DEVELOPMENT OF NAL-NL2 Harvey Dillon, Gitte Keidser, Teresa Ching, Matt Flax, Scott Brewer The HEARing CRC & The National Acoustic Laboratories creating sound valueTM www.hearingcrc.org

Prescribe hearing aids to: Make speech intelligible Make loudness comfortable Prescription affected by other things localization, tonal quality, detection of environmental sounds, naturalness. We know how to model intelligibility and loudness, but not the other goals. Intelligibility and loudness are also arguably more important than the other goals. Consequently, the derivation process is based on intelligibility and loudness.

Deriving optimal gains - step 1 Speech spectrum & level Loudness model Normal loudness Gain-frequency response Amplified speech spectrum Compare Audiogram Intelligibility model Intelligibility achieved Loudness model Loudness (hearing impaired) The bottom left loop acts to adjust gain frequency response (upwards) so that intelligibility is maximized. The bottom right loop acts to adjust gain (downwards) so that loudness is no greater than that perceived by a normal hearer.

The audiograms Rejection criterion : -30<= G <=60 , where G is the slope sum(H(f))/3 <=100 , where f is in the set {0.5, 1, 2} kHz Inverted hearing loss profiles used Some of the audiograms used to derive the optimum gains.

Deriving optimal gains - step 1 Audiogram 1 Speech level 1 Optimal gain frequency response Audiogram 1 Speech level 2 Optimal gain frequency response Audiogram 1 Speech level 3 Optimal gain frequency response Audiogram 2 Speech level 1 Optimal gain frequency response 200 audiograms x 6 speech levels  1200 gain–frequency responses, each at 20 frequencies from 125 Hz to 10 kHz The optimal gains that resulted from each audiogram for each speech level.

Overall prescription approach Psychoacoustics Compare Adjust Theoretical predictions Final formula Assumptions, rationale The theoretical predictions are then compared with empirical findings obtained over the last decade and are adjusted when there is a conflict. Speech science Empirical observations

Limiting compression ratio As an example, compression ratio should not be allowed to become too high, especially for fast-acting compression, or intelligibility is reduced. For normal hearing no compression is needed, and for profound hearing loss, compression ratio should not exceed 1.5 in the low frequencies.

Multi-dimensional equation A neural network H250 H500 H1000 H2000 H8k SPL A neural network, made up of successive layers of perceptrons is used to calculate the gain for any arbitrary audiogram and speech level. The coefficients within each layer of the neural network are trained by using the gains that were obtained from the theoretical derivation, as adjusted by the comparison with empirical findings. G250 G500 G1000 G2000 G8k

The two key ingredients Compare Speech spectrum & level Gain-frequency response Amplified speech spectrum Loudness model Normal loudness Loudness (hearing impaired) Audiogram Intelligibility model Intelligibility achieved We need a loudness model and an intelligibility model.

Psychoacoustics We have studied a lot of psychoacoustics, (dead regions, tuning curves, otoacoustic emissions) but there is not time to talk about in this presentation.

Why are hearing thresholds so useful? Frequency selectivity Temporal resolution Speech Perception proficiency Hearing thresholds Central auditory processing Other Although super-threshold abilities like frequency selectivity and temporal resolution are very important, they are partly predicted by hearing thresholds, so it is feasible to have a prescriptive formula based on hearing thresholds alone. Age Cognitive ability

BKB, VCV and CUNY The key result of the modified speech intelligibility index model is how the maximum amount of information available to people (when it is amplified to be audible) varies with the degree of hearing loss, on average.

Factors affecting prescription Experimental findings, relative to NAL-NL1, allow us to fine tune the NAL-NL2 prescription to agree with empirical findings.

Gain; adults, medium input level (N = 187) On average, adults prefer less gain than prescribed by NAL-NL1. Males prefer more gain than females. For those with a moderate or severe loss, experienced aid wearers prefer more than new aid wearers.

Gain; adults, medium input level (N = 187)

Gain for adults: low & high input levels Suggest that the compression ratio should be slightly higher, at least for clients with mild and moderate hearing loss People prefer higher compression ratios than prescribed by NAL-NL1. For low input levels, they turn the gain up more than NAL-NL1 prescribes. For high input levels, they turn the gain down more than NAL-NL1 prescribes.

Binaural loudness correction Listening with two ears provides more loudness than listening with one. Less gain is therefore used for bilateral fittings, especially at higher input levels.

Empirical evidence: variations from NAL-NL1 Output level NAL-NL1 Children Adults Compared to NAL-NL1, NAL-NL2 provides more gain at low input levels for children, but less gain at high input levels for adults. Input level

Adults – congenital or acquired? The difference in gain preferences between children and adults is just that, rather than a difference between those with congenital hearing loss and those with adventitious hearing loss.

Effect of language Gain at each frequency depends on importance of each frequency Low frequencies more important in tonal languages Two versions of NAL-NL2 Tonal languages Non-tonal languages Tonal languages (like Mandarin) convey more information with the pitch of speech than do non-tonal languages (like English). Consequently, the low frequencies are given more gain for tonal languages than for non-tonal languages.

Tonal versus non-tonal language The gain differences between tonal and non-tonal languages are, however, small, and we are currently gathering more information on the frequency importance function for tonal languages.

Example audiogram: moderate sloping Frequency (Hz) 250 125 500 1k 2k 4k 8k 20 40 60 80 100 120 50 dB 65 dB Hearing threshold (dB HL) 80 dB Example of the insertion gain prescription for a moderate sloping loss.

Example audiogram: flat 60 Frequency (Hz) 250 125 500 1k 2k 4k 8k 20 40 60 80 100 120 50 dB 65 dB Hearing threshold (dB HL) 80 dB

Example audiogram: steeply sloping Frequency (Hz) 250 125 500 1k 2k 4k 8k 20 40 60 80 100 120 50 dB 65 dB Hearing threshold (dB HL) 80 dB

Example audiogram: extreme ski-slope Frequency (Hz) 250 125 500 1k 2k 4k 8k 20 40 60 80 100 120 50 dB 65 dB Hearing threshold (dB HL) 80 dB

Example audiogram: reverse sloping Frequency (Hz) 250 125 500 1k 2k 4k 8k 20 40 60 80 100 120 50 dB 65 dB Hearing threshold (dB HL) 80 dB

Variables in NAL-NL2 Age Vent Tube Comp speed WBCT N Language Depth Gender RECD Bi-uni CR RECD BWC Experience REUG Aid type UCT I/O REDD MLE Transducer REIG REAG AC CG AC' ABG ESG BC' RESR ESCD BC SSPL2cc This is the way that the various quantities come together within the NAL-NL2 software. Blue = User i/p Grey = internal variable Red = effect of saturation Dash-dot = alternatives Green = stored data SSPLES Limiting type

“A challenge for the profession is to devise fitting procedures that are scientifically defensible and the challenge for the individual audiologist is to choose the best procedures from whatever are available” A quote from Denis. Denis Byrne, 1998

Thanks for listening Acknowledgements www.hearingcrc.org www.nal.gov.au This research was financially supported by the HEARing CRC established and supported under the Australian Government’s Cooperative Research Centres Program creating sound valueTM