JP7323936B2 - Fatigue estimation device - Google Patents

Description

TECHNICAL FIELD The present invention relates to an apparatus for estimating a worker's degree of fatigue based on the worker's voice.

A method of estimating the degree of fatigue of a worker using voice has been proposed. For example, Patent Document 1 discloses the following device. First, a speech during fatigue (non-fatigue) is acquired and a vector of its parameters is recorded, and a parameter analysis is performed on the speech during analysis to generate a vector. Furthermore, whether or not the voice is fatigued is estimated by whether or not it resembles the speech parameter vector during fatigue.

Such a device has the advantage that it is possible to estimate the degree of fatigue by capturing the voice of the worker without performing any special work.

JP 2007-3835

In the above-described conventional technology, for example, words uttered by a taxi driver during communication with a base station are acquired to generate a voice parameter vector, which is then converted into a pre-recorded voice parameter vector when fatigued. determined by comparing with That is, the speech during fatigue and the speech during judgment are different in length and sound quality. For this reason, there is a problem that accurate fatigue estimation cannot always be performed even though the complicated processing of calculating voice parameters and generating vectors is performed.

SUMMARY OF THE INVENTION It is an object of the present invention to solve the above problems and to provide an apparatus capable of performing highly accurate fatigue estimation with simple processing.

Some independently applicable features of the invention are set forth below.

(1)(2) A fatigue estimating apparatus according to the present invention comprises first vocalization acquisition means for acquiring a first vocalization of a worker's utterance at a first time point of work, and a second utterance acquisition means for acquiring a second utterance for the same utterance as the utterance of the worker at a later second point in time; and a fundamental frequency of the first utterance and the second utterance. , a fundamental frequency calculating means for calculating average power or duration, and a fatigue calculator for calculating a degree of fatigue based on the difference in fundamental frequency, average power or duration between the first vocalization and the second vocalization. means.

Therefore, the degree of fatigue can be estimated by acquiring the voices at the first time point and the second time point.

(3) The fatigue estimation device according to the present invention is characterized in that the fatigue degree y is calculated based on the following equation.

y=β ₀ +β ₁ (X ₁ '-X ₁ )
where β ₀ and β ₁ are constants, X ₁ is the fundamental frequency of the first utterance , and X ₁ ' is the fundamental frequency of the second utterance .

Therefore, the degree of fatigue can be estimated by calculating the fundamental frequency of the sound at the first time point and the second time point.

(4) The fatigue estimation device according to the present invention is characterized in that the fatigue degree y is calculated based on the following equation.

y=β ₀ +β ₁ (X ₂ '-X ₂ )
where β ₀ and β ₁ are constants, X ₂ is the average power of the first utterance , and X ₂ ' is the average power of the second utterance .

Therefore, the degree of fatigue can be estimated by calculating the average power of the voice at the first time point and the second time point.

(5) The fatigue estimation device according to the present invention is characterized in that the fatigue degree y is calculated based on the following equation.

y=β ₀ +β ₁ (X ₃ '-X ₃ )
Here, β ₀ and β ₁ are constants, X ₃ is the duration of the first utterance, and X ₃ ' is the duration of the second utterance.

Therefore, the degree of fatigue can be estimated by calculating the duration of the voice at the first time point and the second time point.

(6) The fatigue estimating device according to the present invention is characterized in that the presence or absence of fatigue is calculated as the degree of fatigue depending on whether or not ΔX ₁ in the following equation is equal to or greater than a predetermined value α.

ΔX ₁ =100·(X ₁ '-X ₁ )/X ₁
Here, X ₁ is the fundamental frequency of the first utterance, and X ₁ ' is the fundamental frequency of the second utterance.

Therefore, the degree of fatigue can be estimated regardless of subjectivity of the worker.

(7) The fatigue estimating device according to the present invention is characterized in that the presence or absence of fatigue is calculated as the degree of fatigue depending on whether or not _ΔX2 in the following equation is equal to or greater than a predetermined value α.

ΔX ₂ =100·(X ₂ '-X ₂ )/X ₂
Here, X ₂ is the average power of the first utterance, and X ₂ ' is the average power of the second utterance.

Therefore, the degree of fatigue can be estimated regardless of subjectivity of the worker.

(8) The fatigue estimating device according to the present invention is characterized in that the presence or absence of fatigue is calculated as the degree of fatigue depending on whether or not ΔX3 in the following equation is equal to or greater than a _{predetermined} value α.

ΔX ₃ =100·(X ₃ '-X ₃ )/X ₃
Here, X ₃ is the duration of the first utterance, and X ₃ ' is the duration of the second utterance.

Therefore, the degree of fatigue can be estimated regardless of subjectivity of the worker.

(9) The fatigue estimation device according to the present invention is characterized in that half or more of all spoken words are vowels.

Therefore, changes in the fundamental frequency and the like tend to appear, and estimation can be performed accurately.

(10) The estimating device according to the present invention is characterized in that the spoken word is a word having a meaning related to fatigue.

Therefore, it is easy to speak with emotion, changes due to fatigue appear clearly, and estimation can be performed accurately.

(11)(12) The estimating device according to the present invention provides a first utterance and a feeling of fatigue regarding a worker's utterance at a first point in time of past work, past utterance acquisition means for repeatedly acquiring a plurality of sets of a second utterance and fatigue for the same utterance as the utterance of the worker at the time point 2; A differential fatigue calculation means for calculating a value obtained by subtracting the fatigue at the first point from the fatigue, as a differential fatigue, and extracting and extracting a set in which the differential fatigue is equal to or greater than a predetermined value in each of the past sets. For the sets, the difference between the fundamental frequencies of the first utterance and the second utterance is calculated as the past difference fundamental frequency, and the set in which the differential feeling of fatigue is equal to or greater than a predetermined value is extracted from each of the past sets, and the extracted set , the difference between the average powers of the first utterance and the second utterance is calculated as the past difference average power, and in each of the past sets, the set in which the difference fatigue feeling is a predetermined value or more is extracted, and the first utterance past difference element calculating means for calculating a difference in duration between the voice and the second utterance as a past difference duration; Extraction means for extracting as an extraction element a correlation with the differential fatigue feeling of a predetermined value or more; A target utterance acquiring means for acquiring a second utterance for the said uttered word of the worker at a second time after the time, and a fundamental frequency difference between the first utterance and the second utterance of the target. as the target fundamental frequency, the difference in the average power as the target average power, and the difference in the duration length as the target difference duration length; A target difference element extracting means for extracting a difference element selected by the element extracting means out of the duration length; and a correlation calculated by the element extracting means based on the difference element extracted by the target difference element extracting means. a fatigue level calculation means for predicting the fatigue level using the Correction means for correcting which elements are to be extracted by the element extraction means based on the legitimacy given to the fatigue level evaluation means 66 and their correlation.

Therefore, while estimating the degree of fatigue, it is possible to select an appropriate element and dynamically perform accurate fatigue estimation.

The "first utterance obtaining means" corresponds to step S13.

"Second utterance obtaining means" corresponds to step S16.

"Fundamental frequency calculating means" corresponds to step S17.

"Fatigue level calculation means" corresponds to step S19.

"Program" is a concept that includes not only programs that can be directly executed by the CPU, but also programs in source format, compressed programs, encrypted programs, and the like.

1 is a functional block diagram of a fatigue estimating device according to one embodiment of the present invention; FIG. It is a hardware configuration of a fatigue estimation device. 7 is a flowchart showing a process of generating a fatigue level estimation function; They are accumulated voice data and fatigue data. It is a figure which shows the data of a fundamental frequency, a difference fundamental frequency, and a differential fatigue feeling before and during crew duty. It is a flowchart of a fatigue degree estimation process. It is a flowchart of a fatigue degree estimation process. It is a functional block diagram at the time of constructing a fatigue estimation device as a server device. FIG. 3 is a functional block diagram of a fatigue device according to a second embodiment; It is a flowchart of a fatigue degree estimation process. It is a flowchart of a fatigue degree estimation process. It is a flowchart of a fatigue degree estimation process. FIG. 10 is a diagram showing data on differential fatigue, differential fundamental frequency, differential average power, and differential duration for the utterance "This is an experiment". It is the result of having performed t test about the data of FIG. FIG. 10 is a diagram showing data of differential fatigue, differential fundamental frequency, differential average power, and differential duration for the spoken word "yes"; FIG. 10 is a diagram showing data of differential fatigue, differential fundamental frequency, differential average power, and differential continuation duration for the spoken word "tadaima". FIG. 10 is a diagram showing data of differential fatigue, differential fundamental frequency, differential average power, and differential continuation time for the utterance "a~a". FIG. 10 is a diagram showing data of differential fatigue, differential fundamental frequency, differential average power, and differential duration for the utterance "tired"; FIG. 10 is a diagram showing data on differential fatigue, differential fundamental frequency, differential average power, and differential duration for the utterance "This is an experiment". This is the result of performing a t-test on the data of FIGS. 15 to 18. FIG.

1. 1st embodiment
1.1 Functional Configuration FIG. 1 shows a functional block diagram of a fatigue estimating device according to one embodiment of the present invention. The first utterance acquisition means 2 acquires the first utterance of the worker's utterance at the first time point of the work. A second uttered sound acquisition means 4 acquires a second uttered sound of the uttered word of the worker at a second time point after the first time point.

Fundamental frequency calculation means 6 calculates the fundamental frequency of the first utterance and the fundamental frequency of the second utterance. The fatigue level calculation means 8 calculates the fatigue level based on the difference between the fundamental frequency of the first utterance and the fundamental frequency of the second utterance.

Therefore, the degree of fatigue can be estimated based on the change in the fundamental frequency of speech due to fatigue. In addition, since the same utterance is uttered at the first time point and the second time point, it is possible to accurately grasp the change in the fundamental frequency due to fatigue while facilitating the processing.

1.2 Hardware configuration Figure 2 shows the hardware configuration of the fatigue estimation device. In this embodiment, it is configured as a device that can be built into the dashboard of a taxi or the like.

A memory 32 , a speaker 34 , a control circuit 35 , a hard disk 36 , a DVD-ROM drive 38 and a D/A converter 39 are connected to the CPU 30 . The control circuit 35 is a circuit for controlling the engine and electric system of the taxi automobile.

The hard disk 36 stores an operating system 42 and a fatigue estimation program 44 . The fatigue estimation program 44 cooperates with the operating system 42 to exert its functions. These programs are recorded on the DVD-ROM 46 and installed on the hard disk 36 via the DVD-ROM drive 38 .

The microphone 41 outputs the worker's speech as an audio signal. The audio signal is converted into a digital signal by the D/A converter 39 .

1.3 Calculation of Fatigue Level Estimating Function First, the fatigue level estimating function for estimating fatigue will be described.

The fatigue level estimation function used in this embodiment is as follows.

y=β ₀ +β ₁ (X ₁ '-X ₁ )
Here, y is the degree of fatigue, β ₀ β ₁ is a constant derived by actual measurement, X ₁ is the fundamental frequency of the speech signal before the start of work (first point), X ₁ ' is after work progresses (first 2) is the fundamental frequency of the speech signal.

Using this fatigue level estimation function, the fundamental frequency X ₁ of the speech voice signal before the start of the taxi driver's work (first time point) and the speech voice signal X ₁ ' after the start of work (second time point) are given. , the degree of fatigue y can be estimated.

The constants β ₀ and β ₁ are calculated by multiple regression analysis based on utterance voice signals obtained by having many workers utter the same utterance before and after starting work.

FIG. 3 shows a flowchart of the fatigue level estimation function generation process. This processing can be performed using a general computer (PC) different from the computer shown in FIG.

First, the CPU loads measurement data recorded on a DVD-ROM or the like into the hard disk 36 (step S1). An example of measurement data is shown in FIG. A speech signal when a worker utters an utterance "ah" before work, a speech signal when the worker utters the same utterance "ah" after work, and a work It is a record of self-reported fatigue of workers before and after work. In the drawing, the audio signal is shown as a waveform, but actually time-series digital data is recorded.

Fatigue is indicated by a numerical value from 1 to 5 according to self-reported work vehicle, 1 is "good" and the least fatigue, 2 is "somewhat good", 3 is "normal", and 4 is " Slightly fatigued", and 5 is "fatigued", indicating that the feeling of fatigue is greater. The sample numbers indicate voice signals of different workers (different timings for the same worker). FIG. 4 shows 50 samples as an example.

Next, the CPU calculates the fundamental frequencies X ₁ and X ₁ ' of the sound signals of the captured pre-work and post-work sound signals (step S2). Here, the fundamental frequency is the reciprocal of the period observed in the speech waveform due to the opening and closing of the vocal cords. FIG. 5 shows the calculated fundamental frequencies.

Next, the CPU calculates the difference frequency between the fundamental frequency X ₁ ' after work and the fundamental frequency X ₁ before work (step S3). The calculated difference frequencies are shown in FIG.

Furthermore, the CPU calculates the difference in fatigue between the fatigue after work and the fatigue before work (step S4). The differential fatigue is an index representing the fatigue caused by performing the work. Since the degree of fatigue is a relative one felt by each person, in this embodiment, differential fatigue is used as the degree of fatigue. FIG. 5 shows the calculated differential fatigue.

Subsequently, the CPU performs multiple regression analysis on the relationship between the degree of fatigue (difference fatigue) y and the difference frequency δX ₁ to calculate β ₀ β ₁ in the following equation (step S5).

y=β ₀ +β ₁ (δX ₁ )
The fatigue level estimation function is calculated as described above. The degree of fatigue is calculated as a numerical value from 1 to 5 (5 being the most fatigued state).

1.4 Fatigue Estimation Processing Next, fatigue estimation using the fatigue estimation function will be described. 6 and 7 are flowcharts of the fatigue determination program.

When the CPU 30 receives an engine start signal from the control circuit 36 (step S11), the CPU 30 outputs a message requesting vocalization from the speaker 34 (step S12). It should be noted that the start of the engine may be determined when power is supplied to the fatigue estimation device.

For example, the message requesting vocalization is to urge the crew member to generate such as "Please vocalize 'aa'". In response to this, when the flight attendant utters "Ah!", an analog voice signal is output from the microphone 41 and converted by the D/A converter 39 into a voice digital signal. The CPU 30 acquires this and records it in the hard disk 36 (step S13).

When the crew starts work and a predetermined time (for example, 4 hours) has passed from step S13 (step S14), CPU 30 outputs a message requesting vocalization from speaker 34 (step S15). At this time, the CPU 30 requests the same utterance as the utterance requested in step S12 to be uttered. In other words, it is assumed that the driver is urged to generate an error, such as "Please say ``Ah!''".

In response to this, when the flight attendant utters "Ah!", an analog voice signal is output from the microphone 41 and converted by the D/A converter 39 into a voice digital signal. The CPU 30 acquires this and records it in the hard disk 36 (step S16).

Subsequently, the CPU 30 calculates the fundamental frequencies X ₁ and X ₁ ' of the voice recorded in step S13 (before starting work) and the voice recorded in step S16 (present) (step S17). Further, the fundamental frequency X ₁ before the start of business is subtracted from the current fundamental frequency X ₁ ' to calculate the difference frequency δX ₁ .

The CPU 30 estimates the fatigue level y based on the fatigue level estimation function y=β ₀ +β ₁ (δX ₁ ) (step S19). If the estimated fatigue level y does not exceed a predetermined value, it is determined whether a predetermined period of time has elapsed since step S16 was executed. If the predetermined period of time has elapsed, steps S15 to S19 are executed to estimate the fatigue level.

If the estimated degree of fatigue y exceeds a predetermined value, a warning voice (such as "You need to rest") is output from the speaker 34 (step S21).

After receiving this warning, the crew turns off the engine and rests, and the fatigue estimation program 44 is paused. After resting, when the crew member starts the engine again, the fatigue estimation program 44 is activated, and step S11 and subsequent steps are executed.

Further, if the crew continues to work without turning off the engine even after the warning in step S21, steps S15 to S19 are executed after a predetermined time has elapsed from step S16 to estimate the degree of fatigue. If the degree of fatigue exceeds the predetermined value, the warning is output again (step S21).

As described above, it is possible to estimate the degree of fatigue of the taxi driver, output a warning, and prevent accidents due to fatigue.

In the above implementation, "a~a" is used as an utterance. Since "a~a" includes vowels (consisting only of vowels), the fundamental frequency is likely to change, which is preferable. Therefore, even if a spoken word other than "a~a" is used, it is preferable to use a spoken word that includes a large proportion of vowels.

In addition, since "a~a" is a word that has a meaning related to fatigue, it is easy to express emotions. Therefore, it is preferable because the fundamental frequency tends to change according to fatigue. Therefore, even in the case of using utterances other than "a~a", it is preferable to use utterances such as "I'm tired" that are easy to express emotions. On the other hand, it can be said that utterances such as "I'm home" and "This is an experiment" are less suitable than "I'm tired".

1.5 Miscellaneous
(1) In the above embodiment, even if the engine is not stopped in spite of the warning output, the fatigue level is estimated again after a predetermined period of time has elapsed. However, if the engine is not stopped despite the output of the warning, the predetermined time may be set shorter than the normal predetermined time to estimate the degree of fatigue.

(2) In the above embodiment, the fatigue level estimation of a taxi driver was explained as an example, but the method can be applied to workers in general who perform other tasks.

(3) In the above embodiment, the start of business is defined as the first time point, and the time during business is defined as the second time point. However, a point in time during work may be set as the first point in time, and a later point in time during work may be set as the second point in time.

(4) In the above embodiment, a warning is issued, but a process for removing the cause of fatigue, such as stopping the engine, may be executed.

(5) In the above embodiment, the fatigue level is estimated using the difference frequency (difference between the fundamental frequency at the second point in time and the fundamental frequency at the first point in time). However, the degree of fatigue may be estimated using the differential duration length (the difference between the duration length of the audio signal at the second time point and the duration length of the audio signal at the first time point). The fatigue level estimation function in this case is as follows.

y=β ₀ +β ₃ (δX ₃ )
where β ₀ and β ₃ are constants, and δX ₃ is the difference duration. The method of calculating the fatigue level estimation function is the same as in FIG. It is to be noted that utterances such as "tired" that are likely to express emotion when uttered are likely to cause changes in the differential duration due to fatigue.

Alternatively, the degree of fatigue may be estimated using the differential average power (the difference between the average power at the second point in time and the average power at the first point in time).

(6) In the above embodiment, the fatigue level estimation function is calculated based on the correlation between the difference frequency and the differential fatigue feeling to estimate the fatigue level. However, assuming that the difference frequency increases with fatigue, the rate of change ΔX ₁ of the difference frequency may be estimated as being proportional to the degree of fatigue.

The fatigue level estimation function in this case is as follows.

ΔX ₁ = 100·(X ₁ '-X ₁ )/X ₁
For example, if ΔX ₁ exceeds a predetermined value, it is determined that the user is fatigued.

A fatigue level estimation function can be similarly generated for the difference time duration.

(7) In the above embodiment, the fatigue estimation device installed in a taxi has been described. However, a normal computer may be used as the fatigue estimation device. Moreover, you may build as a server apparatus. FIG. 8 shows a functional block diagram when the fatigue estimation device is constructed as a server device.

The terminal device 20 and the server device 10 are connected through a communication line such as the Internet. The terminal device 20 is provided with a microphone and generates an audio digital signal of the spoken word at the first point in time. Similarly, a speech digital signal of the spoken word at a second time point is generated. These voice digital signals are transmitted to the server device 10 by the first vocalization transmission means 3 and the second vocalization transmission means 5 .

The first utterance acquisition means 2, the second utterance acquisition means 4, the fundamental frequency calculation means 6, and the degree of fatigue calculation means 8 in the server device 10 have the same functions as in FIG. The fatigue level calculation means 8 transmits the estimated fatigue level to the terminal device 20 . The warning means 7 of the terminal device 20 that has received the fatigue level outputs a warning from a speaker, display, or the like.

(8) In the above embodiment, the fatigue estimation function is generated based on voice data of many people. However, the fatigue estimation function may also be generated based on the voice data of the crew whose fatigue is to be estimated. This improves the accuracy.

(9) In the above embodiment, the first point in time is before the start of work, and the second point in time is during work. However, any point in time during work may be the first point in time and the second point in time (point after the first point in time).

(10) The above modifications can be combined with other embodiments as long as they do not contradict the essence.

2. Second embodiment
2.1 Functional Configuration FIG. 9 shows the functional configuration of the fatigue estimation device according to the second embodiment. The past utterance acquisition means 50 reads out a plurality of sets of recorded voice digital data of past utterances (speech digital data at the first and second time points and fatigue feeling sets).

The differential fatigue calculation means 52 calculates the difference in fatigue between the first time point and the second time point as the differential fatigue feeling.

The past difference element calculating means 54 calculates the difference between the fundamental frequencies of the audio digital data at the first time point and the second time point of each set, and the average power of the audio digital data at the first time point and the second time point of each set. A difference, the difference in the duration of the audio digital data at the first time point and the second time point of each set, is calculated.

The element extracting means 56 extracts, as an extraction element, one of the difference frequency, difference average power, and difference duration length that has a degree of correlation with the difference fatigue feeling equal to or greater than a predetermined value.

The target utterance acquisition means 58 acquires the utterances of the subject whose fatigue level is to be estimated (speech digital data at the first and second points in time).

The target difference element calculation means 60 calculates the difference between the fundamental frequencies of the speech digital data at the first time point and the second time point, the difference between the fundamental frequencies of the speech digital data at the first time point and the second time point, The difference in the average power of the audio digital data at time 2 and the difference in duration of the audio digital data at the first time and the second time are calculated.

The target difference element extraction means 62 extracts the difference in the fundamental frequency, the difference in the average power, and the difference in the duration calculated by the target difference element calculation means 60, and extracts the degree of correlation with the differential fatigue feeling by the element extraction means 54. Select the one associated with the extracted element as high.

The fatigue level prediction means 64 calculates the fatigue level using the correlation calculated by the element extraction means 56 based on the difference element selected by the target difference extraction means 62 . The fatigue level evaluation means 66 determines the legitimacy of the calculated fatigue level.

The correction means 68 receives data from the target difference element calculation means 58 and the fatigue level extraction means 54 and corrects the correlation in the element extraction means 54 and which difference element is to be extracted by feedback.

Therefore, it is possible to improve the prediction accuracy of the degree of fatigue along with the utilization.

2.2 Hardware configuration The hardware configuration is the same as in Fig.2.

2.3 Fatigue Level Estimation Process In the first embodiment, the fatigue level estimation function was fixed. This embodiment differs in that the fatigue level estimation function dynamically changes.

10 to 12 show flowcharts of the fatigue determination program. Steps S31 to S36 are the same as steps S11 to S16 in FIG.

After acquiring the voice of the crew member at the first time point (before duty) and the second time point (during duty), the CPU 30 calculates the fundamental frequency (step S37). Further, the fundamental frequency at the first point in time is subtracted from the calculated fundamental frequency at the second point in time to calculate a differential fundamental frequency (step S38).

Next, the CPU 30 calculates the average power at the first time point and the second time point (step S39). Further, the average power at the first point in time is subtracted from the calculated average power at the second point in time to calculate the difference average power (step S40).

Next, the CPU 30 calculates the length of duration between the first time point and the second time point (step S41). Further, the difference duration length is calculated by subtracting the duration length at the first time point from the calculated duration length at the second time point (step S42). These are recorded on the hard disk 36 .

In this embodiment, the fatigue level is not necessarily predicted using all of the difference fundamental frequency, the difference average power, and the difference duration time calculated above, but prediction is performed based on the extracted elements among them. I'm trying The extraction process will be explained later, and the following explanation will be given assuming that the elements have been extracted. For example, here, it is assumed that the difference continuation time length is extracted and the following fatigue level estimation function is recorded.

y=β ₀ +β ₃ (δX ₃ )
where β ₀ and β ₃ are constants, and δX ₃ is the difference duration.

In step S43, the CPU 30 estimates the degree of fatigue based on the difference continuation time length calculated in step S42. If the estimated fatigue level y does not exceed a predetermined value, it is determined whether a predetermined period of time has elapsed since the execution of step S42.

If the estimated degree of fatigue y exceeds the predetermined value, a warning voice ("Rest is required", etc.) is output from the speaker 34 (step S45).

After receiving this warning, the crew turns off the engine and rests, and the fatigue estimation program 44 is paused. After resting, when the crew member starts the engine again, the fatigue estimation program 44 is activated and steps S31 and subsequent steps are executed.

Further, if the crew continues to work without turning off the engine even after the warning in step S45, steps S31 to S43 are executed to estimate the degree of fatigue after a predetermined time has elapsed from step S42. If the degree of fatigue exceeds the predetermined value, the warning is output again (step S45).

In this embodiment, after the CPU 30 estimates the degree of fatigue based on the extracted difference elements (step S43), the speaker 34 prompts the crew member to input a self-evaluation of fatigue (step S46). This input can be made from, for example, a touch display (not shown) or a microphone (not shown), and can be entered in five stages from 1 to 5. The self-evaluation fatigue feeling is recorded in the hard disk 36 as a set with the difference fundamental frequency, the difference average power, and the difference continuation time calculated in steps S37 to S42.

In addition, in this embodiment, the CPU 30 uses the previously recorded difference fundamental frequency, difference average power, difference duration length, and fatigue, and the newly added difference fundamental frequency, difference average power, and difference duration length. , the fatigue level estimation function is recalculated based on the feeling of fatigue.

When the fatigue level estimation in step S43 reaches a predetermined number of times (or when a predetermined period of time elapses) since the last recalculation of the fatigue level estimation function, the CPU 30 recalculates the extraction of the difference element (step S48). That is, which of the difference fundamental frequency, the difference average power, and the difference continuation time has the highest correlation with the self-evaluation of fatigue is selected.

Here, it is assumed that the phrase "This is an experiment" is used as an utterance. FIG. 13 shows recorded differential fatigue, differential fundamental frequency, differential power, and differential duration. The bottom six data are the data recorded after the last recalculation. The CPU 30 calculates the degree of correlation with the differential fatigue feeling for each of the differential fundamental frequency, the differential power, and the differential duration including the added data. In this embodiment, the t-test is used to calculate the degree of correlation.

FIG. 14 shows the results of t-test on the data in FIG. According to this, the difference continuation time length is extracted as the difference element because the difference continuation time length has the highest degree of correlation (the numerical value is small). In other words, the difference continuation time length will continue to be used as the difference element.

Subsequently, the CPU 30 performs multiple regression analysis on the relationship between the degree of fatigue (differential fatigue) y and the differential duration δX ₃ to calculate β ₀ β ₁ in the following equation (step S49).

y=β ₀ +β ₁ (δX ₃ )
The CPU 30 records the calculated fatigue level estimation function in the hard disk 36, and uses it for fatigue level estimation from the next time.

As described above, in this embodiment, highly correlated difference elements are re-extracted and the fatigue level estimation function is recalculated. Therefore, it is possible to realize a fatigue estimating device that becomes more accurate through operation.

2.4 Miscellaneous
(1) In the above embodiment, only one highly correlated difference element is extracted in step S48. However, it is also possible to extract a plurality of difference elements (such as extracting those below a predetermined t value) and estimate the degree of fatigue based on the following equation.

y=β ₀ +β ₁ (δX ₁ )+β ₂ (δX ₂ )
Moreover, regardless of the degree of individual correlation, multiple regression analysis may be performed using all the difference elements using the following formula to determine the estimation formula.

y=β ₀ +β ₁ (δX ₁ )+β ₂ (δX ₂ )+β ₃ (δX ₃ )
where _{.delta.X.sub.1} is the difference fundamental frequency, _{.delta.X.sub.2} is the difference average power, and _{.delta.X.sub.3} is the difference duration.

(2) The above modifications can be combined with other embodiments as long as they do not contradict the essence.

3. Experimental results Figures 15 to 19 show the differential fatigue, differential fundamental frequency, differential average power, and Indicates the difference duration length. Further, FIG. 20 shows the respective t-test values.

As shown in FIG. 20, the differential fundamental frequency had a good correlation with the differential fatigue in any utterance. Out of these, utterances such as ``aah'' and ``tired'', which tend to convey emotions, showed a good correlation.

Regarding the difference average power, relatively short utterances such as "Crawling" and "Tadaima" showed a good correlation.

As for the difference duration, utterances that are easy to convey emotions such as "aaaa" and "tired" showed a good correlation.

Claims (13)

a first utterance acquisition means for acquiring a first utterance of the worker's utterance at a first time point of the work;
a second utterance acquisition means for acquiring a second utterance of the same utterance as the utterance of the worker at a second time point after the first time point of the work;
Fundamental frequency calculation means for calculating the fundamental frequency, average power or duration of the first vocalization and the second vocalization;
Fatigue calculation means for calculating the degree of fatigue based on the difference in fundamental frequency, average power, or duration between the first vocalization and the second vocalization;
A fatigue estimation device comprising
The first vocalization acquisition means outputs a message requesting the driver to vocalize the first vocalization upon receiving an automobile engine start signal,
The second vocalization acquisition means outputs a message requesting the driver to vocalize the second vocalization after a predetermined time has elapsed since the engine was started,
The fatigue level calculation means outputs a warning to the driver when the calculated fatigue level exceeds a predetermined value,
It is detected whether or not the engine of the automobile has been stopped after the warning output, and if the engine is not stopped , a message is output to the driver requesting the utterance of the second utterance, and the second utterance obtaining means outputs a message to the driver. A fatigue estimating device, wherein a second utterance is newly acquired, and a degree of fatigue is calculated using the newly acquired second utterance by means of fatigue level calculation means. A fatigue estimation program for realizing a fatigue estimation device by a computer, the computer comprising:
a first utterance acquisition means for acquiring a first utterance of the worker's utterance at a first time point of the work;
a second utterance acquisition means for acquiring a second utterance of the same utterance as the utterance of the worker at a second time point after the first time point of the work;
Fundamental frequency calculation means for calculating the fundamental frequency, average power or duration of the first vocalization and the second vocalization;
A fatigue estimation program for functioning as fatigue calculation means for calculating the degree of fatigue based on the difference in the fundamental frequency, average power or duration of the first utterance and the second utterance,
The first vocalization acquisition means outputs a message requesting the driver to vocalize the first vocalization upon receiving an automobile engine start signal,
The second vocalization acquisition means outputs a message requesting the driver to vocalize the second vocalization after a predetermined period of time has elapsed since the engine was started,
The fatigue level calculation means outputs a warning to the driver when the calculated fatigue level exceeds a predetermined value,
It is detected whether or not the engine of the automobile has been stopped after the warning output, and if the engine is not stopped , a message is output to the driver requesting the utterance of the second utterance, and the second utterance obtaining means outputs a message to the driver. A fatigue estimation program characterized by causing a computer to function so as to newly acquire a second utterance and to calculate a fatigue level using the newly acquired second utterance by a fatigue level calculation means. In the apparatus of claim 1 or the program of claim 2,
A device or a program, wherein the calculation of the degree of fatigue y is performed based on the following equation.
y=β ₀ +β ₁ (X ₁ '-X ₁ )
where β ₀ and β ₁ are constants, X ₁ is the fundamental frequency of the first utterance, and X ₁ ' is the fundamental frequency of the second utterance. In the apparatus of claim 1 or the program of claim 2,
A device or a program, wherein the calculation of the degree of fatigue y is performed based on the following equation.
y=β ₀ +β ₁ (X ₂ '-X ₂ )
where β ₀ and β ₁ are constants, X ₂ is the average power of the first utterance, and X ₂ ' is the average power of the second utterance. In the apparatus of claim 1 or the program of claim 2,
A device or a program, wherein the calculation of the degree of fatigue y is performed based on the following equation.
y=β ₀ +β ₁ (X ₃ '-X ₃ )
Here, β ₀ and β ₁ are constants, X ₃ is the duration of the first utterance, and X ₃ ' is the duration of the second utterance. In the apparatus of claim 1 or the program of claim 2,
A device or program for calculating the presence or absence of fatigue as a degree of fatigue depending on whether or not ΔX ₁ in the following expression is equal to or greater than a predetermined value α.
ΔX ₁ =100·(X ₁ '-X ₁ )/X ₁
Here, X ₁ is the fundamental frequency of the first utterance, and X ₁ ' is the fundamental frequency of the second utterance. In the apparatus of claim 1 or the program of claim 2,
A device or program for calculating the presence or absence of fatigue as a degree of fatigue depending on whether or not ΔX ₂ in the following expression is equal to or greater than a predetermined value α.
ΔX ₂ =100·(X ₂ '-X ₂ )/X ₂
Here, X ₂ is the average power of the first utterance, and X ₂ ' is the average power of the second utterance. In the apparatus of claim 1 or the program of claim 2,
A device or program for calculating the presence or absence of fatigue as a degree of fatigue depending on whether or not ΔX ₃ in the following expression is equal to or greater than a predetermined value α.
ΔX ₃ =100·(X ₃ '-X ₃ )/X ₃
Here, X ₃ is the duration of the first utterance, and X ₃ ' is the duration of the second utterance. In the device or program according to any one of claims 1 to 8,
A device or program, wherein more than half of the total number of words in said spoken words are vowels. In the device or program according to any one of claims 1 to 9,
The device or program, wherein the spoken words are words having a meaning related to fatigue. A first utterance and fatigue for a worker's utterance at a first time point in the past task, and the same utterance as the worker's utterance at a second time point after the first time point in the past task. a past utterance acquisition means for obtaining a set of second utterances and fatigue for a word and repeatedly acquiring a plurality of sets;
differential fatigue calculation means for calculating, as a differential fatigue, a value obtained by subtracting the fatigue at the first time from the fatigue at the second time;
Extracting a set in which the differential fatigue feeling is equal to or greater than a predetermined value from each of the past sets, calculating the difference between the fundamental frequencies of the first utterance and the second utterance as a past difference fundamental frequency for the extracted set, For each set in the past, a set with a differential fatigue feeling of a predetermined value or more is extracted, and for the extracted set, the difference between the average powers of the first utterance and the second utterance is calculated as the past difference average power, and the past A past difference element calculation means for extracting a set in which the difference fatigue feeling is equal to or greater than a predetermined value from each set of , and calculating the difference in the duration length of the first vocalization and the second vocalization as the past difference duration length;
an extracting means for extracting, as an extraction element, one of the past difference fundamental frequency, the past difference average power, and the past difference continuation time that has a correlation with the difference fatigue feeling equal to or greater than a predetermined value;
a first utterance for a worker's utterance at a first time point in the current task; a second utterance for the worker's utterance at a second time point after the first time point in the current task; a target vocalization acquisition means to acquire;
Calculate the difference between the fundamental frequencies of the first utterance and the second utterance of the target as the target fundamental frequency, calculate the difference in average power as the target average power, and calculate the difference in duration as the target difference duration length. a target difference element calculation means for
target difference element extracting means for extracting a difference element selected by the element extracting means from among the target fundamental frequency, the target difference average power, and the target difference duration length;
a fatigue level calculation means for predicting a fatigue level based on the difference element extracted by the target difference element extraction means, using the correlation calculated by the element extraction means;
Fatigue evaluation means for the worker to judge the validity of the predicted fatigue level;
Correction means for correcting which elements are extracted by the element extraction means based on the difference elements calculated by the target difference element extraction means and the legitimacy given to the fatigue level evaluation means, and the correlation between them ,
Fatigue estimator with A fatigue estimation program for realizing a fatigue estimation device by a computer, the computer comprising:
A first utterance and fatigue for a worker's utterance at a first time point in the past task, and the same utterance as the worker's utterance at a second time point after the first time point in the past task. a past utterance acquisition means for obtaining a set of second utterances and fatigue for a word and repeatedly acquiring a plurality of sets;
differential fatigue calculation means for calculating, as a differential fatigue, a value obtained by subtracting the fatigue at the first time from the fatigue at the second time;
Extracting a set in which the differential fatigue feeling is equal to or greater than a predetermined value from each of the past sets, calculating the difference between the fundamental frequencies of the first utterance and the second utterance as a past difference fundamental frequency for the extracted set, For each set in the past, a set with a differential fatigue feeling of a predetermined value or more is extracted, and for the extracted set, the difference between the average powers of the first utterance and the second utterance is calculated as the past difference average power, and the past A past difference element calculation means for extracting a set in which the difference fatigue feeling is equal to or greater than a predetermined value from each set of , and calculating the difference in the duration length of the first vocalization and the second vocalization as the past difference duration length;
an extracting means for extracting, as an extraction element, one of the past difference fundamental frequency, the past difference average power, and the past difference continuation time that has a correlation with the difference fatigue feeling equal to or greater than a predetermined value;
a first utterance for a worker's utterance at a first time point in the current task; a second utterance for the worker's utterance at a second time point after the first time point in the current task; a target vocalization acquisition means to acquire;
Calculate the difference between the fundamental frequencies of the first utterance and the second utterance of the target as the target fundamental frequency, calculate the difference in average power as the target average power, and calculate the difference in duration as the target difference duration length. a target difference element calculation means for
a target difference element extraction means for extracting a difference element selected by the element extraction means from among the target fundamental frequency, the target difference average power, and the target difference duration length;
a fatigue level calculation means for predicting a fatigue level based on the difference element extracted by the target difference element extraction means, using the correlation calculated by the element extraction means;
Fatigue evaluation means for the worker to judge the validity of the predicted fatigue level;
Based on the difference element calculated by the target difference element extraction means and the legitimacy given to the fatigue level evaluation means, which element is extracted by the element extraction means and as a correction means for correcting the correlation Fatigue estimation program to make it work. In the device or program according to any one of claims 1 to 10,
The apparatus or program, wherein acquisition of the new second utterance by the second utterance acquisition means and calculation of the fatigue level by the fatigue level calculation means are executed after a time period shorter than the predetermined time has elapsed.

Images