New HealthRecSys September 20th Rethinking Hearing Aids as … hearing... · 2019. 10. 1. ·...

Post on 17-Oct-2020

0 views 0 download

Transcript of New HealthRecSys September 20th Rethinking Hearing Aids as … hearing... · 2019. 10. 1. ·...

Rethinking Hearing Aids

as Recommender Systems

HealthRecSys ’19 – September 20th, 2019

Alessandro Pasta – Technical University of Denmark (alpas@dtu.dk)

Michael Kai Petersen – Eriksholm Research Centre (mkpe@eriksholm.com)

Kasper Juul Jensen – Oticon A/S (kjen@oticon.com)

Jakob Eg Larsen – Technical University of Denmark (jaeg@dtu.dk)

10%By 2050

Hearing Loss

will have disabling hearing loss 1

2 1 WHO (https://www.who.int/news-room/fact-sheets/detail/deafness-and-hearing-loss)

In 2019

33% of people over 65

have disabling hearing loss 1

of the world’s population

Treating Hearing Loss

3

Rethinking Hearing Aids as Recommender Systems

4

Item User

Complex device

configuration

Multiple parameters

influence the experience of

hearing-impaired users

Hearing is a

subjective sense

Users perceive the sounds

differently and might

benefit from a fully

personalized hearing aid

configuration

Rating

Learning user

preferences

How to effectively gather

them in multiple real-world

situations?

Context

Constantly changing

sound environment

Users might have different

preferences in different

situations

Rethinking Hearing Aids as Recommender Systems

5

1. Fitting SpaceIt’s the space defined by the different

possible combinations of settings that

can be applied to a hearing aid, based on

the audiogram of the user.

Rethinking Hearing Aids as Recommender Systems

65 Mead C Killion. (2002). “New thinking on hearing in noise: a generalized articulation index”, Seminars in Hearing, vol. 23 no. 16 Marozeau, J., and Florentine, M. (2007), “Loudness growth in individual listeners with hearing losses: A review”, Journal of

the Acoustical Society of America: JASA express letters, vol. 122, pages EL81-EL87

1. Fitting SpaceIt’s the space defined by the different

possible combinations of settings that

can be applied to a hearing aid.

2. Default ConfigurationThe default configuration is a

combination of medium settings, adopted

when user preferences are not known.

Previous research showed that people

have different characteristics and hearing

preferences. 5,6

Default

Configuration

Rethinking Hearing Aids as Recommender Systems

7

1. Fitting SpaceIt’s the space defined by the different

possible combinations of settings that

can be applied to a hearing aid.

2. Default ConfigurationThe default configuration is a

combination of medium settings, adopted

when user preferences are not known.

3. Recommended ConfigurationA personalised configuration can be

recommended, based on some specific

characteristics and preferences of the single

user.

HOW TO RECOMMEND A PERSONALISED

CONFIGURATION?

✓ Simplifying the complex audiological space

✓ Gathering user preferences in real-world

situations

Default

Configuration

Recommended

Configuration

1 6

Gender

15 16

75

19 20 19

0

20

40

A B C D E F G

Years

User

Experience with Hearing Aids

The Study Participants

8

The Study Timeline

9

Level 1Soft Gain

Level 2Soft Gain

Level 3Soft Gain

Level 4Soft Gain

Level 1Noise reduction and

directionality

Level 2Noise reduction and

directionality

Level 3Noise reduction and

directionality

Level 4Noise reduction and

directionality

Level 1Brightness

Level 2Brightness

Level 3Brightness

Level 4Brightness

Week 1

Evaluation of Parameter A

Noise reduction and directionality

Week 2

Evaluation of Parameter B

Brightness

Week 3

Evaluation of Parameter C

Soft gain

Week 4

Final test of preferencePrescribed

Configuration

Personalized

Configuration

Level 1Soft Gain

Level 2Soft Gain

Level 3Soft Gain

Level 4Soft Gain

Level 1Noise reduction and

directionality

Level 2Noise reduction and

directionality

Level 3Noise reduction and

directionality

Level 4Noise reduction and

directionality

Level 1Brightness

Level 2Brightness

Level 3Brightness

Level 4Brightness

1What is the

perceived

usefulness of the

parameters?

Usefulness of the 3 Parameters

11

3,4

4,0

1,2

2,3

3,3

1,6

2,7

4,0

4,6

3,33,6

4,3

3,3

4,4

3,6

4,5

2,6

1,2

3,2 3,1

4,2

1

2

3

4

5

A B C D E F G

Ave

rage

use

fuln

ess

User

Average usefulness of the 3 parameters

Noise Reduction and Directionality (n=94) Brightness (n=98) Soft Gain (n=103)

12

2Do people have

different

preferences?

13

3Does the same

person have

different

preferences?

Noise Reduction and DirectionalityUser preferences when the parameter is considered to be useful (Usefulness > 2 out of 5)

14

0%

20%

40%

60%

80%

100%

A(n=5)

B(n=9)

C(n=0)

D(n=4)

E(n=13)

F(n=1)

G(n=20)

Pre

fere

nce

User

Preferred Noise Reduction and Directionality Levels

Lev. 4

Lev. 3

Lev. 2

Lev. 1 Incre

ase in N

R

BrightnessUser preferences when the parameter is considered to be useful (Usefulness > 2 out of 5)

15

0%

20%

40%

60%

80%

100%

A(n=5)

B(n=10)

C(n=7)

D(n=8)

E(n=7)

F(n=12)

G(n=42)

Pre

fere

nce

User

Preferred Brightness Levels

Lev. 4

Lev. 3

Lev. 2

Lev. 1

Incre

ase in B

rightn

ess

Soft GainUser preferences when the parameter is considered to be useful (Usefulness > 2 out of 5)

16

0%

20%

40%

60%

80%

100%

A(n=6)

B(n=16)

C(n=4)

D(n=0)

E(n=11)

F(n=6)

G(n=37)

Pre

fere

nce

User

Preferred Soft Gain Levels

Lev. 4

Lev. 3

Lev. 2

Lev. 1

Incre

ase in S

oft G

ain

17

Personalized

Configuration

4Is real-world

personalization

preferred to how

hearing aids are

fitted in a standard

clinical workflow?

Prescribed

Configuration

Test of Preference

18

✓ 6 out of 7 users preferred the Personalized Configuration

✓ Some users fine-tuned the hearing aids for speech situations

✓ Participants liked to have more than one configuration

Conclusions

19

Item User

We simplified the

audiological design

space and isolated the

most important

parameters.

Users exhibited

different audiological

preferences.

Rating

The device configuration

learned in multiple real-

world environments was

preferred to a traditional

configuration.

Context

The same user

exhibited different

preferences in different

contexts.

Th nk you

Icons made by FreePik and PixelBuddha from www.flaticon.com