KR20120098383A - Apparatus and method diagnosing health using voice - Google Patents

Abstract

PURPOSE: A health status diagnosis apparatus and method thereof are provided to automatically diagnose health status using voice in a communication process. CONSTITUTION: A voice sensing unit(110) senses voice by monitoring the voice. A voice analysis unit(120) extracts voice characteristics according to a health status from the voice sensed by the voice sensing unit. A voice diagnosis unit(130) diagnoses the health status by comparing a diseased condition standard with the voice characteristic extracted from the voice analysis unit. The voice diagnosis unit monitors a health change status. A diagnosis output unit(140) outputs signals for the health status and the health change status information. [Reference numerals] (110) Voice sensing unit; (120) Voice analysis unit; (130) Voice diagnosis unit; (140) Diagnosis output unit

Description

Apparatus and method diagnosing health using voice}

The present invention relates to an apparatus and method for diagnosing the state of health of body organs, and more particularly, to an apparatus and method for diagnosing the state of health of body organs using voice.

Recently, there have been increasing cases of complaints of laryngeal pain and voice changes due to air pollution, limited space activities, and increased use of mobile phones. Laryngeal disease can be difficult to communicate through speech utterance, if it is not treated early, and then metastasize to malignant, it is impossible to recover from surgery and restore normal voice. Therefore, early diagnosis of laryngeal disease is a very important factor in preventing metastasis to malignant tumors, preserving the function of the larynx, and increasing the success rate of treatment and the survival rate of patients.

The range of negatively diagnosed diseases is widening, and another example of negative diagnosis is Parkinson's disease. Parkinson's disease is caused by the destruction of the neurotransmitter dopamine-producing cells secreted by the brain regions that control movement, which can be diagnosed with muscle stiffness, tremor, slow motion, and loss of balance. When it is diagnosed as a dopamine-producing cells are already destroyed a lot, to prevent the destruction of dopamine-producing cells to have a high therapeutic effect, early diagnosis is a very important factor. For this early diagnosis, research results suggest that Parkinson's disease, which is a central nervous system disease, can be diagnosed early by analyzing patterns that are difficult to distinguish with the human ear. Can lose.

Therefore, there is a need to provide a method and apparatus for diagnosing the state of health of body organs using voice.

An object of the present invention is to provide an apparatus and method for diagnosing a state of health of body organs, which can identify and diagnose a state of health of body organs using a voice of a user.

 In order to solve the above problems, in order to solve the above problems, a medical condition diagnostic apparatus according to an embodiment of the present invention,

A voice detector for detecting a voice by monitoring the voice;

A voice analyzer extracting a voice feature for each type of health condition to be diagnosed from the voice detected by the voice detector;

A voice diagnosis unit for diagnosing a state of health and monitoring a state of health change by comparing a voice feature extracted from the voice analyzer with a pathological state criterion;

And a diagnosis output unit configured to output the health state and health change state information diagnosed by the voice diagnosis unit as a predetermined type of signal.

Preferably, the voice detector monitors a voice change state of the user.

The voice detector may include an analog to digital converter for converting an input analog signal into a digital signal; And

It is preferable to include a voice discrimination unit for discriminating the voice from the digital signal converted by the analog-digital converter.

The voice analysis unit

A health state selector configured to determine a health state type with respect to the voice detected by the voice detector;

A voice section selector configured to select a voice section for diagnosis according to the health condition type selected by the health condition selector;

A voice feature selector which selects a voice feature corresponding to the health state type selected by the health state selector; And

Preferably, a voice feature extracting unit extracts the voice feature from a user voice input according to the voice feature selected by the feature selector.

Preferably, the speech feature extractor includes a normal speech database and a pathological speech database to distinguish between normal speech and pathological speech.

The voice diagnosis unit

A comparison unit comparing the voice feature extracted from the voice analysis unit with a pathological diagnosis criterion;

A monitoring unit for detecting a change in the voice state of the user over time; And

It is preferable to include a diagnostic processing unit for diagnosing the state of health of body organs according to the comparison result of the comparison unit and the voice change of the monitoring unit.

It is preferable to further include a storage unit for storing and updating the diagnostic result and the diagnostic time information diagnosed by the diagnostic processing unit.

The diagnostic output unit

A digital signal processor converting the health state and health change state information diagnosed by the voice diagnosis unit into audio and video signals;

A display unit for outputting a video signal processed by the digital signal processor; And

It is preferable to include a speaker unit for outputting the audio signal processed by the digital signal processing unit.

In order to solve the above other problem, the medical condition diagnosis method according to an embodiment of the present invention, the process of detecting and collecting voice;

Extracting voice features from the collected voices according to types of diagnostic health conditions;

Diagnosing a health condition and monitoring a health change state by comparing the voice characteristic with a pathological condition criterion;

And outputting the diagnosed health state and health change state information as a notification signal of a predetermined form.

The voice collecting process

Converting the input voice into a digital voice signal;

It is desirable to have a process for discriminating speech from the converted digital speech signal.

The process of extracting the state of health features

Determining a health state type for the collected voices;

Selecting a voice section for diagnosis according to the determined health state type;

Selecting a voice characteristic corresponding to the selected health state type;

And extracting the speech feature from the speech section according to the selected speech feature.

The medical condition diagnosis process

Analyzing a change in the health status of the user over time;

It is preferable to include a process of differently notifying the recommendation message according to the change in the health state of the user over time.

The process of diagnosing the state of health and monitoring the state of health change,

Comparing the extracted voice features with pathological diagnostic criteria;

Detecting a change in a user's voice over time;

It is preferable to have a process for diagnosing the state of health of body organs according to the comparison result and the negative change.

The method may further include collecting, storing, and updating the diagnosed diagnosis result and diagnosis time information.

It is preferable to further include a step of deriving the diagnostic criteria for the pathological voice and the normal voice in order to compare the extracted voice features and the pathological diagnostic criteria.

Deriving the diagnostic criteria for the pathological voice and normal voice,

Constructing a pathological speech database and a normal speech database;

Extracting pathological speech features and normal speech features from the pathological speech database and the normal speech database;

Performing comparative training on the pathological and normal speech features;

According to the comparative training, it is preferable to have a process for deriving diagnostic criteria for pathological voice and normal voice.

The output process

Converting the diagnosed health state and health change state information into audio and video signals;

It is preferable to have a process of outputting the converted audio signal and video signal.

1 is a block diagram of an apparatus for diagnosing a state of health according to an embodiment of the present invention.
FIG. 2 is an embodiment of the voice detector of FIG. 1.
3 is an embodiment of the voice analyzer of FIG. 1.
4 is an embodiment of the negative diagnosis unit of FIG. 1.
5 is an embodiment of the diagnostic output unit of FIG. 1.
6 is a flowchart illustrating a method of diagnosing a state of health according to an embodiment of the present invention.
7 is a detailed flowchart of a medical condition diagnosis method according to an embodiment of the present invention.
8 is a detailed flowchart of a medical condition diagnosis method according to another embodiment of the present invention.
FIG. 9A illustrates an example of deriving diagnostic criteria for pathological and normal voices of FIG. 6.
FIG. 9B illustrates diagnostic criteria of GMM parameters modeled with two Gaussian using GMM.
10 is a flowchart of an automatic diagnostic method using diagnostic criteria according to the present invention.
11 is a flowchart illustrating a medical condition diagnosis method using a mobile phone according to another embodiment of the present invention.

Hereinafter, exemplary embodiments of the present invention will be described with reference to the accompanying drawings.

1 is a block diagram of an apparatus for diagnosing a state of health according to an embodiment of the present invention.

The medical condition diagnosis apparatus of the present invention corresponds to an electronic device capable of voice input, recording, or making a call. For example, an electronic device capable of voice input, recording, or calling may be a TV, a mobile phone, a smartphone, a monitor, a Voice over Internet Protocol (VoIP) phone, a portable multimedia player (PMP), a personal digital assistant (PDA), or a notebook computer. However, it is not limited thereto.

The health state diagnosis apparatus of FIG. 1 includes a voice detector 110, a voice analyzer 120, a voice diagnosis unit 130, and a diagnosis output unit 140.

The voice detector 110 monitors and detects the voice of the user through a recording or a call of the user. In addition, the voice detection unit 110 adds a function that can detect the moment the user speaks (call, game using voice).

In this case, the voice detector 110 may include a function of detecting a moment when the user speaks, and continuously monitors the voice change state of the user even when the user does not recognize the voice.

The voice analyzer 120 extracts a voice feature for each type of diagnosis health state from the voice detected by the voice detector 110. At this time, the voice analysis unit 120 selects a voice feature to increase the efficiency and performance of the diagnosis, and extracts the voice feature in consideration of the characteristics of the electronic device that can have a limited amount of calculation and limited memory compared to the medical device.

The voice diagnosis unit 130 compares the health condition features extracted from the voice analyzer 120 with the pathological condition criteria to diagnose a health condition and monitor a health change state. In addition, the voice diagnosis unit 130 includes a function of collecting / update / save the diagnosis result.

The diagnosis output unit 140 converts the health state and health change state information diagnosed by the voice diagnosis unit 130 into a predetermined signal and outputs the signal to a speaker or a monitor. In this case, the predetermined signal may be, for example, a graphic or audio and / or video signal.

2 is a diagram illustrating an example of the voice detector 110 of FIG. 1.

The voice detector 110 of FIG. 2 includes an analog-digital converter 112 and a voice discriminator 114.

The analog-digital converter 112 converts an input analog signal into a digital signal.

The voice discriminator 114 determines a voice signal from the digital signal converted by the analog-digital converter 112.

As an example of a method of determining a voice signal, voice activity detection (VAD) or end point detection (EPD) may be mentioned.

3 is an embodiment of the voice analyzer 120 of FIG. 1.

The voice analyzer 120 of FIG. 3 includes a health state selector 122, a voice section selector 124, a voice feature selector 126, and a voice feature extractor 128. The voice analyzer 120 may configure each component block differently according to a diagnostic function to be applied to the device, and selectively apply each component block.

The health state selector 122 determines a diagnoseable health state type associated with the voice to diagnose various health conditions from the voice. For example, the medical condition type may be laryngeal cancer or Parkinson's disease.

The voice section selector 124 selects a voice section of the user according to the health condition type selected by the health condition selector 122. The voice section selector 124 may selectively select a voice section of the user according to the selected health state type. For example, when laryngeal cancer is selected, only the vowel portion of the user's voice is selected, and when Parkinson's disease is selected, the entire sentence of the user's voice is selected.

The voice feature selector 126 selects a voice feature corresponding to the health state type selected by the health state selector 122.

For example, as an example of speech features for distinguishing normal speech from pathological speech, JITT (Jitter), SHIMM (Shimmer), HNR (Harmonic-to-Noise Ratio), and MFCC (Mel-Frequency Cepstral Coefficient) But it is not limited thereto.

Here, JITT represents an average change in pitch, SHIMM represents an average change in amplitude, HNR represents a ratio of Harmonic and Noise, and MFCC represents a mathematical coefficient for sound modeling.

Usually, the average change amount of pitch of pathological speech becomes larger than the average change amount of normal speech pitch.

Typically, the average change in the amplitude of pathological speech is greater than the average change in the normal speech amplifier.

Typically, the ratio of the noise component to the harmonic component of the pathological speech is greater than the ratio of the noise component to the harmonic component of the normal speech.

As an example, when laryngeal cancer is selected, JITT (jitter) and SHIM (Shimmer) may be selected as voice features for distinguishing pathological voice.

The speech feature extractor 128 extracts speech features such as JITT, SHIMM, HNR, and MFCC selected by the speech feature selector 126 from the user's speech. The speech feature extractor 128 may extract speech features using general software or Mel-Frequency Cepstral (MFC).

In this case, the voice feature extractor 128 includes a normal voice database and a pathological voice database to distinguish between the normal voice and the pathological voice.

FIG. 4 is an embodiment of the voice diagnosis unit 130 of FIG. 1.

The voice diagnosis unit 130 of FIG. 4 includes a comparison unit 132, a diagnosis processing unit 134, a storage unit 136, and a monitoring unit 138.

The comparison unit 132 compares the voice feature extracted from the voice analyzer 120 with the pathological diagnosis criteria. At this time, the pathological diagnosis standard refers to a standard for determining a health state of body organs through an objective voice feature. In this case, a diagnostic criterion for distinguishing a normal voice from a pathological voice may be learned through a Gaussian Mixture Model (GMM) / Hidden Markov Model (HMM) or a neural network.

The monitoring unit 138 detects a user's voice change over time.

The diagnosis processor 134 diagnoses the health state of the body organs according to the comparison result of the comparator 132 and the user voice change of the monitoring unit 138.

The storage unit 136 stores the diagnosis result diagnosed by the diagnosis processing unit 134 and updates the diagnosis result. The storage unit 136 may be, for example, a magnetic recording medium such as a hard disk, or a nonvolatile memory such as an EEPROM or a flash memory, but is not limited thereto.

5 is an embodiment of the diagnostic output unit 140 of FIG. 1.

The diagnostic output unit 140 of FIG. 5 includes a digital signal processing unit 142, a display unit 144, and a speaker 146.

The digital signal processor 142 converts the health state and health change state information diagnosed by the voice diagnosis unit 130 into audio and video signals.

The display unit 144 displays the video signal processed by the digital signal processor 142 as a graphic or video.

The speaker unit 146 outputs the audio signal processed by the digital signal processor 142 as sound.

6 is a flowchart illustrating a method of diagnosing a state of health according to an embodiment of the present invention.

First, a voice of a user input through recording or a call is detected and collected (step 610).

Next, a voice feature is extracted for each type of health state from the collected user voices (620).

Subsequently, the health condition is diagnosed and the health change state is monitored by comparing the extracted negative features with the pathological criteria (step 630).

In operation 640, the diagnosed health state and health change state information are output as a notification signal of a predetermined form.

As a result, according to an exemplary embodiment of the present invention, an early diagnosis of the health state of the body organs may be performed by diagnosing and informing the user of the health state of the body organs that can be derived by voice from the electronic device.

First, it is checked whether a user's voice is input through recording or a call (step 710).

Subsequently, when a user's voice is input, a voice feature for checking a state of health is extracted from the input voice (S720).

Next, a voice diagnosis is performed by comparing the extracted voice features with the pathological condition criteria (step 730).

In operation 740, the voice diagnosis result and the diagnosis time information are stored for the continuous management of the medical condition. Here, the diagnostic time information may include, but is not limited to, a diagnostic date and a diagnostic history.

At this time, the case where the user wants to diagnose the health state is set as a one-time diagnosis.

Then, it is checked whether the medical condition diagnosis is a one-time diagnosis (step 750).

Subsequently, if the medical condition diagnosis is a one-time diagnosis, the user is notified of the diagnosis result and recommended treatment appropriate for the diagnosis result (step 770).

However, if the medical condition diagnosis is not a one-time diagnosis, the health condition is continuously monitored through the voice of the user. In this case, the controller continuously checks whether a change has occurred in the state of health of the user over time (step 760).

Subsequently, if the user's health condition changes over time, the user is notified of the diagnosis result, and treatment recommended for the diagnosis result is recommended (step 770).

8 is a detailed flowchart of a medical condition diagnosis method according to another embodiment of the present invention.

 First, it is checked whether a voice of a user is input through recording or a call (step 810).

Subsequently, when a user's voice is input, a voice feature for checking a state of health is extracted from the input voice (S820).

Subsequently, a voice diagnosis is performed by comparing the extracted voice features with the pathological condition criteria (step 830).

In operation 840, the voice diagnosis result and the diagnosis time information are stored for the continuous management of the health condition. Here, the diagnostic time information may include, but is not limited to, a diagnostic date and a diagnostic history.

Then, it is checked whether the medical condition diagnosis is a one-time diagnosis (step 850).

Subsequently, if the medical condition diagnosis is a one-time diagnosis, the user is notified of the diagnosis result and recommended treatment appropriate for the diagnosis result (step 852).

However, if the medical condition diagnosis is not a one-time diagnosis, the health condition of the user is analyzed over time by continuously monitoring the health condition through the user's voice (step 860).

Subsequently, the recommendation message is displayed differently according to the change in the health state of the user over time.

For example, if the change in the user's state of health passes 1 to 2 days, the user is notified of the message, "Do not oversee" (step 862). Notify the hospital recommendation message (step 864), and when the user's health condition changes more than two weeks, determine the worsening of the disease and notify the user actively recommending the hospital (step 866).

Then, it is checked whether the health state of the user is improved by the diagnosis result (step 870).

Subsequently, if the user's health condition is improved, the user is notified of the improved state (S880).

FIG. 9A illustrates an example of deriving diagnostic criteria for pathological and normal voices of FIG. 6.

First, a pathological speech database and a normal speech database are constructed.

Then, the pathological speech feature and the normal speech feature are extracted from the pathological speech database and the normal speech database (steps 910 and 920).

Next, comparative training is performed on pathological speech and normal speech features using well-known analytical modeling (step 930).

Then, the comparison training derives the diagnostic criteria for the pathological voice and the normal voice (step 940).

FIG. 9B illustrates diagnostic criteria of GMM parameters modeled using two Gaussian models using a Gaussian Mixture Model (GMM).

Referring to FIG. 9B, the GMM parameters modeled by the first and second Gaussian (shimmer, MFCC2, MFCC3) are shown as a standard of diagnosis for normal voice and pathological voice.

10 is a flowchart of an automatic diagnostic method using diagnostic criteria according to the present invention.

First, a voice of a user is input through recording or a call (step 1010).

Subsequently, a voice feature for a laryngeal state is extracted from the input voice (step 1020).

Subsequently, an automatic diagnosis of a laryngeal state is performed by comparing the extracted voice feature with a diagnosis criterion (step 1030).

Subsequently, the diagnosis result is notified to the user (step 1050).

11 is a flowchart illustrating a medical condition diagnosis method using a mobile phone according to another embodiment of the present invention.

First, a user's call is detected using a mobile phone (step 1110).

In operation 1120, the user's voice is detected.

Subsequently, the voice feature for diagnosis is extracted from the user voice by analyzing the user call voice in real time (step 1130).

Subsequently, the health state is extracted using the extracted voice feature (step 1140).

Subsequently, the diagnosis result is notified to the user (step 1150).

As a result, according to an exemplary embodiment of the present invention, the user may automatically diagnose a health state by using a voice call during a mobile phone call. In addition, the user can save and check the diagnosis result for each call.

Meanwhile, the above-described embodiments of the present invention can be written as a program that can be executed in a computer, and can be implemented in a general-purpose digital computer that operates the program using a computer-readable recording medium. The computer-readable recording medium may include a storage medium such as a magnetic storage medium (eg, a ROM, a floppy disk, a hard disk, etc.), an optical reading medium (eg, a CD-ROM, a DVD, etc.).

So far I looked at the center of the preferred embodiment for the present invention. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the scope will be construed as being included in the present invention.

Claims (18)

A voice detector for detecting a voice by monitoring the voice;
A voice analyzer extracting a voice feature for each type of health condition to be diagnosed from the voice detected by the voice detector;
A voice diagnosis unit for diagnosing a state of health and monitoring a state of health change by comparing a voice feature extracted from the voice analyzer with a pathological state criterion;
And a diagnostic output unit configured to output the health state and health change state information diagnosed by the voice diagnosis unit as a predetermined signal. The apparatus of claim 1, wherein the voice detector monitors a voice change state of the user. The apparatus of claim 1, wherein the voice detector comprises: an analog to digital converter configured to convert an input analog signal into a digital signal; And
And a voice discriminating unit for discriminating a voice from the digital signal converted by the analog-digital converting unit. The method of claim 1, wherein the voice analysis unit
A health state selector configured to determine a health state type with respect to the voice detected by the voice detector;
A voice section selector configured to select a voice section for diagnosis according to the health condition type selected by the health condition selector;
A voice feature selector which selects a voice feature corresponding to the health state type selected by the health state selector; And
And a voice feature extractor configured to extract the voice feature from a user voice input according to the voice feature selected by the feature selector. The apparatus of claim 4, wherein the speech feature extractor comprises a normal speech database and a pathological speech database to distinguish between normal speech and pathological speech. The method of claim 1, wherein the voice diagnostic unit
A comparison unit comparing the voice feature extracted from the voice analysis unit with a pathological diagnosis criterion;
A monitoring unit for detecting a change in the voice state of the user over time; And
And a diagnosis processing unit for diagnosing a state of health of body organs according to a comparison result of the comparing unit and a change in voice of the monitoring unit. The apparatus of claim 6, further comprising a storage unit configured to store and update the diagnosis result and diagnosis time information diagnosed by the diagnosis processing unit. The method of claim 1, wherein the diagnostic output unit
A digital signal processor converting the health state and health change state information diagnosed by the voice diagnosis unit into audio and video signals;
A display unit for outputting a video signal processed by the digital signal processor; And
And a speaker unit configured to output an audio signal processed by the digital signal processor. Detecting and collecting voice;
Extracting voice features from the collected voices according to types of diagnostic health conditions;
Diagnosing a health condition and monitoring a health change state by comparing the voice characteristic with a pathological condition criterion;
And a method of outputting the diagnosed health state and health change state information as a notification signal of a predetermined form. The method of claim 9, wherein the collecting of the voice is performed.
Converting the input voice into a digital voice signal;
And a step of discriminating the voice from the converted digital voice signal. The method of claim 9, wherein the extracting the health state characteristic
Determining a health state type for the collected voices;
Selecting a voice section for diagnosis according to the determined health state type;
Selecting a voice characteristic corresponding to the selected health state type;
And extracting the speech feature from the speech section according to the selected speech feature. The method of claim 9, wherein the medical condition diagnosis process
Analyzing a change in the health status of the user over time;
And a method of notifying the recommendation message differently according to the change of the health state of the user over time. The method of claim 9, wherein the diagnosing the state of health and monitoring the state of health change,
Comparing the extracted voice features with pathological diagnostic criteria;
Detecting a change in a user's voice over time;
And a method of diagnosing a state of health of body organs according to the comparison result and the change of voice. The method of claim 13, further comprising collecting, storing, and updating the diagnosed diagnosis result and diagnosis time information. The method of claim 13, further comprising deriving diagnostic criteria for pathological voice and normal voice to compare the extracted voice features with pathological diagnosis criteria. The method of claim 15, wherein the derivation of diagnostic criteria for pathological and normal voice comprises:
Constructing a pathological speech database and a normal speech database;
Extracting pathological speech features and normal speech features from the pathological speech database and the normal speech database;
Performing comparative training on the pathological and normal speech features;
A method of diagnosing a health condition using a voice, the method comprising deriving diagnostic criteria for pathological voices and normal voices according to the comparative training. The method of claim 9, wherein the output process
Converting the diagnosed health state and health change state information into audio and video signals;
And a step of outputting the converted audio signal and the video signal. A computer-readable recording medium having recorded thereon a program for implementing the method of claim 9.

Images