It's been a while since I blogged about hearing aids -- I think I've been waiting for something "interesting and new" to come up, and the past 2 weeks talking about DSP haven't quite done it for me. However, Wednesday's talk on digital noise reduction (DNR) completely fired me up, so I'm going to write about that, then backfill in the DSP bits as needed.
"Why do we have DNR?" Dr. Alexander asked the class. "What do you tell your patients about it?" The audiology students volunteered a few answers, and then I gave mine: "Because I don't want to keep turning my hearing aids off and on and off and on when I walk through the hallways and the student union and around campus."
"Aha," he said. "And why would you do that?"
This would be obvious to anyone who's ever worn a hearing aid, I thought. "Because it's too damn loud!"
DNR exists for one reason: noise is annoying. In fact, it's (been empirically proven to be) more annoying to those with hearing loss than those without; when you're used to an absence of background noise, putting on hearing aids can occasionally lead to the reaction of what the hell is that the sink the microwave the dog ah god cars passing ergh is that the air conditioning WHY DID I TURN THESE ON AGAIN and then the small devices get torn from ears and shoved in pockets so the listener can continue on in blessed, blessed quiet. I did this a lot as a small child; lunchrooms were a particularly awful variety of auditory hell.
Pediatric audiologists have a tough job; schools have awful SNR (signal-to-noise ratio) because of classroom floor tiles, hallways lined with lockers -- think about the surfaces of schools and how hard and reflective they are, then put in tons of kids making tons of noise that bounces off those surfaces. Kids are also more bothered by noise than adults. And so kids with even a slight hearing loss are statistically way more likely to repeat a grade, even if they're aided (I'm starting to pick up more audiology lingo -- they say "aided" and "unaided" instead of "wearing hearing aids" or not).
(A side note for my parents: The more I learn about this, the more I realize why everyone was freaking out about discovering my hearing loss immediately before kindergarten. Statistically speaking, I would have been likely to fall behind. I just... didn't.)
Anyhow! Noise is annoying. Noise is so annoying that one of the main criteria for determining one's candidacy for hearing aids is "how much noise can you tolerate?" also known as the ANL, "acceptable noise level." This isn't simply a factor of masochism; hearing-impaired listeners tend to need a higher SNR to understand speech in noise than normal-hearing listeners, and the required SNR increases as the severity of the loss goes up -- someone with a 30db loss needs a 4dB difference between signal and noise, someone with a 90dB loss needs around 17dB. But it does get into factors of individual cognition that are independent from one's audiogram. If, for whatever reason, you can't bear anything other than an impossibly high SNR, hearing aids will just frustrate you by making the noise louder ("increasing the BNL, or background noise level"), and you'll be likely to reject them outright. ANL is the predictive measure of hearing aid success.
Therefore, one great boon of DNR for audiologists is that it may drastically expand their potential market. It might make nonusers into users -- people who previously would have rejected HAs on the basis of their low noise tolerance might actually wear them now.
So what is DNR? As the name implies, it's digital -- so this is relatively new, taking advantage of a DSP that can run algorithms for "noise detection." Analog hearing aids can't do this sort of manipulation of the signal; they can basically only amplify. But if the background noise is constant, like wind, or an air conditioning unit, or the steady drone of a motor, DNR can find that noise and then -- poof! take it out of the equation. Improved comfort.
It's amazing stuff; my new hearing aids have it. Sometimes I'll be driving with the windows open, and I'll hit the highway and the breeze will turn into a howler, and then -- poof! My left ear will suddenly feel quiet, empty, and I'll have a microsecond of confusion before remembering that ah, that's the ear by the window -- it was getting wind noise -- DNR caught that and has scootched the wind noise out, and I'll notice my muscles relaxing slightly, imperceptibly, in relief.
And then tensing up again, because now my left ear can't hear the radio.
Which brings us to a problem with DNR; Digital Noise Reduction works by... well, reducing noise. Specifically, the frequencies of noise. It drops the gain down. Alas, speech and noise tend to be in similar frequency regions, so oftentimes de-amplifying the noise... de-amplifies the signal. DNR algorithms can detect noise, but they can't separate it from signal; it's like trying to get the peanut butter out of a peanut butter cookie, as Dr. Alexander put it.
And that's with a constant background noise (the wind whipping past my car) that looks very different from the signal (oldies music on the radio). Variable background noise is even more impossible, because how can you predict when noise will arrive in the future? And variable background noise of voices are worse yet, because then you're trying to filter speech from... well, speech. To carry onwards the peanut cookie analogy, we're now stirring two jars of peanut butter together and then trying to separate them out again. It's just... yeah. No. This is why we use directional microphones, which are the only technologies shown to improve speech intelligibility in noise as of today. They're trying not to pick up the noise in the first place.
Wait. Backtrack a little there. Did you catch the implications of the second-to-last sentence? "...directional microphones, which are the only technologies shown to improve speech intelligibility in noise as of today..."
That means DNR does not improve speech intelligibility in noise. It's entirely for comfort. Entirely. Which sounds, perhaps, like a wimpy thing to do -- hearing-impaired people whining about the auditory equiavlent of not having plush upholstered chairs with massage features.
Except it's not. When you look at studies that compare DNR-enabled to not-enabled hearing aids, you find that DNR makes hearing aid users less averse to sound (as measured by a scale called "APHAB" which I know nothing else about but now want to dig into, because I'm curious what "aversion to sound" means and how it's measured). DNR gives them more tolerance for noise. And it does this without degrading speech intelligibility; it doesn't help, but it doesn't hurt either. Moreoever, some of these studies were done in the real world -- users wore the hearing aids around their daily lives, logged sound environments, and so forth -- so the findings are from outside the lab.
It's not because hearing aid users are wimps. It's because hearing aid users get tired. The word to use here is "effortfulness" -- it's the opposite of "effortlessness," -- and the feeling will be familiar to anyone who's learned a foreign language, had to listen to a lecture while sleep-deprived, or has ever had someone try to talk with them when they're in the middle of something. What do you say in that last scenario? "Hang on, I can't do this and listen to you at the same time." In other words, wait 'till I'm done so I can concentrate on what you're saying.
People with hearing loss don't know what that feels like. We have to work so hard to hear and understand what you're saying that there's not much processing power left to concentrate on it. We are forever stuck in Russian 401, perfectly able to understand a native speaker rattling away at a normal clip, but desperately thinking gee, I wish this lecture were in English because then I could keep track of who killed what and married who when how where wait was that a fairy tale I'm lost now. (See why hearing impaired kids are more likely to repeat a grade?)
Except that, for us, the lecture never is in English. I am a native English speaker. That is what English lectures feel like. Don't even ask me what it takes to keep up with someone in Mandarin or German. Look at me when I'm trying to watch an uncaptioned YouTube video. My eyes, I've been told, are staring so hard that they look like they're trying to burn a hole in the screen. (This may not be too far off, actually. Pupil dilation is a measure of cognitive effort, and has been used to gauge cognitive exhaustion in several studies with hearing aids. Naturally, I want to get my hands on or cobble together a pupil dilation gauge now.)
There's an "Effortfulness Hypothesis" that says "hey, hearing-impaired folks have shifted their attention/cognition powers to the initial stages of perception, and so they have fewer resources to do the processing later on, like remembering things, or reacting in time, or whatnot." And so we don't remember things. And we don't react in time to get a conversational response in, so we seem "slow" or "quiet during group discussions." It's like requiring a kid to juggle, do jumping jacks, or watch an active toddler during math class; nobody would be surprised if that kid's math grade tanked.
(Combine the effortfulness that accompanies a severe-to-profound hearing loss with ADHD and you get a recipe for academic disaster. I am now gobsmacked that I ever made it out of high school, and very, very, very thankful for books. I'm willing to bet that without books, I would have straggled, probably repeated grades, not so sure about the whole "college" thing. As it is, with books, I skipped a grade in order to attend a magnet school, went to college on an academic scholarship, and am now mid-PhD. And it pains me to wonder how many people could have been like me, or better -- and fell through some stupid crack somewhere. If it sounds like I'm angry, I am: the restraint that keeps my energy tightened towards "coping with the world" is loosening a bit, and the high pressure fissures out briefly through that crack before I silence it again.)
In any case, cognition is a big deal. The role of cognitive effort in "hearing and understanding things" (there may be a better technical term for this, but I don't know it) can't be underestimated. Hearing-impaired folks need to concentrate and pay more attention when listening, and DNR might be able to reduce the need for that, freeing up their mind for other things. See, the weird thing about DNR-processed speech in noise is that it isn't measurably more intelligible -- your score on a word identification test won't change -- but it sounds better. The current theory is that you're offloading the noise reduction from your brain -- so you can do the same stuff, just with less effort(fulness).
In the short run, this isn't a big deal. People are capable of concentrating pretty hard for a few minutes (for instance, in Russian 401), just like they're capable of sprinting pretty hard for a few minutes. The question is what happens afterwards, hour after hour and day after day; is there a point where your brain starts panting from exhaustion? (Yes.) If you don't have to work so hard to mentally separate the signal from the noise, can you last longer? (Probably. When I got hearing aids this summer, I was stunned to discover that I'd actually listened to a complete lecture instead of zoning out and reading halfway through.) If so, this is called "improving secondary task performance" -- there's a lot of fascinating things to tackle around the area of cognition, auditory processing, multitasking, and education, both for those with normal hearing and for those without.
Of particular interest to my polyglot aspirations is an experiment by Marcoux et al in 2006 that showed improved novel word learning in adults using hearing aids with DNR. The same seemed to hold true for kids, but only after a certain age: 8-9 year-olds showed no improvement, but 11-12 year-olds did. (Language learning windows: fascinating.) In the other age direction, older hearing aid users may actually be using more cognitive power to compensate for what is often late-onset hearing loss; since they've already acquired language and the ability to decode contextual cues, the prediction part of their auditory processing engine might just be overclocking. (This is based on a study that tested the intelligibility of low and high context sentences in noise; normal-hearing and hearing-impaired listeners both young and old were able to understand high-context sentences once the noise was soft enough, and normal-hearing listeners were able to understand low-context sentences once the noise was softer still, but older hearing-impaired listeners never did get the low-context sentences.)
So that's the impact of DNR. Now, in order to understand how DNR works on a technical level, we're going to need to step back for a moment and look at the "Digital" in "Digital Noise Reduction" -- how does DSP work? How does it take and manipulate the signal to detect noise and suppress it and have all these nifty consequences in one's brain and in one's life?
This will be the topic of our next post.