Verification

The rationale of DSL v4.1 was to place the DR of sound within a HI listener's residual DR

"Fit everything in"

NAL also found that kids need more gain than adults But only about...

4 dB more, not 7 or 8 dB

With NAL-NL1, the audibility of speech spectra are never more than ___ dB SL

40 dB SL

When we consider average talkers, speech varies a lot in level Soft speech is about ... Average speech is about ... Shouted speech is about ... The DR for one talker, at a single vocal effort is ...

52 dB 62 dB 82 dB 30 dB

NAL-NL1 suggests ____% conductive component for conductive/mixed losses

75% quite a bit more gain add‑in for conductive component for NAL than it is for DSL.

Who decided to keep the hearing aids?

80% of people preferred the verified prescription and over half of these people chose to purchase the hearing aids Those that preferred the first fit, very few decided to keep hearing aids

NAL tried to maximize speech intelligibility based on ***Also juggles the slope of the gain a little bit to really make it as comfortable as possible

AI or SII procedure *Ended up not doing it as well as it possibly could after adjusting for comfort

With reverse slopes, cookie bites, rising slopes, those configurations tend to be more likely congenital Fit more gain?

Also DSL fits more mid frequency gain, might have better luck with those patients But Providing mid-frequency gain, you can get some speech masking from the LF onto the HFs

At this time, there was also data that using a CR above 3:1 was not good -People didn't like it -People didn't perform as well

And so that was the other piece that went into NAL‑NL2 is 1st, we've got to get the gain right for average But then we're going to try to fit soft and loud but we're not going to use compression ratio than is higher than about 3:1.

NAL-NL2 became a multidimensional equation The only way Harvey could solve it was to use a neural network

And this is why, if you just change one frequency, the threshold of one frequency, one of the things you see in NAL is the gain prescription and slope of the gain prescription changes at other frequencies. This is also also the reason why NAL‑NL2 took a few years to get into probe mic systems.Because the probe mic systems didn't have a processing power to do calculations based on a neural network.

Also, when you try to maximize audibility (i.e., make everything fit), it will maximize performance but people will start saying it's too loud and they'll reject their hearing aids

As clinicians we are not trying to make all speech audible. It's a balancing act

Why...?

Audibility

Why...?

Audibility If you fit each premier product to NAL-NL2, there's an improvement across the board

Since DSL was developed with kids in mind, the focus was on

Audibility rather than sound quality *Quickly became the most popular prescription for kids

Gain for first fits is loser than for validated prescriptive methods ... Does this actually matter? Pilot study done ~7 years ago Published in the Hearing Review Took premier products form the Big 6 with fancy special features, put on 2 programs 1) First fit 2) NAL-NL2 fit Took an old single-channel analog instrument and programmed to NAL-NL1 Completed aided QuickSIN presented in sound field at a softer speech level

Bars show average SNR disadvantage compared to individuals with normal hearing Found that the old analog hearing aid fit to NAL-NL1 performed better than a state-of-the-art new premier hearing aid with first fit Why...?

Ok, now let's enter TDs Chose a TD of 90, which is about 10 dB below average

Black bars = what happens in software display Red bars = what happens in real ear For a couple of manufacturers, what they show on the screen and what happens in the real ear are completely different from each other. For some manufacturers, gain for average will change For others, it wont Know your manufacturer

Do these gain differences affect calculated speech intelligibility based on audibility (SII)?

Calculated audibility, or SII, for DSL and NAL Left-hand panel is not accounting for hearing loss desensitization Right-hand panel is accounting for hearing loss desensitization The conclusion of this author is that DSL and NAL basically give you the same calculated audibility They're really giving you equal calculated audibility, so you expect the same speech recognition

Step 1 of NAL

Calculating SII and trying to maximize it

How does NAL-NL1 derive gain?

Choose an input level Calculate normal loudness Choose an audiogram Juggle insertion gains until SII is a maximum and total loudness is normal or less

You might choose the proprietary method in the software (and then fit the NAL-NL2 targets) if the manufacturers turn on special features slightly different (i.e., it'll implement special features the way you intend to), BUT you'll want to...

Confirm that this is true with the manufacturer and prove that's the right way to do it

Dr. Ricketts thinks that maybe preferred listening levels aren't just higher for kids Maybe they're higher in

Congenital hearing loss

Valente et al 2018 Previous 2012 study was NAL-NL1 Now we have NAL-NL2, which is a little lower in level Manufacturers have also had a chance to update their procedures 24 adults New hearing aid users Premium high-end RIC hearing aids with custom ear tips Randomly assigned to "first fit" or "verified fit" group

Consonant recognitive was better with a verified prescription That was particularly true for soft speech where it's often under fit (75% of time) Effects might be greater if participants had more hearing loss (most had normal LF hearing)

Live speech mapping- what can we use it for?

Counseling Hey, look, even though we have done a good job of fitting, not everything is going to be audible. You're still going to have to watch talkers' faces and really use good communication strategies. We're not fixing your hearing.

The only two popular US prescriptive fitting methods that have been systematically researched

DSL NAL

____ Has been used since 1984 Initially developed for fitting children The current version is v5 and is available on most probe-mic equipment and hearing aid fitting software Not available as stand alone software

DSL (Desired Sensation Level)

What do the prescriptions actually look like?

DSL Child vs. DSL Adult Quiet vs. Noise You can see quite a bit less gain prescribed for adults than there are for kids. Why those big differences are important is this question that we sometimes have of transitioning kids from DSL child to DSL adult. It's a pretty big change.

Has new targets for adults and children; Different targets for quiet and noise Children quiet targets = DSL v4.1 7dB less gain and lower CTs prescribed for adults (not congenitally impaired) Slightly less gain and lower CTs for noise CFs for ABR thresholds (nHL) and newest age specific (by month) RECD estimates built in *Thresholds can be entered in HL or nHL Built in corrections for binaural (-3 dB) and conductive (+25% of PTA CC) Changed gain in LFs based on "curvilinear compression rules" *Less LF gain

DSL v5

You also might want to chose the proprietary method in the software if it saves you time

E.g., If you're going to fit to NAL-NL2 using probe mic, then you want to select whichever one gets you closest to those targets

In the late 1980s, we started to see an influx of compression hearing aids Prior to that time, hearing aids were linear What non-linear protocols were available?

E.g., LGOB (loudness growth in octave bands) Resound had one of the first multi-channel compression hearing aids on the market The fitting process was was to have individual listeners rank order, soft, average and loud sounds and give them loudness ratings. They used this prescriptive method to then fit the hearing aid.

What NAL did differently and what DSL came around to was...

Empirically based methods They start out with kind of this theoretical "let's do it this way," and then they fit a bunch of patients and see what they like and how well they do. Modify and improve their prescriptive methods based on patient data and patient outcome data.

Why is there an NAL-NL2?

Empirically derived procedures keep collecting data And in one of their studies, they found 33% (1/3 of people) were saying the gain was too high

So which one of these two methods: restoring normal loudness or equalizing loudness ...works better?

Equalizing loudness You would think restoring loudness would because that's what NH listeners have. Then people with HL lose the compressive mechanism of OHCs, which is what we want to make up.

If you don't CHOOSE new vs. experienced user in the probe-mic system, it will automatically assume

Experienced That may be a little loud, even compared to NAL's own data for those new users.

Abrams et al. (2012) Ricketts likes this study because they didn't rig it to look a certain way; They tried to do things like people do clinically First fit on 1st group NAL-NL1 fit on 2nd group Both groups came in after wearing it for a period of time, and they did fine-tuning Then completed the APHAB

For Ease of Communication, Reverberation, and Background Noise sub-scales, scores obtained with the verified prescription were significantly higher (indicating greater benefit) than those for the initial-fit approach 15/22 preferred their hearing aids programmed to verified prescription

Data from Bentler et al (2006)

For adults... You can see manufacturer's first fits are quite under target Particularly in the HFs compared to NAL-NL2

Do gain differences between the two affect calculated loudness?

For high level inputs but even medium level inputs, DSL is louder. DSL is willing to have it be a little louder because they think sensation level is important. NAL is really focused on comfort, and controlling loudness while still trying to get things as high level as possible. I think these data are probably reflective of why most U.S. clinicians choose NAL instead of DSL

"War of children" between NAL and DSL Which is more appropriate?

Found kids in Australia prefer NAL Kids in Canada prefer DSL Not surprising ... kids like what they were fit with.

Let's talk about the DSL philosophy and where it came from Richard Seewald is in Canada. He's an audiologist there working in the '70s. And they have an MMR outbreak and there's lots and lots of kids with hearing loss, and he starts going, I really need to do a better job in taking an analog hearing aid off the shelf that says this is the fitting range and putting on a kid. Started looking at preferred listening levels in loudness and started looking at discomfort studies. One that really caught his eye was this data from Erber and Witt in 1977 where you can see in individual kids what sensation level led to the percent correct words.

He was really concerned about kids that are learning language for the first time Unlike what we might see in a post‑lingually deafened adult, kids need a pretty high SL before hitting maximum speech recognition. At relatively low sensation levels even + 6 +8 +10, there's still a lot of kids that speech recognition is much poorer than we're seeing at, you know, + 15 or 18, you know, for some kids even +20. So looking at these data, he developed the desired sensation level procedure where he was trying to provide speech at a high enough SL where the kids' performance was pretty close to optimal as high as it was going to be Also considered loudness and wanted to make sure that things weren't too loud, but really he was most interested in audibility.

Now its possible to fit manufacturer's first fit. How?

Implementations of auto‑fit. E.g., some probe mic systems allow you to fit to the manufacturers' fit. *But most people are not using the manufacturer's first fit because they believe in it, they're using it because its easy So most people who chose this automation are not going to use probe mic to fit first fits

Why can't I just use live speech mapping?

It's not reliable Preferred level depends on talker mood

Same paper Comparing DSL and NAL in adults

Largest differences are for high level inputs More gain in LFs for DSL

It was a very different thought process because for the first time you could apply different gains for soft average and loud speech. LGOB really was one of several methods that had this idea of, well, now that we've got compression, let's make soft speech soft, average sounds average and loud sounds loud.

Let's restore normal loudness

But authors did not see a difference in speech recognition in noise performance between NAL-NL2 and first fit Why?

Levels did not trigger directional microphone (needed 70 dB SPL before directional was triggered)

Are there differences between these validated prescriptive methods and manufacturer's first fits?

Manufacturers tend to under fit Why? Want a low return for credit OR There are manufactures that give different gain for speech in quiet and speech in noise (e.g., match to target for quiet, under fit for speech in noise)

Many kids when they become adults tend to like more gain than the DSL adult

Maybe its less gain than the DSL child, but its more gain than the DSL adult

The only complication with doing that is if this listener that you just fit did this and then they had a remake The hearing aid comes back and they see a different audiologist because its a big office The person refitting them selects "experienced," puts in the same program from before, and doesn't verify with probe mic

Might end up above targets Using experienced is going to turn up the gain relative to new.

NAL makes assumptions about the compression knee point So sometimes, the compression in the HA won't match up to the compression NAL is assuming Result?

Might not be able to hit targets E.g., if a manufacturer has higher knee points, slower time constants, you won't be able to hit both average and soft

____ Has been used since 1976 Became quite popular after the introduction of NAL-R in 1986 The current version is NAL-NL2 This is available with most fitting software and most probe-mic equipment and is available stand alone

NAL (National Acoustics Laboratories)

What's left in the US?

NAL based DSL based Manufacturer's Proprietary Cambridge based (Brian Moore's formula)

One of the arguments of the early '90s was whether we should, from a theoretical standpoint restore normal loudness or the idea behind NAL originally was to have equal loudness across frequencies.

NAL didn't attempt to restore normal loudness at all. They did something very different, which was to make high frequency speech sounds and low frequency speech sounds equal in loudness.

SII has importance functions for: -Frequency (red dots) -Nonsense syllables (yellow dots)

NAL's calculations of important audibility are based on Burns' nonsense syllable importance. If you're learning speech, it's important to hear each one of those phonemes.

If you open up the hearing aid fitting software and select proprietary fitting method, tell them it's a new use, do probe microphone and tell them it's an ITE, NAL-NL2 and then fit to targets What did you just fit to?

NAL-NL2 In terms of audibility, what matters is what happens in the ear. Not what you select in the manufacturer fitting software.

Based on trying to maximize calculated audibility Uses loudness equalization (equal loudness across frequencies)

NAL-R

Applies more gain for LFs, less for HFs *Less HF audibility than seems right

NAL-RP

What are the current validated prescriptive methods?

National Acoustics Laboratory-Nonlinear (NAL-NL2) Desired Sensation Level (DSL v5) Cambridge Fitting Formula (CAMFIT2)

So late '80s to early '90s with non-linear/compression hearing aids... we saw this plethora of manufacturer specific, as well as standalone ways to "normalize loudness"

Next to each it says in parentheses, "individual or average." What that means is they either required an individual patient's measurements of loudness, or they used average loudness growth curves for people with that level of hearing loss. *There were lots of them out there and basically all of them have went away.

Was NAL-NL1 a simple speech intelligibility maximizer?

No Believed speech should not be perceived as any louder for HI listeners than perceived by NH

That's been done over and over and over again in the literature, because people keep thinking, well, manufacturers will eventually get closer to prescriptive methods.

Nope...

DSL was a bit different Rather than restoring normal loudness, they wanted to restore...

Normal dynamic range Fit everything in between threshold and TD

3rd philosophy: make sure everything is comfortable ... even if we have to sacrifice some audibility for average speech

Not going to hear soft speech at all (not even the peaks) Not going to hear the valleys of average speech Audibility in this case is very limited; there's concern for poor speech recognition

There's a level distortion factor that SII applies

Once thresholds reach 73 dB SPL, there's more distortion to high speech levels Above that level, NAL is turning gain down a little bit for higher speech levels

V1 of DSL created in 1984 with pen/paper DSL didn't take off until 1990. That's when it went into software (DSL v3). Very quickly DSL in the early '90s became the most popular prescriptive method for kids. Once we started getting compression hearing aids, they started developing this idea and the DSL IO/v4 came out in 1996, and this was optimized a little bit in DSL Version 4.1, which was the standard fitting procedure for kids for a long period of time (90s-2005) In 2005, Version 5.0 of DSL was finally released.

One of the biggest differences between v4 and v5 was considering, do we also want to fit adults? And should/could adults be fitted differently?

NAL found that experienced users, particularly those with more hearing loss, wanted a little more gain.

People in the severe to profound range, and are an experienced hearing aid user, they would suggest up to 3 dB more gain. Whereas if you're inexperienced user, those same listeners might want almost 7 dB less gain.

So they pooled data from five preferred-gain studies

Preferred gain as a function of LF HL Within +/- 3 dB of NAL-NL1, 49% of people said was just right 5% were saying too soft 46% were saying it's too loud So they asked what would happen if they turned gain down 3-4 dB *Turning down by 3 dB is presented by that blue box (3 dB LESS than NAL-NL1 +/- 3 dB) So now when you take a look at that data, 60% say its just right

Around 2005, DSL asked if children and adults need different prescriptions Data from 2005 (Scollie et al.)

Preferred listening level vs. DSL recommended listening level For kids, they were pretty much on the money But for adults, they preferred significantly less gain than they were prescribing. About 7 or 8 dB less. This was really used in DSL 5 to come up with different prescriptions for adults and kids where the adult DSL prescription is prescribing way less gain than the pediatric DSL prescription.

The idea of fitting hearing aids by looking at the output of speech and visualizing audibility came from DSL AudioScan was down the road from where DSL came up. They partnered to coin the SPL-o-gram, which was speech mapping on to the residual DR

Revolutionary for DSL

And so the idea behind NAL was to use your loudness in an effective manner rather than trying to normalize your loudness.

So rather than trying to make it normal, where things might end up being too loud, we only have so much loudness we can use. Let's use the loudness for the areas of speech where it's really important.

NAL-NL1 provides equal loudness across

Speech frequency bands

So from these data and feedback from all of these patients, they came to the conclusion that NAL-NL1 was too loud, and to make it JUST RIGHT they need to reduce gain

Started out with assumptions/rationales (psychoacoustics and speech science) Made theoretically predictions Made empirical observations Adjusted gain Compared to original assumptions

The binaural correction for NAL-NL2 (i.e., how much gain the fitting formula takes away for 2 hearing aids vs. 1) depends on the level of the input signal Why?

Supra-threshold loudness summation depends on the level of the signal For input levels <40 dB, they only take away 2 dB when wearing 2 hearing aids vs. 1 When you get up to 90 dB or so, they're taking away 6 dB for 2 hearing aids vs. 1 So it's a sliding scale for 1 vs. 2 hearing aids

Which of the four philosophies is closest to current prescriptive methods?

The 1st Weight audibility while trying to control for comfort

If you don't input male/female, quiet/noise What is the audio scan doing?

The audio scan assumes is that you're fitting males in noise So the calculation for gain, for soft, average and loud, is based on that you're fitting males in noise. And the assumption for audio scan when fitting kids is that you're fitting kids in noise, that you're not fitting them in quiet.

Why still talk about DSL 4.1/DSL I/O?

The philosophy in DSL 5 for kids is very similar to the philosophy in DSL 4.1.

DSL first fits tend to be a little bit closer to actual DSL than the manufacturer's implementation of NAL Why?

The way NAL and DSL do business. Essentially DSL wants their own people to help manufacturers implement it, to make sure they get it right. Where that's not so true of NAL. They basically send the algorithm and the manufacturer implements it.

Without hearing aids, someone with a moderate to moderately-severe hearing loss is not going to have very good audibility But, how do we make all of this fit between threshold and TD?

There are different philosophies for doing so

If you focus only on pleasantness...

There will be no audibility It's a balancing act

Manufacturers will argue, "You should really use our first fit, because we provide different gains depending on the input. And you really are not getting the right kind of gain if you're not really using our default." The problem with that is ...

There's really actually no peer‑reviewed published data to support that actually works. You're basically trusting the manufacturer that that is actually a good thing to do without data.

What else was problematic with loudness normalization procedures?

They didn't consider supra-threshold summation effects

Data from Seewald et al., 2008

Things didn't look any better for kids... Not comparing to DSL here. Merely just comparing to each other If you're only doing first fit, you are affecting audibility quite significantly just by picking manufacturer You wouldn't know that if you didn't do probe mic

Which validated prescriptive method should I pick for adults? NAL-NL2? DSL v5? CAMFIT2?

This is a paper that Earl Johnson published back in 2012. A1‑A5 or five different configurations of hearing loss. Most of the speech information is 500-4000 Hz. So let's cover everything else up 0 is the difference between DSL and NAL You can see in general these prescriptive methods in terms of gain for adults are plus minus 5 of each other through 500 to 4000 Hz This shouldn't surprise us, because these are empirically derived approaches. So for average speech, they have got more and more and more similar for adult listeners

The last time this was done, it was in 2015 (Sanders et al.), and they evaluated first fit versus NAL across five manufacturers, and looked at it with 8 people.

Under fit for soft 3/4 of cases Under fit for average 1/2 of the cases *About 7-10 dB below In the data here, zero‑line for soft and average would be the manufacturers' first fit equals actual NAL‑NL2 fit. Some manufacturers are above that line. That's a little closer to a DSL fit. Giving a little bit more LF/bass boost sound than the NAL does.

In terms of the APHAB, there was less of a problem in background noise with the

Verified prescription

So even from a sound quality kind of standpoint, there were more that liked the

Verified prescription

Manufacturers will also argue, "We're trying to focus on pleasantness and comfort in noise rather than audibility. We don't want the hearing aid to be uncomfortably loud" And why do they want to make sure the hearing aid is REALLY comfortable?

Want a low return for credit

2nd philosophy: let's make everything fit and overlap This would be a high CR with slow time constants

We made sure everything is audible and nothing is uncomfortable BUT, two things: 1) With lots of gain for soft, we're making sounds that patient's don't want to hear audible 2) If a loud talker and a soft talker were talking at the same time, the hearing aid would adjust gain for the loud talker and you might lose the soft talker

4th philosophy: let's make everything fit and NOT overlap This would be a high CR with fast time constants

We're still going to have the complaint of soft sounds being too loud Now we DONT have the issue where if a loud talker and soft talker are talking at the same time, you soft talker is lost. Instead, the soft segments will be brought up and loud segments will be brought down. BUT, the complaint is going to be that every sounds like they're the same level. This is going to really hurt speech understanding

What would you do for music? Musicians and non‑musicians broken up into low frequencies, mid frequencies and high frequencies. WFL and SSS are two different music selections. The CST is running sentences.

What you can see is in general they wouldn't really change from NAL‑NL2. Non‑musicians showed more variability and an interest in turning down the mid and high frequencies a little bit more and having a little bit more LF emphasis

1st philosophy: let's focus on average speech Make sure most of average is in the audible and comfortable range

When we do this, there's some audibility for soft, and the peaks of shouted speech remain comfortable (below the TD) *No discomfort BUT, if we do this most of the soft speech minima are not audible

There's a lot of variables/information going into these calculations

Which is why it took a neural network initially to calculate it

With maximum outputs, we know that whether or not assigning MPO (i.e., entering in a TD) affects gain depends on the prescriptive method

With DSL, gain for average is affected by the entered TD E.g., If a patient has a narrow dynamic range or has hyperacusis where they have considerable sensitivity to loud sound and you enter a lower threshold of discomfort, DSL will prescribe less gain for average. NAL‑NL2 does not do that.

What do manufacturers do with maximum output? Mueller et al (2008) examined differences among the Big 6 MPO settings for a 50 dB HL *When TD was and was NOT entered

With no TD entered, MPOs differed by 20 dB

If you select manufacturer first fit and push "go" have you just fit to first fit?

You have no idea Without probe mic, you don't know WHAT you've fit To really fit with manufacturer's first fit, accounting for coupling/geometry differences, you ACTUALLY have to do probe mic ... it's just like the other prescriptive targets

How important is "pleasantness"?

You want the hearing aid to be pleasant enough If the patient isn't going to wear the hearing aid, you've lost. They're not going to get any benefit if they don't wear it.

What drives pleasantness? Pilot study that Ricketts did with 5 people

Zero-line is an NAL-NL2 fit Did paired comparisons Had listeners rank them in terms of pleasantness for both speech and music Here's the rank ordering Least gain is most pleasant Most gain is least pleasant Black line was the hearing aid not providing any actual amplification (just enough to make up for insertion loss). Added this at the end, and people said THIS was the most pleasant

For ____ speech inputs, the DSL i/o was similar to the previous DSL (v3). The only difference was it tried to squish things into the dynamic range. So if the TDs were lower than average, the gain for average would also go down. If TDs were equal to average, then the DSL i/o target was exactly the same as the gain for the DSL v3.

average

Differences between DSL and NAL are much larger for ____. DSL is going to have considerably more gain for kids because of this emphasis on the importance of audibility Comparatively to NAL, which is putting more emphasis for kids on the issue of loudness and comfort.

children

Why not 100% of conductive component?

limitations of the hearing aid