So just to start with: Children live in a noisy world. And when I went to prepare this version of this talk about a year ago I went on to Google and I just did a quick search on the family meal, and this was actually the first one that came up. I know, right, they’re color-coordinated, all the vegetables are really impressive. What I wanted to focus on here is that this family’s doing a lot of things that we think promote good communication. So there’s no other, as far as I can tell, sources of competing sounds in the background, they don’t have the TV or other electronic media in the background, they’re looking at the child who’s talking. They look to be maybe slightly more engaged than is comfortable. But they’re involved. But at least in my house this is more like what the family meal is like. So this is an illustration from the Washington Post. And you can see we’ve got the TV on, the the teenager — this is a very good example at my house — a teenager’s looking at her iPad. We’ve got the dog — they do have vegetable so maybe a slight improvement from many days at my house. But this is really likely the better example of what listening environments are like for the developing child.

And then we break it down further. What we’re focused on a lot in my lab is what types of background sounds are in the environment, and how might they affect hearing and learning. And we’ve known for a long time, and there’s been quite a lot of study, about what I would call relatively steady noise in the background. And so this is a schematic of some sources of potential noise in the classroom, and it’s the things that we think about a lot when we think about our educational audiology classes, or classroom acoustics. So we have fans and we have fish tanks I think those are really out of favor now in most classrooms. We have noise from traffic outside, and the relatively steady state. And you can see just by looking at the time intensity wave form that these are relatively steady state. They do fluctuate over time, but not in ways that are overly dramatic just looking at it in this framework.

So when we think about steady noise, the way that we like to think about it in our lab, is it really impacts audibility. It impacts the fidelity with which the peripheral auditory system can represent targets sound. So if I’m listening to, say, Lauren, but there’s a lot of background sound I’m going to actually physically not — some parts of her message are not going to be physically transmitted to my brain because the background noise and her speech portions are overlapping on the basilar membrane. And my example here, we have the teacher who’s reading to her class, and we have an air conditioner. We call the teacher the target the air conditioner in this case is a masker, and it really physically obscures parts of the target in this case. In the literature this is often called energetic masking, and it’s the masking we almost always are dealing with an audiology clinic. So if you’re doing masking of PBK words, if you’re doing masking of tones, you know your traditional audiometric masking, these are the types of sounds that you almost always are using.

So the issue that I’m gonna focus on almost exclusively today is really what happens when the competing background sounds are speech. And this is a chart that’s taken directly from my first grader’s classroom. So he frequently comes home from school, and he’s been told to talk in a 0 or a 1. Which I love because he’s very loud, not unlike somebody he is related to. And this is a very common. So they have a lot of noise in the classroom, but most of it’s coming from the children speaking. And here’s another example.

As Karen I think did a really nice example of showing, if you look at the time waveform this is actually two competing sounds. They are summed and delivered to the peripheral auditory system. And you can see that they fluctuate. It’s very difficult to know looking from this waveform in particular what the target and what the masker is. And this is the focus of most of my talk, and if I had known that Karen was gonna do that awesome slide of transparent overlaid slides on top of one another — that’s informational masking. I love that example.

And the problem really for the developing child is not only do they, but people talk a lot. So there are emerging data coming out of our field and the field of speech-language pathology as well, they’ve actually tried to document how often children are exposed to competing speech. It turns out, most of the time. So conservative estimates would indicate that at least half the time children are in their home they’re listening to some form of competing speech. And when they’re away from home ,its most of the time. And my colleagues Sophie Ambrose and her colleagues at Boys Town also documented that TV and other electronic media introduce additional speech streams. So I think the estimate in their study was at sixty percent of homes in the U.S. have a television or some other form of competing speech on in the background most of the day. And the issue is that there are also data linking exposure to both noise and speech to academic outcomes and learning, especially language learning. And so this is a problem and something that we need to really get a good handle on.

So again I’m just going to basically do an uglier version of what Karen’s awesome slide was like. So when we think about this other form and what competing speech might do, in this case we have the teacher who’s the target, but now we have an assistant teacher and another student off to the side reading aloud. And they’re competing speech, and the issue is there is some competition and some overlap in the peripheral auditory system, but by most accounts at least in a typically developing child, the information that is sent to the brain should be sufficient for them to be able to parse apart the auditory scene. But as you can tell it’s it’s confusing. They’re similar, it’s hard to not only segregate what belongs to the target, what belongs to the masker, but also to stay focused and stay attended to the target. And we call that in the field often informational masking. So i put this in there because I’ve tried to do the talk without using some of the jargon, but if I get nervous or if I talk fast I’ll often include it so this is what I mean. So informational masking, one of the simple ways to think about it is what is the brain doing with the information that it receives from the peripheral auditory system.

I mentioned that children must learn about speech in the presence of all these different sounds, and one of the things that’s really important to know that I’m gonna sort of hit you over the head with a little bit, is that the ability to hear and understand speech matures at different rates for these different competing background sounds. And we think that’s because there are different processes that are developing that mediate these different abilities. The other thing — and if you fall asleep for the rest of my talk because you had really good ice cream at lunch — if all you kind of take home from my talk is this is a really important point. These central processes even though they’re not talking about the peripheral representation on the basilar membrane, these higher-order processes affect directly what you hear. So if you cannot segregate a sound, if you are not selectively attending to it, you don’t hear it. So functionally in some ways it may not matter where in the system this is occurring. But if children are poor at segregating sounds they’re not hearing the message. And I think that that is a really important take-home message to think about. So what a child, an infant and especially a young child hear in a background noisy situation is not what we hear — most of us hear. Maybe by the time we get to Karen’s participants we’re doing a u-shaped distribution.

Now I’m going to talk a little bit about, mostly work that we’ve done in our lab, to really get a handle on: first, how well do they hear, and when in development are they adult like in these different backgrounds. So this summarizes probably about a decade of work from our lab and other labs. And as a general rule a lot of our paradigms use some form of speech recognition. In this particular study I’m going to show you a study that I did with my colleague Emily Buss we looked at forced-choice consonant identification. So the children had a touchscreen monitor, actually they held what looks like a little iPad, and they were just asked to touch the speech sound that they heard, and the background was either a steady noise like a shhh or it was two competing talkers. And I want to use this as an example, but just to let you know people have done, including our lab at many other labs have looked at constant identification, word identification, open set word identification, sentence recognition — the results are the same. So I’m showing examples that I think kind of highlight the main points, but this has generally been a consistent and overwhelming finding in the literature.

So if we think about children, in this particular study we measured consonant identification. We measured it at a fixed signal-to-noise ratio of 0 DB signal-to-noise ratio, and we looked at performance in two different maskers. And we looked at a wide age range of children. And you can see here on the x-axis that I looked at performance for these for age groups, and that’s somewhat an artificial distinction but this was a pretty big N study. And when I started looking at some of these effects in children we used to test five to ten year olds, and then we realized, well the ten-year-olds weren’t mature. So let’s include up to 13. And now we are including children up to the age of 16. And part of our fear was that we’d start seeing that the 18 year olds who are normal hearing sort of adult controls weren’t mature. But in fact that is not the case, so we think that we see most of the maturation around 16, fortunately. What’s shown here is percent correct identification. Just to orient you, higher up means better means you identified more consonants than lower on the y-axis. And if we look first at speech shaped noise one of the things you can see is a trend that many other people have observed in the literature: That young children maybe five, six, even seven year olds, often perform worse. They need a higher signal-to-noise ratio to perform as well as adults. But their performance was worse in the steady noise. And that’s a fairly consistent finding, these are normal hearing children. But you can see if you look at the eight to ten and eleven to 13 year old age groups they are more or less mature. What was really interesting in this study, and I think it’s pretty obvious, if you look at the filled bars, not only were the younger children substantially worse relative to the adults — so it’s a really pronounced age effect they’re really surprisingly bad at this — but you can also see that even the 11 to 13 year old age group were not performing in an adult-like manner, and that’s again, many studies have replicated this with different materials. So the ability to perceive speech in a speech background we think matures sometime after 13 or 14 years of age. And that’s in contrast to the ability to perceive speech in sort of a steady noise background.

And I wanted to show these data these are from Nicole Corbin who’s this wonderful PhD student works with Emily Buss and I. And it’s not just that the effects are bigger, but the whole time course of development doesn’t look to be the same for these two different recognition abilities.

This was a study that looked at open set monosyllabic word recognition. So kind of like PBK words, but we had this corpus of 800 words that were within the lexicon of a five and six-year-old child. What’s shown here this time is we did an adaptive procedure. So opposite of the figure I showed you before. We measured the signal-to-noise ratio where that listeners needed to get about 71 percent correct recognition. So lower is better, it means you could handle more background. And again we use noise and speech backgrounds. And what you can see, these are data in the steady noise masker and what you can see it’s sort of a general trend for improvement with age, with the youngest children showing an increase in susceptibility. If you look closely at these data, the line doesn’t help because it makes you look like, it makes it look like there’s a real linear trend. But if you look at the data for children probably about eight years and older, they look really indistinguishable from the adults. And there’s a lot of variability.

In contrast, these are data from, I think Nicole tested like 55 children, which is a lot of children to test, and what we see in the two-talker masker is really a different pattern of development all together. So we see kind of a really gradual, that’s a significant age effect but the slope is more shallow. And what was really interesting and we honestly don’t know what is responsible, is we see these effects around puberty we think where performance just drops to adult like levels. And that was something we hadn’t expected. We thought we would see a linear trend, and we actually wanted to fit functions that let us get an estimate of when you see mature performance, but we’ve talked to some people about what about what might be going on, but we think this is an interesting trend. But fortunately for us 16 year olds do look mature, so all of my previous studies where we used 18 to 24 year old adults as our baseline for mature performance I think are still okay.

The other thing that supports this idea that what’s going on when we ask children to perceive speech in a steady noise versus a two-talker masker or speech masker, is that they are uncorrelated. That the performance in one the threshold in the noise, is not correlated with performance in the two-talker from the same children. So that’s really interesting, and I think was surprising to us. We had some level we thought we would see the kids who were bad in one condition were bad in the other, but that’s not true. And so you can see the r value here is not very high, it’s not a significant correlation.

I was going to say before I transitioned to the slide was that one of the things we’ve seen repeatedly from our studies of school-aged children is that they’re much, they show these greater susceptibility effects when the background is two streams of speech, something very similar to the target speech, than in steady noise. And these data really though give us maybe a different insight into the problem. So we did a study that looked at infant speech perception. And rather than looking at recognition, obviously, we looked at just speech detection threshold. And we looked at 7 to 11 month old infants and adults and we also looked at school aged children because we wanted to get an estimate of how they performed using our traditional infant procedure. And in this particular example, if we look first at the open squares — so again we have an adaptive threshold so lower means they could tolerate more background — if you look first at the two-talker speech we really see the same effect that we saw with the recognition performance. And in fact the age effect for both this detection task and the recognition task I showed you in the previous slide, if we look at the average effect between the children and adults at the same age was a 7 dB effect both studies. Which is somewhat encouraging that we’re tapping what may be the same abilities. But you can see that for the school-age child– but you can see that for the infants they’re considerably worse. This ability to detect speech in the presence of other talkers, infants are really poor at.

What was surprising to us is that the children — the school age children — they look like adults in terms of speech detection thresholds, again not too surprising. They were a little better than they were at recognition and detection, and that maybe isn’t all that surprising. So they were able to do these higher order processing that we’ve heard from the other speakers. But what’s really interesting to us is they were no better at detecting speech when it was broadband noise, and that’s really kind of thrown us for a loop in terms of what might be going on. And our current explanation is that they have not yet at this age learned the ability to determine what’s target what’s background, and they tend to as a general rule we think try to integrate and listen to everything. And so these are some ongoing studies, we have some thoughts on doing a study where the talker is actually their mother or their father, and there’s some indication in the literature that we will see those familiarity effects pop out. And also, when in development does this ability mature? And we can them — not really two-year-olds, we can certainly test three and four-year-olds in the lab pretty routinely now. And so those are some of the questions that were really interested in looking at.

Okay, so why? Why are they so poor as compared to adults at these tasks? And this is really reiterating some of the mechanisms in processes you heard about from the earlier speakers. Most of study developmental effects in speech perception agree that at least some of the problem is a failure just to segregate the target from the masker. So if you’re an infant or a young child, how do you know what the target talker is and what the background talker is. And how do you perform that task, what information do you use? We also think they have unrefined and undeveloped selective attention ability. So they might be able to use some of the same primitive or early cues that Dr. Sussman mentioned, but if they can’t then focus their attention on the target and disregard the masker it it’s not going to show up they’re functionally going to not be able to do the task as well.

And finally this is an area that really, we talked with colleagues who know a lot about these issues more and more. But the role that just general cognitive processing plays, so things like working memory and other executive functions and how do those interplay? Is this an auditory specific developmental process or is this just a general cognitive processing maturation. And we don’t know the answer to that.

So I’m going to stop for a second and give you a bit of a model of how we like to think about the process. So this is sort of my theoretical framework. For Dr. Chatterjee over here. Whenever I give a talk — Monita is one of my colleagues — if there’s not a hypothesis or a really clear theoretical framework, we have to stop and do that. So just an important note for those of you on the RMPTA if Anita is ever going to be in your audience — or any of us really, it’s a good thing to do.

So here’s sort of a really simplified model of how we like to think about development. So as Barb mentioned, we have these sources in the environment and they all sum before they arrive at the ear. So in this case we’ve got the teacher, we have the assistant and the student, and we have an air conditioner. So all kinds of different sounds, and they arrive at the ear of this developing child. And one way we like to think about this early stage is: what’s the fidelity with which that information is represented in the sensory system. So I’ve represented it here like a spectrogram, but really that’s what we’re getting at. How is the timing, the frequency, the intensity of information — is it sufficient to even do the task. So in the case of typically developing infants and children, the answer is yes. By all accounts the development of the peripheral auditory system is precocious. I wrote here by at least six months of age we think this peripheral representation is mature. I think more recent data from Carolina Dalla’s lab would indicate probably three months. So certainly the data I’m showing, the issue for a typically developing child isn’t their peripheral auditory system. It’s sending the brain a really nice — maybe a bit fuzzy at times when they’re young infants, but the representation of sound is thought to be more or less adult-like very early in life.

So the brain gets sent this information, and the question is what does the brain do with this information. We wanted to reconstruct the auditory scene so can segregate each of these sounds that arrived at its ear into three separate objects. And then if so, what about trying to selectively attend to one while disregarding the others. And so those are the mechanisms that we spend most of our time trying to parse apart slowly. And it’s tricky and kids of course because often we don’t know which of these processes we’re looking at because it’s really hard to get an infant to tell us how many streams of soundly heard or what they’re attending to. But there are some methods we think that we can employ that are getting us closer to trying to parse the specific mechanisms apart. And basically the punch line here is that the ability to use information the brain is sent from the sensory auditory system takes years of experience and neural maturation. So we’re talking not until adolescence.

So one of my favorite things to talk about is, okay so we know that they’re very poor in general at recognizing speech when other people are talking. So for a long time in my lab, we wanted to look at — well, if we introduce a cue that we know adults rely on to segregate a sound, whether that’s space or common onset time, or whatever that sort of acoustic cue might be. So one of the things we’re really interested in: At first we were interested in, okay, they’re really poor, so can we help them. And now a lot of our studies are really looking at the segregation types of studies as a way to understand what the fundamental mechanisms that are driving children’s performance are. So one of the initial studies we did in this area was done in combination with Lauren Calandruccio and Emily Buss, and we wanted to know whether children take advantage of differences in targeted masker sex. So it’s been known for many years that there’s a number of vocal characteristics between men and women, and that seems to be a really powerful segregation cue. So if you measure adults speech-on-speech recognition and you have the same sex, its matched you might have two men. That’s much harder than if the background talkers are a different sex. So if you have a male talker and a female background. You tend to get what we call a huge release from masking and it’s a very robust effect. And so we wondered, well do children take advantage of that information as well. And so the data here are shown again, their adaptive thresholds — so I didn’t put the arrows on any more from this point on — but lower is better. This shows data in a two male-talker masker. So it’s two men talking in the background. The blue circles are matched, it means that the target was also male. So everybody’s male. So you can see that those blue dots are higher than the red triangles which we then flipped and we made the target a female. So we have age as a function of years, and you can see that kids performed more poorly than adults. Universally. On both conditions. We’ve known that for a long time. But they do benefit from this sex mismatch, and just to prove to ourselves that this wasn’t due to some unique stimulus combination we did read the experiment with the same target stimuli, but we used a two female-talker masker and what you can see is the effect flips.

So this is evidence — and these are really big effects if you look at the difference between the two sets of symbols. As other people have observed for adults. The task is a four alternative forced-choice picture pointing response where even a 3 dB effect I think is a pretty meaningful effect. And we’re seeing effects on the order of 10, 15 dB. So it’s a whopping effect really. And so that’s encouraging that they use some of the same information that adults use to segregate sounds.

So what about babies? I told you that they’re really poor. This is really promising, if we knew that this benefited infants, this might have some implications for what types of environments and how we want to structure. Well it’s very surprising to us, we did a study that were just writing up now. And in part we started this study in 2013 and we’re under like our seventh experiment to convince ourselves that this is actually how infants behave. But it is. In this particular plot what I’ve shown are group data. They are infants and adults, the infants on the left panel, the adults in the right panel. We measured a speech detection threshold. So this is just: can detect the sound or not? Lower again is better. And what you can see the masker is a two female-talker masker, the red would be a female target talking. You can see that — and I have the legend backwards so it should be adults and infants, which I think is a bad error. So adults were better — so replace infants with adults. One of the things you can see that the infants were the same in whether it was a male or female target talker. And that was true for individual data as well. See what I did in the next one — yeah I did the same thing, so reverse it. We see it flips the other way around.

So these are really hot off the press data. These are Mary Flaherty’s data. And one of the things we wanted to look at then — so we did the study where we had a male and female target mismatches, and one of the things we wanted to do is really get at this a little bit more fine detail. So we wanted to look at what part of male and female speech were children and adults relying on. So what we did is we manipulated fundamental frequency. So we took a male talker, and he produced both the target and the masker speech. And we tested children with original speech — so in this case it’s the same target talker just like what Barb showed, really hard. And I’ll play that. So this is going to be a monosyllabic — word no it’s a disyllabic word in a two-talker masker. All the same talker. I’ll warn you it’s not one of the four pictures shown here. So see if you can hear what the target is. It’s really hard though, and it’s hard for adults. Then what we did is we took the same target stimuli and we shifted them with this nice fancy algorithm. So this is a, I believe this is a nine semitone shift, so this is a large shift. And see if you can hear what the target word is now. So the masker stays the same but we move the fundamental frequency of the adult higher in frequency.

What we’ve shown here is these are thresholds so their speech recognition threshold as a function of age, and what you can see if you look at the difference between the original — which are shown in blue — and the red I think the really interesting finding here is that young children, really children under 10 years of age, they don’t benefit from a 9 semitone separation. Which was really not what we expected. In fact this was a precursor study to a study that we want to do in listeners with hearing loss as children because we think that maybe if we introduce FO separation difference we can run that through an FM system, and that’ll be really helpful in natural environments. But the problem is of course if their typically developing peers don’t benefit from it, it’s unlikely that the children with hearing loss that that would be an effective strategy. So again these are the same talker, so we tried to control everything but FO, and there’s lots of work to be done. But I think this is a really interesting initial result.

One thing I’ll point out is I’m not showing these data here, but Mary’s also collecting data on children with mild-to-moderate/moderately severe hearing loss who wear hearing aids. And these children can represent this fundamental frequency difference. It’s a really hit you over the head difference. It’s a low-frequency cue. And we’re not sure if it takes them longer to use this cue or not, but certainly the older children with hearing loss sort of in the 13 to 14 year-old age range, they do seem to benefit so that’s really encouraging. But we’re just starting that study. Did I represent that correctly Mary? Okay.

Our working hypothesis

Followup and future work