature is judged to be in a sub-health state.
  • the health detection model can be trained again by using the above-mentioned new sub-health data and corresponding sound feature, and adding the target sound feature and corresponding new sub-health data To the health database.
  • the above disease database and health database can be obtained by referring to the above-mentioned health database generation process. Specifically, sound characteristics and disease data corresponding to the sound characteristics can be input into a preset HMM model for training to obtain a disease detection model to predict Obtain a disease database; input sound features and health data corresponding to the sound features into a preset HMM model for training, obtain a health detection model, and predict a health database.
  • the method further includes:
  • sample data Acquiring a specified amount of sample data, and dividing the sample data into a training set and a test set, wherein the sample data includes extracted sound features and health status data corresponding to the sound features;
  • the result training model is recorded as the health state model.
  • the health database used to match the corresponding physical health state can be predicted only after the training is completed.
  • a large amount of sample data needs to be obtained, and the sample data is divided into a training set and a test set, where the sample data includes extracted sound features and corresponds to the extracted sound features.
  • Health data The sample data of the training set is input into a preset hidden Markov model for training, and a training model is obtained as a result of performing health detection.
  • an embodiment of the present application further provides a computer device.
  • the computer device may be a server, and its internal structure may be as shown in FIG.
  • the computer device includes a processor, a memory, a network interface, and a database connected through a system bus.
  • the computer design processor is used to provide computing and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system, computer-readable instructions, and a database.
  • the memory provides an environment for operating systems and computer-readable instructions in a non-volatile storage medium.
  • the database of the computer equipment is used for data such as a preset health state model.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer-readable instructions are executed by a processor to implement a health management method based on voice recognition.
  • the processor executes the steps of the sound management method based on sound recognition: acquiring sound information; extracting a target sound feature of the sound information; inputting the target sound feature into a preset health database for matching, wherein, the The health database includes multiple sound features and health status data corresponding to each sound feature; the health status data corresponding to the sound features in the health database is a prediction result obtained by predicting the sound feature through a health status model; the health status The model is a model trained on sound characteristics of multiple known health status data; a matching result is obtained, and the health status of the target vocal creature corresponding to the sound information is determined according to the matching result.
  • the above computer equipment is trained based on a hidden Markov model to obtain a health state model.
  • the target sound features are obtained.
  • the sound features are input into a preset health database for matching, and the matching result can be obtained according to The matching result determines the health status of the vocal creature corresponding to the sound information.
  • the method after acquiring the sound information, includes: extracting a voiceprint feature of the sound information; and inputting the voiceprint feature through a voiceprint recognition model to obtain a target vocal creature corresponding to the sound information. The identity of the target vocal creature and the health status of the target vocal creature are combined, and a combination result is output.
  • the preset health databases include multiple ones, each health database corresponds to a biological species one by one, and before the inputting the target sound feature to the preset health database for matching, the method includes: The voiceprint feature judges the biological type of the target vocal creature corresponding to the sound information; and invokes a health database corresponding to the biological type corresponding to the sound information.
  • the health status data includes disease data, sub-health data, and complete health data.
  • the above-mentioned step of inputting the sound characteristics into a preset health database for matching includes: judging the target sound characteristics and the above-mentioned sound characteristics.
  • the disease data of the health database matches; if the target sound feature matches the disease data, it is determined that the target vocal creature corresponding to the sound information is a disease state; if the target sound feature does not match the disease data , Determine whether the target sound feature matches the sub-health data of the health database; if the target sound feature matches the sub-health data, determine that the target vocal creature corresponding to the sound information is in a sub-health state; If the target sound feature does not match the sub-health data of the health database, determine whether the target sound feature matches the complete health data of the health database; if the target sound feature matches the complete health data, determine the above The above target corresponding to the sound information The sound is completely healthy biological state.
  • the method includes: if the target sound feature does not match the complete health data, inputting the target sound feature into the health State model to obtain the health status data corresponding to the target sound feature, and add the target sound feature and corresponding health status data to the health database, to avoid misjudgment caused by the failure of the previous health database to match the new sound feature .
  • the method includes: scoring the physical health of the target vocalizer according to the health status, and By matching the recuperation suggestions corresponding to the physical health level, the user can more intuitively know the health condition of the body.
  • FIG. 8 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied.
  • An embodiment of the present application further provides a computer non-volatile readable storage medium, which stores computer-readable instructions.
  • a health management method based on voice recognition is implemented, specifically: Acquiring sound information; extracting a target sound feature of the sound information; inputting the target sound feature into a preset health database for matching, wherein the health database includes a plurality of sound features and corresponding ones of the sound features Health state data; the health state data corresponding to the sound features in the health database is the prediction result obtained by predicting the sound feature through a health state model; the health state model is obtained by training the sound features of multiple known health state data Model; obtain a matching result, and judge the health status of the target vocal creature corresponding to the sound information according to the matching result.
  • the computer-readable storage medium is based on a hidden Markov model, and is trained to obtain a health state model.
  • the target sound feature is obtained by extracting the characteristics of the sound information to be predicted, and the sound feature is input into a preset health database for matching to obtain a matching result.
  • the health status of the vocal creature corresponding to the sound information may be determined according to the matching result.
  • the method after acquiring the sound information, includes: extracting a voiceprint feature of the sound information; and inputting the voiceprint feature into a voiceprint recognition model to obtain a target vocal creature corresponding to the sound information. Identity; combining the identity of the target vocal creature and the health status of the target vocal creature, and outputting a combination result.
  • the preset health databases include multiple ones, each health database corresponds to a biological species one by one, and before the inputting the target sound feature to the preset health database for matching, the method includes: The voiceprint feature judges the biological type of the target vocal creature corresponding to the sound information; and invokes a health database corresponding to the biological type corresponding to the sound information.
  • the health status data includes disease data, sub-health data, and complete health data.
  • the above-mentioned step of inputting the sound characteristics into a preset health database for matching includes: judging the target sound characteristics and the above-mentioned sound characteristics.
  • the disease data of the health database matches; if the target sound feature matches the disease data, it is determined that the target vocal creature corresponding to the sound information is a disease state; if the target sound feature does not match the disease data , Determine whether the target sound feature matches the sub-health data of the health database; if the target sound feature matches the sub-health data, determine that the target vocal creature corresponding to the sound information is in a sub-health state; If the target sound feature does not match the sub-health data of the health database, determine whether the target sound feature matches the complete health data of the health database; if the target sound feature matches the complete health data, determine the above The above target corresponding to the sound information The vocal creature is fully healthy.
  • the method includes: if the target sound feature does not match the complete health data, inputting the target sound feature into the health state
  • the model is used to obtain the health status data corresponding to the target sound feature, and the target sound feature and the corresponding health status data are added to the health database, so as to avoid misjudgment caused by the failure of the previous health database to match the new sound feature.
  • the method includes: scoring the physical health of the target vocalizer according to the health status, and By matching the recuperation suggestions corresponding to the physical health level, the user can more intuitively know the health condition of the body.
  • Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory can include random access memory (RAM) or external cache memory.
  • RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Biomedical Technology (AREA)
  • Pathology (AREA)
  • Biophysics (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Veterinary Medicine (AREA)
  • Medical Treatment And Welfare Office Work (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

L'invention concerne un procédé et un appareil de gestion de la santé basés sur la reconnaissance vocale, un dispositif informatique et un support de stockage. Le procédé consiste à : acquérir des informations vocales ; extraire une caractéristique vocale cible des informations vocales ; entrer la caractéristique vocale cible dans une base de données de santé prédéfinie pour une mise en correspondance, la base de données de santé comprenant de multiples caractéristiques vocales et des données d'état de santé correspondant respectivement à diverses caractéristiques vocales ; et acquérir un résultat de correspondance et fournir l'état de santé d'un être cible produisant la voix correspondant aux informations vocales en fonction du résultat de correspondance.
PCT/CN2018/108388 2018-06-22 2018-09-28 Procédé et appareil de gestion de la santé basés sur la reconnaissance vocale et dispositif informatique Ceased WO2019242155A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810652310.X 2018-06-22
CN201810652310.XA CN109171644A (zh) 2018-06-22 2018-06-22 基于声音识别的健康管理方法、装置、计算机设备和存储介质

Publications (1)

Publication Number Publication Date
WO2019242155A1 true WO2019242155A1 (fr) 2019-12-26

Family

ID=64948728

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/108388 Ceased WO2019242155A1 (fr) 2018-06-22 2018-09-28 Procédé et appareil de gestion de la santé basés sur la reconnaissance vocale et dispositif informatique

Country Status (2)

Country Link
CN (1) CN109171644A (fr)
WO (1) WO2019242155A1 (fr)

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110136743A (zh) * 2019-04-04 2019-08-16 平安科技（深圳）有限公司 基于声音采集的健康状态监控方法、装置及存储介质
CN110189757A (zh) * 2019-06-27 2019-08-30 电子科技大学 一种大熊猫个体识别方法、设备及计算机可读存储介质
CN110368005A (zh) * 2019-07-25 2019-10-25 深圳大学 一种智能耳机及基于智能耳机的情绪及生理健康监控方法
CN110473616B (zh) * 2019-08-16 2022-08-23 北京声智科技有限公司 一种语音信号处理方法、装置及系统
CN110853294A (zh) * 2019-10-29 2020-02-28 广东美的白色家电技术创新中心有限公司 利用家用电器进行监控的方法、装置和计算机存储介质
CN111210915A (zh) * 2020-04-17 2020-05-29 北京海益同展信息科技有限公司 一种疫情管控方法及系统
CN112509697B (zh) * 2020-12-04 2023-08-22 南京大学 一种多模块智慧健康管理信息系统
EP4493045A4 (fr) * 2022-03-13 2026-03-25 Sonaphi Llc Systèmes et procédé de fourniture d'informations de santé par le biais de l'utilisation de la voix d'une personne
CN116030562A (zh) * 2022-11-17 2023-04-28 北京声智科技有限公司 数据处理方法、装置、设备及介质
CN116439661B (zh) * 2023-03-24 2023-09-29 中国人民解放军总医院第六医学中心 一种围产期的产妇生理状态监测评估方法及系统
CN119650034A (zh) * 2024-11-18 2025-03-18 成都理工大学 用于中医诊断的基于语音分析的嵌入式健康管理系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2005002433A1 (fr) * 2003-06-24 2005-01-13 Johnson & Johnson Consumer Compagnies, Inc. Systeme et procede de formation personnalisee pour la comprehension correcte de la parole humaine au moyen d'une prothese auditive
CN103810374A (zh) * 2013-12-09 2014-05-21 中国矿业大学 一种基于mfcc特征提取的机器故障预测方法
CN106847293A (zh) * 2017-01-19 2017-06-13 内蒙古农业大学 设施养殖羊应激行为的声信号监测方法
CN106941005A (zh) * 2017-02-24 2017-07-11 华南理工大学 一种基于语音声学特征的声带异常检测方法

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9055861B2 (en) * 2011-02-28 2015-06-16 Samsung Electronics Co., Ltd. Apparatus and method of diagnosing health by using voice
CN102222500A (zh) * 2011-05-11 2011-10-19 北京航空航天大学 结合情感点的汉语语音情感提取及建模方法
CN105895105B (zh) * 2016-06-06 2020-05-05 北京云知声信息技术有限公司 语音处理方法及装置
US10276187B2 (en) * 2016-10-19 2019-04-30 Ford Global Technologies, Llc Vehicle ambient audio classification via neural network machine learning
CN106725329A (zh) * 2016-12-30 2017-05-31 天津森宇科技发展有限公司 根据发音判别人体健康状态的智能识别系统
CN107704549A (zh) * 2017-09-26 2018-02-16 百度在线网络技术（北京）有限公司 语音搜索方法、装置及计算机设备

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2005002433A1 (fr) * 2003-06-24 2005-01-13 Johnson & Johnson Consumer Compagnies, Inc. Systeme et procede de formation personnalisee pour la comprehension correcte de la parole humaine au moyen d'une prothese auditive
CN103810374A (zh) * 2013-12-09 2014-05-21 中国矿业大学 一种基于mfcc特征提取的机器故障预测方法
CN106847293A (zh) * 2017-01-19 2017-06-13 内蒙古农业大学 设施养殖羊应激行为的声信号监测方法
CN106941005A (zh) * 2017-02-24 2017-07-11 华南理工大学 一种基于语音声学特征的声带异常检测方法

Also Published As

Publication number Publication date
CN109171644A (zh) 2019-01-11

Similar Documents

Publication Publication Date Title
WO2019242155A1 (fr) Procédé et appareil de gestion de la santé basés sur la reconnaissance vocale et dispositif informatique
US12230369B2 (en) Systems and methods for mental health assessment
US12347563B2 (en) System and method for assessing physiological state
CN110516161B (zh) 一种推荐方法及装置
US20210110895A1 (en) Systems and methods for mental health assessment
CN114206361A (zh) 一种用于语音属性的机器学习的系统和方法
Das A comparison of multiple classification methods for diagnosis of Parkinson disease
JP7390268B2 (ja) 認知機能予測装置、認知機能予測方法、プログラム及びシステム
US20260094689A1 (en) Adaptive artificial intelligence system for identifying behaviors associated with mental illness and modifying treatment plans based on emergent recognition of aberrant reactions
CN115955932A (zh) 基于语音特征的阿尔茨海默病预测方法和装置
US20240420725A1 (en) Audio Analytics System And Methods Of Use Thereof
Verde et al. Dysphonia detection index (DDI): A new multi-parametric marker to evaluate voice quality
US20220409089A1 (en) A cough detection system and method
Milani et al. A real-time application to detect human voice disorders
Connaghan et al. A preliminary investigation of acoustic features for remote monitoring of respiration in ALS
Yamada et al. A deep-learning-based method of estimating water intake
CN117373658B (zh) 一种基于数据处理的抑郁症辅助诊疗系统及方法
US20250226066A1 (en) Systems and methods for mental health assessment
Choi et al. Measurement of level of consciousness by AVPU scale assessment system based on automated video and speech recognition technology
IBRAHIM et al. Integration of Machine Intelligence and Internet of Things in Smart Healthcare for Voice Pathology Detection (IoT
Gorli et al. Proposal of ASLR for Voice Disorders
Tripathi et al. Acoustic Analysis for Identifying Fake Cough and Cold Symptoms via AI Models
Cazarin et al. Phoniatric System Based on Acoustical Analysis for Early Detection of Anomalies in Voice Production
Brueckner et al. Audio-Based Detection of Anxiety
Zhang Multimodal attention network for trauma activity recognition from spoken language and environmental sound

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18923030

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18923030

Country of ref document: EP

Kind code of ref document: A1