Eating Habit Improvement System Using Dietary Sound

Haruka Kamachi

1 a

, Sae Ohkubo

2 b

and Anna Yokokubo

1,2 c

and Guillaume Lopez

1,2 d

Graduate School of Science and Engineering, Aoyama Gakuin University, Sagamihara, Japan

Department of Integrated Information Technology, Aoyama Gakuin University, Sagamihara, Japan

Keywords:

Eating Activity Detection, Dietary Sound, Wearable Devices, Behavior Transformation.

Abstract:

Obesity may cause lifestyle diseases such as diabetes and high blood pressure. Eating slowly and chewing

well are essential to prevent obesity. This research aims to improve the consciousness of dietary behavior

based on eating habits by quantifying eating behavior. It proposes “ChewReminder,” a smartphone application

software that detects eating activities in real-time under a natural meal environment and gives feedback based

on detected activity. ChewReminder detects four activities: chewing, swallowing, talking, and other.The

smartwatch gives feedback using vibration depend on chewing count per one bite which information was

linked from the smartphone. Also, the total feedback about the meal was displayed on the smartphone after

ﬁnishing the meal. The chewing count for 70% subjects and chewing pace for more than half subjects was

improved with using ChewReminder by the result of total chewing count, average of chewing count per bite

and chewing pace. ChewReminder is effective especially people who are aware of fast eating. Also, the

result of long-term experiment indicated that feedback displayed on a smartphone was effective to improve

consciousness of eating activity. Therefore, the result of both experiment shows that ChewReminder is a valid

system to improve consciousness of eating activity especially chewing activity.

1 INTRODUCTION

According to the World Health Organization (WHO),

overweight means BMI 25 or more and obesity means

BMI 30 or more. In 2016, 39% of those aged 18

years or older were overweight and 13% of them were

obese. Also, obese people are increasing about three

times since 1975(Obe, ). Obesity may cause lifestyle

diseases such as diabetes and high blood pressure.

Chewing well induces more saliva secretion and

faster blood-sugar level increase. As a result, it works

on the satiety center and hungry feeling is satisﬁed.

That leads prevention of obesity(che, ). Also, Fig-

ure 1 shows that the quicker one eats, the higher BMI

is(Ando et al., 2008). It means that obesity is strongly

related to eating quickly. Therefore, eating slowly and

chewing well is essential to prevent obesity.

Furthermore,Past studies revealed that conversa-

tion during the meal is related to health. So, in-

creasing conversation during the meal is desirable.

(Kishida and Kamimura, 1993).

https://orcid.org/0000-0002-9269-1026

https://orcid.org/0000-0002-4802-834X

https://orcid.org/0000-0003-2657-4961

https://orcid.org/0000-0002-9144-3688

Figure 1: The relationship between BMI and BMI increase

rate with eating speed (quoted from(Ando et al., 2008)).

Recently, many wearable devices available on the

market. However, none of these devices can detect

automatically dietary behavior, especially multiple di-

etary behavior such as swallowing and talking in ad-

dition to chewing in a natural meal environment yet.

If the number of chewing during the meal can be

detected, it become possible to give feedback to peo-

ple in real-time that they do not chew enough or eat

too quickly, leading to prevent obesity. Also, it can be

expected that detecting a conversation during meals

can be used to make feedback promoting the increase

of conversation during the meal. The detection of

more speciﬁc eating behavior can contribute to the

promotion of healthier dietary habits.

346

Kamachi, H., Ohkubo, S., Yokokubo, A. and Lopez, G.

Eating Habit Improvement System Using Dietary Sound.

DOI: 10.5220/0011676300003414

In Proceedings of the 16th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2023) - Volume 5: HEALTHINF, pages 346-353

ISBN: 978-989-758-631-6; ISSN: 2184-4305

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

2 RELATED RESEARCH

Mitsui et al. suggested a system that judges the num-

ber of chewing and the status of speaking in real-time

by using a bone conduction microphone and gives

real-time feedback to the user to improve his/her eat-

ing behavior(Mitsui et al., 2018). They could count in

real-time chewing behavior with 91% accuracy and

utterance length with 96% accuracy. However, they

did not evaluate the performances in the natural meal

environment yet, and the only used speciﬁc foods.

Nakaoka et al. developed a system “eat2pic” that

encourages users to eat healthier by sensing what they

eat using a chopstick IoT equipped with a camera

and IMU sensors, and dynamically changing digital

artwork based on this information.(Nakaoka et al.,

2021).

However special devices are needed. The feed-

back system using a smartphone need to look a dis-

play during the meal. Beside, almost all systems

have not detected speaking in addition to chewing and

swallowing for feedback in real-time under a natural

meal environment.

From above, this research aims to the feedback of

speciﬁc dietary intake behavior in a natural meal envi-

ronment based on the detection eating activity in real-

time. Therefore, it develop and suggest feedback sys-

tem including real-time feedback with a smartwatch

from audio data by using a commercially and readily

available bone conduction microphone placed inside

the ear.

3 PROPOSED SYSTEM:

ChewReminder

3.1 System Overview

The overall of “ChewReminder” is shown in Figure2.

The user put a bone conduction microphone into the

ear and wear a smartwatch on the wrist during a meal.

The eating sound is collected using a bone conduction

microphone that communicates via Bluetooth with

the smartphone. Next, the section of eating activity

is segmented automatically in real-time and predicted

type of activity using classiﬁcation model. The pre-

dicted activity leads some information about the meal

such as meal time and chewing count. These informa-

tion is used as feedback. The feedback is performed

by two types: real-time and after the meal.

The expert said that doing feedback immediately

after the action is effective. Therefore, a smartwatch

is used to feedback in real-time to give a feedback im-

mediately after the action. The smartphone is used to

Figure 2: The system overview.

feedback whole meal information after meal because

that a smartphone can display a lot of information.

The smartphone used was a Google Pixel 3 and 3a

(produced by Google co. Ltd.), the bone conduction

microphone was a MOTOROLA Finiti HZ800 Blue-

tooth Headset (produced by Motorola co. Ltd.), and

the smartwatch was a POLAR M600 (produced by

POLAR). The sound signal sampling from the micro-

phone was 8kHz.

3.2 Algorithm of Eating Activity

Detection

3.2.1 Classiﬁcation Model

The machine learning was used to build models

that can classify chewing, swallowing, speaking, and

other sounds. The classiﬁcation model with Ran-

dom Forest using 31 extracted features suggested by

Kondo et al. was selected to implement the Android

app feedback system.(Kondo et al., 2021).

3.2.2 Method of Segmentation

The eating activity was detected by the Android app

implemented algorithm of detection eating activity

using dietary sound collected by a bone conduction

microphone. This algorithm segmented audio data

automatically corresponding to the detailed eating ac-

tivity suggested by Kamachi et al. (Kamachi et al.,

2021).

3.3 Feedback System

3.3.1 Eating Activity Detection App

The proposed feedback system in this research

(ChewReminder) consists of an Android app using

both a smartphone and a smartwatch. The smartphone

app detects four eating activities: chewing, swallow-

ing, speaking, and others. First, the user connects

a bone conduction microphone to a smartphone via

Eating Habit Improvement System Using Dietary Sound

347

Bluetooth. Next, after he/she taps Bluetooth ON but-

ton, he/she can start the system.

The bone conduction sound is automatically seg-

mented by segmentation method described in 3.2.2.

After segmentation, the features are extracted and eat-

ing activity is predicted using implemented classiﬁca-

tion model by method described in 3.2.1. The classiﬁ-

cation model implemented in the app was built using

Weka(wek, ), which is a machine learning software.

In this research, chewing and bite are deﬁned as fol-

lowing.

• Chew/Chewing: the action/activity of masticating

food between teeth once.

• Bite: the process composed of several chews from

putting food into the mouth to swallowing it.

3.3.2 Feedback Content

The smartphone app measures following information

in real-time. These information are used to do feed-

back. Also, the app records the each detection time,

detection activity, the number of bite and the each

chewing count per bite in csv ﬁle.

• The total meal time between app start when tap-

ping START button to stop when tapping STOP

button

• The total chewing count in the meal

• The chewing count per one bite and calculate av-

erage chewing count per bite from all chewing

count per bite

• The chewing pace as average time per one chew-

ing (seconds / chewing) calculated using time of

one bite and chewing count per bite

• The total talking time by summarizing time of de-

tection speaking

• The total eating time of chewing calculated by

summarizing each time of one bite

The feedback system is consist of two types.

These are real-time feedback using a smartwatch and

visual feedback using a smartphone after the meal.

3.3.3 Real-Time Feedback on Smartwatch

The real-time feedback on a smartwatch uses a vi-

bration depend on chewing count per one bite to do

natural meal preventing looking a visual feedback all

time like a display of smartphone during the meal.

The smartwatch vibrates long when swallowing is de-

tected if chewing count per bite is less than 20. Also,

the smartwatch vibrates short when chewing count

Figure 3: Screenshots of feedback display examples.

per bite achieve 20 to tell user about reaching the stan-

dard value. The standard count per bite is set 20 be-

cause that modern people chew from 10 to 20 times

per one bite(ave, ).

3.3.4 Feedback on Smartphone

The smartphone app display feedback of whole meal

after ﬁnishing the meal as taping the STOP button.

The item of feedback is the total meal time, total

chewing count, average of chewing count per bite, av-

erage of chewing pace, rate of eating time and rate of

talking time. Figure 3 shows an examples of smart-

phone feedback display.

The feedback display was used an icon, bar chart,

text of number value and text sentence of advise to

understand easily. The each item has a standard value

and shows a double circle if the each value exceed

the standard value. The standard value of total meal

time is 20 minute because of from 15 to 20 minute

is the ideal meal time(mea, ). The feedback of meal

time changed bar chart color and advice message de-

pend on three steps: less than 15, between from 15

to 20 and 20 or more. The standard value of aver-

age chewing count per bite is 20. The reason why is

described in 3.3.3. An icon of apple and advice mes-

sage are changed depend on three steps: less than 10,

between from 10 to 20 and 20 or more. The stan-

dard value of chewing pace and last item was deter-

mined by data collected from four people during the

meal. From these result and environment of meal, the

standard value of chewing pace is 1.0 second per one

chewing. The last item of feedback is based on talk-

ing rate and eating rate. The good meal have 20% or

more eating rate calculated by eating time and total

meal time from above result. Also, this research de-

ﬁned that good meal include 10% or more talking rate

calculated by detection speaking time and total meal-

time. The icon and advice message is changed depend

on three steps: condition of good meal, condition of

bad meal and other.

HEALTHINF 2023 - 16th International Conference on Health Informatics

348

4 EVALUATION EXPERIMENT

The real-time feedback experiment was performed to

evaluate veriﬁcation of effectiveness real-time feed-

back with smartwatch. Also, long-term experiment

was performed to evaluate veriﬁcation of effective-

ness feedback displayed on a smartphone.

4.1 Overview of the Experiment

This experiment was performed to verify if proposed

feedback system affect the consciousness of eating

activity. The purpose of the real-time feedback exper-

iment was investigation of changes for eating activity

and consciousness especially by aspect of real-time

feedback. The purpose of the long-term experiment

was investigation of changes for eating activity and

consciousness especially by aspect of displayed feed-

back.

4.2 Experimental Method

4.2.1 Real-Time Feedback Experiment Method

Data from 14 men and women aged from 21 to 26

years old were collected. Two or three subjects were

participating together at the same time because they

were required not only eating but also talking to each

other during the meal. The subjects were required to

eat twice a lunch box on different days. Once with

real-time feedback on a smartwatch, and the other

time without real-time feedback.

The subjects wore an in-ear bone conduction mi-

crophone and a smartwatch in both experiment.

Before the experiment, subjects answered ﬁve

grade evaluation of the questionnaire items to conﬁrm

the result for people who has consciousness of eating

fast.

The subject experimented case with real-time

feedback was required to be conscious of chewing

count per bite based on vibration feedback types. Af-

ter experiment case with real-time feedback, subjects

answered the questionnaire about ChewReminder.

Figure 4 shows the experiment environment. The all

experiment was performed around lunch time.

The following indices were chosen for evaluation:

total meal time, total chewing count, average of chew-

ing count per bite and chewing pace displayed in ﬁ-

nal feedback on a smartphone. These values are com-

pared between case with real-time feedback and non

real-time feedback.

The evaluation result is good when these values is

increased. The better eating habit, the longer the meal

time at least 20 minute. The many chewing count is

Figure 4: Snapshot of the data collection environment.

good for healthy meal. Also, chewing count per bite

need increase for good meal. The chewing slowly is

good to prevent obesity, so bigger value of chewing

pace is good to eat.

Also, subjects answered the questionnaire about

this feedback system and ChewReminder after exper-

iment case with real-time feedback.

4.2.2 Long-Term Experiment Method

This experiment was performed to verify if ChewRe-

minder affects the eating activity especially aspect of

displayed feedback on a smartphone.

Data from three women (subject 1, 2 and 3) aged

from 21 to 42 years old were collected. They used

ChewReminder consecutively around a week during

the meal regardless of type of meal like lunch and din-

ner.

• Subject 1 used this system without real-time feed-

back on a smartwatch for a week (seven days) dur-

ing almost every meal. The total data is 19.

• Subject 2 used this system with real-time feed-

back on a smartwatch for four days during lunch

and dinner with her family. The total data is six.

• Subject 3 used this system with real-time feed-

back on a smartwatch for ﬁve days during lunch

and dinner eaten alone. The total data is seven.

Subjects answered the questionnaire same as real-

time feedback experiment before start the data collec-

tion and after the all data collection.

5 EVALUATION RESULTS

5.1 Real-Time Feedback Experiment

5.1.1 The Total Meal Time

The number of people who the total meal time was

longer value obtained with real-time feedback than

that without real-time feedback was six out of total

14. Figure 11 shows the result of all subjects.

Eating Habit Improvement System Using Dietary Sound

349

Figure 5: Total meal time per subject regardless the order of

the experiments.

Figure 6: Average total

meal time when the ﬁrst

meal was using real-time

feedback.

Figure 7: Average total

meal time when the ﬁrst

meal was without feedback.

The box plot of the result are showed in following

ﬁgure. Figure 6 shows the result of case with real-

time feedback at ﬁrst and Figure 7 shows the result of

case without real-time feedback at ﬁrst. The box plot

result shows that the mean and median value when

the meal was using feedback increased compared to

the value of non real-time feedback in group that ex-

perimented using real-time feedback at ﬁrst. Also, the

median value when the meal was using real-time feed-

back is increased compared to value when the meal

was without real-time feedback in the group that ex-

perimented without feedback at ﬁrst.

The following ﬁgure shows the result of total meal

time per subject. Figure 8 shows the result of case

with real-time feedback at ﬁrst and Figure 9 shows the

result of case without real-time feedback at ﬁrst. The

four out of seven people improved the value of total

meal time from obtained without real-time feedback

to obtained using real-time feedback in the group that

experimented ﬁrstly using real-time feedback. The al-

most all people is over the standard time as 20 minute.

The total meal time of less than half people for oth-

ers was shorter or not changed. The reason why less

than half of the participants increased their meal time

may be that environment was different during the two

meals: difference the rate of talking and eating pace

depend on number of people who eat together.

5.1.2 The Total Chewing Count

The number of people who increased the total chew-

ing count data with real-time feedback from data

without real-time feedback was nine out of total 14.

Figure 8: Total meal time

per subject when the ﬁrst

meal was using real-time

feedback.

Figure 9: Total meal time

per subject when the ﬁrst

meal was without feedback.

Figure 10: Total chewing count per subject regardless the

order of the experiments.

Figure 10 shows the result of all subjects.

The box plot of the result are showed in follow-

ing ﬁgure. Figure 12 shows the result of case with

real-time feedback at ﬁrst and Figure 13 shows that

the result of case without real-time feedback at ﬁrst.

The box plot result shows that the mean value when

the meal was using real-time feedback increased com-

pared to the value when the meal was without real-

time feedback in both group. It was conﬁrmed that

chewing count can be improved by using ChewRe-

minder.

The following ﬁgure shows the result of total

chewing count per subject. Figure 14 shows the result

of case with real-time feedback at ﬁrst and Figure 15

shows the result of case without real-time feedback at

ﬁrst. The four out of seven people improved the value

of total chewing count from obtained without real-

time feedback to obtained using real-time feedback

in the group that experimented ﬁrstly using real-time

feedback. The more than half of participants (four

out of six) who answered agree or strongly agree to

the questionnaire before experiment items of “Do you

think you eat fast (eating fast)?” increased total chew-

ing count.

Figure 11: Total meal time per subject regardless the order

of the experiments.

HEALTHINF 2023 - 16th International Conference on Health Informatics

350

Figure 12: Average total

chewing count when the

ﬁrst meal was using real-

time feedback.

Figure 13: Average total

chewing count when the

ﬁrst meal was without feed-

back.

Figure 14: Total chewing

count per subject when the

ﬁrst meal was using real-

time feedback.

Figure 15: Total chewing

count per subject when the

ﬁrst meal was without feed-

back.

5.1.3 The Average Chewing Count

The number of people who increased the average of

chewing count per bite with real-time feedback from

data without real-time feedback was 10 out of total 14

which is over than 70%. Figure 16 shows the result of

all subjects.

The box plot of the result are showed in fol-

lowing ﬁgure. Figure 17 shows the result of case

with real-time feedback at ﬁrst and Figure 18 shows

the result of case without real-time feedback at ﬁrst.

The box plot result shows that the mean and median

value when the meal was using real-time feedback in-

creased compared to the values of non real-time feed-

back in both group. It was conﬁrmed that chewing

count per bite can be improved by using ChewRe-

minder.

Four out of seven people who used ChewRe-

minder ﬁrst and six out of seven who used it last had

higher average chewing count per bite when using

ChewReminder real-time feedback. Almost all the

participants (ﬁve out of six) who answered agree or

strongly agree to the questionnaire before experiment

items of “Do you think you eat fast (eating fast)?” in-

creased their average chewing count per bite.

Figure 16: Average chewing count per subject regardless

the order of the experiments.

Figure 17: Average chew-

ing count when the ﬁrst

meal was using real-time

feedback.

Figure 18: Average chew-

ing count when the ﬁrst

meal was without feedback.

Figure 19: Chewing pace per subject regardless the order of

the experiments.

5.1.4 The Eating Pace

The number of people who increased the chewing

pace with real-time feedback from data without real-

time feedback was eight out of total 14 which is over

than half. Figure 19 shows the result of all subjects.

The box plot of the result are showed in fol-

lowing ﬁgure. Figure 20 shows the result of case

with real-time feedback at ﬁrst and Figure 21 shows

the result of case without real-time feedback at ﬁrst.

The box plot result shows that the mean and median

value when the meal was using real-time feedback in-

creased compared to the values of non real-time feed-

back in both group. It was conﬁrmed that chewing

pace can be improved by using ChewReminder.

The following ﬁgure shows the result of chewing

pace per subject. Figure 22 shows the result of case

with real-time feedback at ﬁrst and Figure 23 shows

the result of case without real-time feedback at ﬁrst.

Five out of seven people improved the value of chew-

ing pace from obtained without real-time feedback to

obtained using real-time feedback in the group that

experimented ﬁrstly using real-time feedback. The

three out of seven people improved the value of chew-

Figure 20: Chewing pace

when the ﬁrst meal was us-

ing real-time feedback.

Figure 21: Chewing pace

when the ﬁrst meal was

without feedback.

Eating Habit Improvement System Using Dietary Sound

351

Figure 22: Chewing pace

per subject when the ﬁrst

meal was using real-time

feedback.

Figure 23: Chewing pace

per subject when the ﬁrst

meal was without feedback.

ing pace from obtained without real-time feedback

to obtained using real-time feedback in the group

that experimented ﬁrstly without real-time feedback.

There are no signiﬁcant difference. The almost all

participants (ﬁve out of six) who answered agree or

strongly agree to the questionnaire before experiment

items of “Do you think you eat fast (eating fast)?”

increased chewing pace. The difference of chewing

pace between case with real-time feedback and with-

out real-time feedback for other participants who did

not increase the value was from 0.1 to 1.0 seconds per

one chewing.

5.1.5 The Questionnaire

The average of the System Usability Scale was 83.6.

It is signiﬁcantly above the average score of SUS,

which is 68. The SUS score of 13 out of 14 people

is over the average score. It indicates the usability of

ChewReminder. Also, More than half of the people

in the questionnaire showed an increase in conscious-

ness of chewing and talking compared to their an-

swers in the questionnaires.It indicates the effective-

ness of ChewReminder in improving the conscious-

ness of eating behavior.

5.2 Long-Term Experiment

Figure 24 and ﬁgure 25 shows the changes in the av-

erage chewing count per bite and the chewing pace

for subject 1. From subject 1, all results are over 20, a

commonly used reference value of sufﬁcient chewing

count per bite. Subject 1 did not use real-time feed-

back on a smartwatch. This result indicates that the

displayed feedback on the smartphone is sufﬁcient to

maintain a good eating habit of chewing well, such

as chewing over 20 bites.The chewing pace increased

from the ﬁrst time to the last time. Although ev-

ery meal is different, the chewing pace has a rising

trend. It indicates the possibility that ChewReminder

improves the eating pace by using it.

Figure 26 and ﬁgure 27 shows the changes in the

average chewing count per bite and the chewing pace

for subject 2. From result of subject 2, chewing count

per bite is decreasing trend at ﬁrst, last three data is

Figure 24: Average chew-

ing count of subject 1.

Figure 25: Chewing pace of

subject 1.

Figure 26: Average chew-

ing count of subject 2.

Figure 27: Chewing pace of

subject 2.

increasing. The 5th data was obtained after pause of

few days, so it is considering that the average chew-

ing count per bite was decreased.Chewing pace is in-

creasing trend in ﬁrst half. It is also considering that

the pause of few days caused decreasing of chewing

pace.

Figure 28 and ﬁgure 29 shows the changes in the

average chewing count per bite and the chewing pace

for subject 3. From result of subject 3, average of

chewing count per bite is rising trend for all data. This

result indicate that the ChewReminder using real-time

feedback and displayed feedback after the meal can

improve chewing count per bite. Chewing pace is not

rising trend for all data. The decreased data is meal of

noodle.

The participants of this long-term experiment an-

swered the questionnaire. SUS score of subject 1 was

67.5, subject 2 was 70.0 and subject 3 was 87.5. The

average of their result was 75.0. It indicate the usabil-

ity of ChewReminder.

Figure 30 shows the result of questionnaire of pos-

itive aspect which is good when selecting strongly

agree. Figure 31 shows the result of questionnaire of

negative aspect which is good when selecting strongly

disagree. The result of questionnaire shows that only

one people was conscious of talking and eating slowly

during the meal by using ChewReminder. More than

half people answered agree or strongly agree to the

other items of questionnaire. From these, it is re-

vealed that ChewReminder improved the conscious-

ness of eating activity during a meal and chewing.

Also, two people answered a higher number in the ﬁve

Figure 28: Average chew-

ing count of subject 3.

Figure 29: Chewing pace of

subject 3.

HEALTHINF 2023 - 16th International Conference on Health Informatics

352

Figure 30: Result of questionnaire of positive aspect.

Figure 31: Result of questionnaire of negative aspect.

grade evaluation to the questionnaire items about con-

scious of chewing in the meal compared to conscious

of chewing to the answer of questionnaire before the

experiment, indicating effect especially conscious of

chewing activity by using the ChewReminder.

6 CONCLUSIONS

This research proposed ChewReminder which is a

feedback system using dietary sound by real-time

feedback on a smartwatch and displayed feedback

on a smartphone to improve eating habit. The An-

droid app on a smartphone detect eating activities by

three steps: segmentation dietary sound, extraction

features and prediction using a classiﬁcation model.

The smartwatch gives feedback using vibration de-

pend on chewing count per one bite which informa-

tion was linked from the smartphone. Also, the total

feedback about the meal was displayed on the smart-

phone after ﬁnishing the meal.

The objectives of this research was achieved.

ChewReminder can detect eating activity in real-time

under a natural meal environment and provide feed-

back based on detected eating activity. The chewing

count for 70% subjects and chewing pace for more

than half subjects was improved with using ChewRe-

minder by the result of total chewing count, average

of chewing count per bite and chewing pace. The pro-

posed system chewReminder is effective especially

people who are aware of fast eating. Also, the re-

sult of long-term experiment indicated that feedback

displayed on a smartphone was effective to improve

consciousness of eating activity. Therefore, the re-

sult of both experiment shows that ChewReminder is

a valid system to improve consciousness of eating ac-

tivity especially chewing activity.

REFERENCES

Eating slowly has many beneﬁts! (in japanese). Tech-

nical report, Sawaki Internal Medecine and Dia-

betes Clinic. https://www.osaka-tounyoubyou.jp/ ry-

ouri/yukkuritabe/ (Accessed on 01/25/2022).

Eating too fast can lead to obesity and metabolic

syndrome. six measures to help you chew your

food well (in japanese). Technical report, Japan

Preventive Association of Life-style related Dis-

ease. https://seikatsusyukanbyo.com/calendar/2017/

009495.php.

Keep your biting power (in japanese). Techni-

cal report, Japan Dental Hygienists’ Associa-

tion. https://www.jdha.or.jp/pdf/health/hatookuchi/

20200401/2.pdf (Accessed on 1/25/2022).

Obesity and overweight. Technical report, World Health

Organization. https://www.who.int/news-room/fact-

sheets/detail/obesity-and-overweight.

Weka. https://www.cs.waikato.ac.nz/ml/weka/index.html

(Accessed on 1/25/2022).

Ando, Y., Hanada, N., and Yanagisawa, S. (2008). Does

”eating slowly” lead to prevent obesity ? Health Sci-

ence and Health Care, 8(2):51–63.

Kamachi, H., Kondo, T., Hossain, T., Yokokubo, A., and

Lopez, G. (2021). Automatic segmentation method of

bone conduction sound for eating activity detailed de-

tection. In Adjunct Proceedings of the 2021 ACM In-

ternational Joint Conference on Pervasive and Ubiq-

uitous Computing and Proceedings of the 2021 ACM

International Symposium on Wearable Computers,

pages 310–315.

Kishida, N. and Kamimura, Y. (1993). Relationship of con-

versation during meal and health and dietary life of

school children. The Japanese Journal of Nutririon

and Dietetics, 51(1):23–30.

Kondo, T., Kamachi, H., Yokokubo, A., and Lopez, G.

(2021). Eating activity detection method by automatic

segmentation using bone conduction sound. Multime-

dia, Distributed, Cooperative, and Mobile Symposium

2021, 2021(1):310–316.

Mitsui, H., Ohara, J., Yokokubo, A., and Lopez, G. (2018).

Method to improve real-time chewing and speaking

detection accuracy from bone-conduction sound. Mul-

timedia, Distributed, Cooperative, and Mobile Sym-

posium 2018, 2018:562–566.

Nakaoka, R., Nakamura, Y., Matsuda, Y., Misaki, S., and

Yasumoto, K. (2021). eat2pic: Food-tech design

as a healthy nudge with smart chopsticks and can-

vas. In 2021 IEEE International Conference on Perva-

sive Computing and Communications Workshops and

other Afﬁliated Events (PerCom Workshops), pages

389–391. IEEE Computer Society.

Eating Habit Improvement System Using Dietary Sound

353