13
Directivity for real world benefit Michael Syskind Pedersen Ph.D., Hearing aid developer, Oticon

Michael Syskind Pedersen Ph.D., Hearing aid developer, …hadf.hoertech.de/2017/downloads/Pedersen_HADF_2017.pdf · Directivity for real world benefit Michael Syskind Pedersen Ph.D.,

Embed Size (px)

Citation preview

Directivity for real world benefit

Michael Syskind Pedersen Ph.D., Hearing aid developer, Oticon

Michael Syskind Pedersen

Front microphone

Rear microphone

Available microphones for directionality

Michael Syskind Pedersen

[Kuklasiński et al., 2016], [Kuklasiński and Jensen, 2017]

Local and binaural directicvity patterns

Michael Syskind Pedersen

Target direction = steering direction Steering direction = 0 degrees

Target direction (Azimuth angle) [degrees] Target direction (Azimuth angle) [degrees]

[Kuklasiński and Jensen, 2017]

Objective intelligibility prediction of MVDR + Single channel Post filter (MWF)

• Higher intelligibility using 4-microphone binaural MWF compared to two-microphone local MWF • 4-microphone binaural MWF has sensibility to steering vector errors (should be kept <15degrees) • High benefit of 4-microphone binaural MWF towards the sides compared to two-microphone local MWF

Local two microphone directivity

SNR improvement mainly of frontal talkers

Limited SNR improvement, when target is towards the side – especially in diffuse noise fields.

Michael Syskind Pedersen

Local directionality

Four microphone directivity

Possible to achieve a higher SNR

Possible to improve SNR for off-axis target sounds

Michael Syskind Pedersen

Binaural directionality

Four microphone directivity

Possible to achieve a higher SNR

Possible to improve SNR for off-axis target sounds

Risk of missing out target sound due to narrow beampatterns [Mejia et al. 2015]

Spatial cues affected

Off-target directions levels easily become too low – even though the SNR has been improved.

Michael Syskind Pedersen

?

Binaural directionality

Michael Syskind Pedersen

Directional noise management system

8

w1

w2

b

Fixed beamformers Adaptive beamformer

SNR Post filter

SNR estimate Noise removal

Example target talker direction

Front microphone

Rear microphone

Reaching the Opn directivity

• Early version • Good results in laboratory tests (ideal conditions), but…

• Beams were quite aggressive.

• Too much noise was removed also noise which did not have to be removed.

• Field test showed that end-users did not liked it.

• Possible reasons: • Insufficient loudness

• Awareness of the surroundings/Removed the ”good” noise

w1

w2

b

Fixed beamformers Adaptive beamformer

SNR Post filter

SNR estimate Noise reduction

Front microphone

Rear microphone

Problem: The omnidirectional response of a BTE

instrument tends to listen more backwards compared to the natural pinna response

Solution: Select a beamformer directivity pattern that

has similar properties as the natural pinna, i.e. a stable beampattern with a slight, frequency dependent dampening of sounds from the back.

0dB-5dB

-10dB-15dB

0dB-5dB

-10dB-15dB

0dB-5dB

-10dB-15dB

0dB-5dB

-10dB-15dB

BTE Pinna

Pinna BTE

Michael Syskind Pedersen

Directional system in easy enironments

w1

w2

b

Fixed beamformers Adaptive beamformer

SNR Post filter

SNR estimate Noise removal

Front microphone

Rear microphone

Automatic control

Reaching the Opn directivity

• We have a powerful machine which can remove a lot more noise.

• We had to turn it down. • Adapting the machinery to increase loudness and spatial awareness. • Many people preferred Pinna directionality in the majority of the time. • Noise reduction only to be applied, when it is needed. • Individual preferences. • Objective measures and models help, but cannot take us all the way. Fine tuning is subjective . • Noise management system should act as seamless as possible

Conclusions • Ability to reduce noise is important. • Even more important is knowing when to apply noise reduction

• How much noise reduction? • Individual differences

• The beamformer, which removes the most noise is not necessarily the one which people prefer • In the majority of the time, many hearing impaired prefer a stable beampattern that

do not remove too much noise.

• Binaural multi-microphone allows improves ability to remove noise • Narrow beams are not always good • The perceived sound environment will change much more aggressively the more

narrow the beam is, if the user and the environment are dynamic. • Tradeoff between SNR improvement and risk of removing target sound

References • A. Kuklasiński and J. Jensen “Multichannel Wiener Filters in Binaural and Bilateral Hearing Aids —

Speech Intelligibility Improvement and Robustness to DoA”, AES JOURNAL OF THE AUDIO ENGINEERING SOCIETY, vol. 65 no.1/2, pp. 8-16, Jan. 2017

• A. Kuklasiński, S. Doclo, S. H. Jensen, and J. Jensen, “Multichannel Wiener Filter for Speech Dereverberation in Hearing Aids—Sensitivity to DoA Errors,” presented at the AES 60th International Conference: DREAMS (Dereverberation and Reverberation of Audio, Music, and Speech), conference paper 2-2. 2016 Jan.

• Mejia et al., 2015: “Lost of speech perception in noise – cause and compensation”, Proceedings of the 5th symposium on Auditory and Audiological Research (ISAAR)