 Tena koutou katoa. Good morning and Slamar Pagi. I'm Gregor Byrne from the University of Canterbury and I've been asked this morning to talk to you about trends in speech perception testing. What I'd like to focus on are the ways in which we can make the tests that we deliver in speech audiometry more rapid, more accessible and more equitable. So speech perception testing is a major research interest of mine. I've developed several speech tests both in New Zealand English, Tareo Maori and also with the help of Dr. Seifu Jamaluddin in the Malay language. But in recent years there have been numerous advancements that have been aimed at making speech perception testing more rapid, more accessible and more equitable. So when we say rapid, essentially it's ways in which we can decrease the amount of time that's required for a speech test and one of the ways in which we can do that is to increase the amount of information we get from each test. If we can get more information from a single test we don't have to repeat it multiple times in multiple different ways. What do we mean when we say we want to make a test more accessible? Essentially we're trying to remove those barriers to the availability of speech testing and part of that can be to do with the cost of the test. That also counts as an equity issue. If we can make a test more affordable then it's available to more of the population and that reduces the inequities that we have in health provision. Another way we can make tests more suitable is to improve the suitability of the test and the test material to the test participants. Most of the tests that we have developed form part of a single software package called the UCAS, the University of Canterbury Adaptive Speech Test Platform. You can see the various buttons here representing the different modes of delivery of a number of different tests that we've developed. The UCAS FW is an adaptive filtered words speech test for the diagnosis of auditory processing disorders. The digit triplet test as many of you will know is a hearing screening test which uses numbers in noise and the auditory visual matrix sentence test is a sentence in noise test that can be delivered for diagnostic speech audiometry purposes in a number of different formats and you'll hear more about those from from Seifle and from other in the next day or two. Okay but first what I want to do is focus on the matrix sentence test and the ways in which we can improve the suitability of that test material for the people who are taking the test. So we know that speech recognition measures are a fundamental component of the audiometric test battery and they provide valuable information regarding the the functional impact of hearing impairment on an individual's communication difficulties and it gives us information that isn't relayed by the audiogram. So an audiogram itself gives some clues as to the way someone will perceive speech but in a lot of cases there's quite a poor correlation between the pure tone audiogram and how someone performs in a real-world environment in in a noisy environment for example. So we developed the University of Canterbury auditory visual matrix sentence test in New Zealand English that's another key thing is we need to develop these tests in the language of the people who are taking the test and the accents and was with the goal of giving us an accurate portrayal of the communication difficulties that people encounter in real-world scenarios. So just as a bit of background matrix sentence tests were first developed by Hargamon for the Swedish language and they consist of syntactically correct but semantically unpredictable sentences. So they're in the format of a name, a verb, a numeral, an adjective and an object. So that's that's the order. In English obviously in Malay the adjective and the object are around the other way and we've had versions created in multiple languages worldwide. Down the bottom here you can see Bahasa Malaysia which was that version of the test was developed by Dr. Seifel Jamaluddin as part of his PhD and he'll be talking about that test a little bit later. I want to focus on the New Zealand English test for the moment. This is what it looks like. So you can see we've got five columns here and each column has ten options for the particular word and you can put them together to form a sentence. Amy likes six old books. Oscar sees 12 good shirts. And so you can see those words get pieced together and they form a sentence. And because we've got five columns each of which contains 10 options, we've got 10 to the power of five possibilities which is 100,000 possible sentences. So we normalised this test for use in constant noise, which is noise which has the same speech, same spectrum as the underlying speech, babble noise, which is multi talk of babble, and also we normalised it in quiet so that you can deliver that test as as just a standard speech in quiet test. The response modes that are available are in the open set response mode, which is where the participant hears the sentence and they just repeat back what they thought they heard, or the closed set mode where you actually select the options that you think you heard by pushing a button on a screen. So this is what the panel looks like from the software. So you're presented actually with an array of 50 buttons that you can see here. And when you as the person taking the test hear the sentence, you then just using a touch screen, for example, push each button that corresponds to the word that you have heard. So let's give that a go. So that's Thomas likes three good books. So obviously I'm very familiar with this matrix. And so I knew where to push those buttons. But what we find is that people who are taking the test do require a little bit of practice. It usually takes about two practice lists for people to get the hang of where the test items are on the screen. So for the items that involve words, we list them in alphabetical order. And the ones that involve numbers, we list them in numerical order, as you can see there. So that middle column there is in numerical order, all of the rest are in alphabetical order. But despite that, it may take a little bit of hunting. So if I find the word Thomas, I look down and there's likes three goods. And then by the time I've gone to the last word, I may have entirely forgotten what that final word was the word books might have fallen out of my memory. If I was just repeating it backwards in the open set mode, it might be a lot rap more rapid as I'm not having to hunt for those items on the screen. So obviously that my ability to perform that test doesn't just depend on my ability to perform speech perception in noise. But there's also the working memory and cognitive capabilities of finding these items on the screen and holding what we've heard in our working memory long enough for us to to find all of those options and to push those buttons. So how demanding is that task? We want it to be able to be performed by people across the age range. And it imposes different demands on different age groups. So obviously young children aren't going to do as well with this test because it is longer. And because it contains more words. So that just simply cognitively and where they are in their stages of language development, they may not be able to perform that test. Now one one thing that is often cropped up with people who are elderly performing this test is we find that their ability to perform such a test decreases with age due to cognitive and working memory demands. So the five word matrix sentence format, particularly in that close set response mode, we find it works well with most adults, but it may impose significant cognitive demands on some participants, particularly children, but also the elderly. So how can we improve the suitability of tests like that? One thing is that we have to take into account those effects for cognition and working memory. So we know that if we put hearing loss to one side, forget about your peripheral auditory sensitivity and the functioning of your cochlear, the ability to perceive speech in quiet doesn't really change all that much with age. It's more a function of your hearing loss. But older listeners do have more difficulty perceiving speech in difficult listening conditions such as noise. So that's you can have two listeners with the same auditory configuration, but different ages. And the one that is older may have more difficulty perceiving speech in noise. So when we're applying speech tests for audiological purposes, we have to think about why we're doing it. What is what's the purpose of this test? Usually where we're using it as a proxy for peripheral hearing sensitivity. And we're not actually interested in those confounding factors such as cognition and working memory. We've got separate tests for those if we want to test those things specifically. So when we're doing speech audiometry, it's usually for the purposes of determining how well someone understands speech. And because working memory is highly correlated with language perception, particularly in adverse conditions, we should ensure that the test material that we're using is within the capability of the older clients that we're testing. So for this reason, in a number of different languages, simplified matrix tests have been used. And these work by reducing the size of that 50 word matrix in order to reduce the cognitive and linguistic demands of the test and they can be used for both children and for elderly clients. So previously, these tests were released under the name of the as a pediatric version of the test. And in most cases, that that's what they are primarily used for. Some of them do that explicitly by use of picture pointing tests, for example. But it has been found that these have been quite useful also for the elderly clients. So some example of simplified matrix sentence tests, we have the Aldenberger kinder's test or Kisa. Now the word kinder, that means child. So it's explicitly saying it's a pediatric children's speech test. And what we do is we use three word pseudo sentences. So not full sentences, but just a number and adjective and an object. And that test has been validated for use with children over four years of age. So an example of a pseudo sentence that they use here is think they sublumen, which means five white flowers. And other simplified matrix tests have appeared in all sorts of European languages as well. So in Finnish, French, Italian and their examples there. The Polish one is slightly different. It is again, if you're a child over seven, then they use a verbal response mode. But they've actually constructed the test so that they can use a picture pointing method for children aged between three and six. And as tomorrow, we'll be telling you a bit more about the introduction, sorry, giving you an introduction to the Malay pediatric sentence test that we've been collaborating on together. OK, so how did we create our New Zealand English matrix sentence test? Well, we took our five by 10 word matrix and took out, firstly, the first two columns. So we got rid of the names and the verbs that left us three columns. And then we reduced the number of options from having 10 options for each to down to six options of each. So we ended up with a three by six matrix. And when we chose those words, the way we did it was by prioritising the words with a lower lexical difficulty. So they appear in children's vocabulary at an earlier age. They had better slopes. By that, I mean they are more suitable for in their psychometric properties for speech testing. And we also chose ones that required less editing because we not only did we choose the words based on the existing adult matrix sentence test, but we used the same recordings. We're able to just slice up the recordings in a slightly different way to to just isolate those three words. And so anything where there was a conjunction between the second word and the third word that made that difficult, we removed those items. And so this is what our matrix looks like in the end. So you can see example sentences might be two small hats, nine old shoes, for example, 10 old books, eight red spoons. And so with those set of 18 words, we're able to get 216 possible pseudo sentences. So that's six to the power of three. And we identified which of those sentences were most suitable for being incorporated into our test by evaluating the psychometric properties of 162 of those sentences. From that, we're able to generate a series of lists that had relatively similar performance between them. So we're trying to make them as homogenous as possible. And we did that using an iterative computer program to to maximize the homogeneity of the the list. So making sure they all had the same average SRT and also the same slope. So that's very important to ensure that they all behave similarly. So how does it look? Well, we evaluated these lists and we did that with 43 normal hearing adult participants. So these are all young adults with normal cognition and normal hearing. And what you can see is that their behavior is really quite homogenous. So this is in the open set test mode. And we found a mean speech reception threshold of minus 9.5 dB signal to noise ratio averaged across lists with a standard deviation of only 0.5 of a dB. And the mean slope was actually very high. It was around 15.5% per dB, which is which is quite high. So if we compare this to the five word version, the adult version of the test, not only is the standard deviation of the SRTs in both of those tests the same, but so are the slopes. And in fact, the slope for the three word version is marginally better. When we look at the closed set response mode, we get similar results. So here we've got the standard deviation of the SRT is very low. You can see those lists are pretty much overlying each other. It's only about 0.4 of a dB and the mean slope is 11.3 dB and that corresponds again very favorably to the five word version of the test. So what this means is that these tests are not only suitable for use with adult clients, but they perform just as well as the ones with five words. So that has actually raised the question of us do we need that five word version at all. Now there are other advantages to having five word sentences and that is that it gives us more scoreable items per sentence and that increases the the speed with which we can arrive at a certain answer. So when we're performing this test in an adaptive mode, if we've got five words in every delivery, we can get a more precise estimate of the speech reception threshold in a shorter amount of time than if we're just using the three word version. So how do these advantages translate for use with children? So one of the things we wanted to look at was how the maturation of the auditory system and your vocabulary knowledge as you age how does that affect your performance on this test and what ages can we perform this test with. So we examined the effect of age and maturation on the SNR thresholds in a number of normally hearing children in schools around Christchurch. So this we had 144 children age between 5 and 13 and these are the results that we get. So here are the results from the ages of 6 to 12 in terms of the average speech reception threshold for a monorail stimulus delivery and what you can see is that there is a steady increase in the child's ability to do that test until they reach about age 10 and then from the ages 10 to 12 we have a plateau of performance. So the child is improving in their ability to perform this test between 6 and 9 and then once they reach 10 the results basically plateau at that point there. So to sum up how can we ensure that the test material is suitable for an individual first we should be testing the people in the language that they are most familiar with because in adverse conditions particularly in noise for example the effect of the native language actually becomes more significant. So it's really important in a particular country to develop versions of tests not only in the majority language but also in minority languages. We also need to ensure that both the test material and the response mode is suited for the language skills the level of cognition and the level of working memory of our participants and the lesson here is that simplified versions of tests can often be just as sensitive and also that that open set response mode may have advantages on the closed set response mode that requires you to hunt for options on a screen. Okay so back to our speech test now I want to talk a little bit more detail about the digit triplet tests that we have developed. So the New Zealand hearing screening test is the name that we give to the the digit triplet tests that are in New Zealand English and in Torreo Maori. For those of you who aren't familiar with a digit triplet test cycle we'll be giving you more information about that later but just very briefly you hear a set of digits so three digits presented in background noise so here it is in New Zealand English for example the digits 391 you enter 391 and in Torreo Maori we've produced Te Whakamatoto Whakaromo or Aotearoa and this is what you hear and you enter 095 or kore iwa and rima and the Torreo Maori digit triplet test was actually the first that we have that was produced in a non-European indigenous language. Right so um Cypher will be talking to you in a bit more detail about the Malay version here's a here's a quick preview and you enter Kusung Dua Antiga okay so as Cypher will explain more this afternoon the test is an adaptive one so it uses a simple one up one down adaptive procedure to find the level of signal to noise ratio in which the person scores 50 percent when we are scoring at the triplet level so here's the adaptive track and what we do is disregard the first seven responses which is where they're getting down to threshold take the average of the remainder and you can see here that this participant score is minus 14 dB signal to noise ratio so before we go ahead with the testing we have to ensure that the stimuli are normalized and by that we mean that all the digits are equally difficult and that helps us maximize the sensitivity test sensitivity of the test and then we can go ahead and validate it so here are the results for the New Zealand English version so what you can see here on the x axis is the main better ear hearing thresholds between 250 Hertz and 8 kilohertz those who are at 20 or below on that measure are shown here in green and those that are above that measure are shown here in red and on the y axis we can see the score that they achieved the decibel signal to noise ratio for the digit triplet test and the aim in producing a test like this is to choose a cutoff on the y axis here it's that black line which best separates those who we want to pass the test from those who we think should be a refer on the test and there's a technique which is to develop a receiver operating characteristic or ROC curve that you can see here in this panel and we want to be in the top left hand corner of that of that plot and that indicates that we have high sensitivity and specificity and with a particular cutoff for this test we achieve a sensitivity of 94 percent and a specificity of 88 percent so that's great I mean this means the test is working how well you know is working really well and is separating those people who we should pass the test and those who refer so we have released that test a number of years ago since 2017 it's been licensed to a private hearing aid chain within New Zealand called Triton Hearing and they have implemented the test on sort of stands that you can see here where we've got an iPad on a screen and a set of headphones you might be waiting at the pharmacy for your prescription and in the meantime you see that over there the the test stand you put the headphones on and you can test your hearing and so these are through hundreds of pharmacies throughout New Zealand they also take it on roadshows for you know fairs and show days and things like that and set up stands for people to come along and have their hearing testing tested so it's proven very popular we've also created an online version it's marketed as the great big hearing check and each year around 10,000 people take that test so it's you know getting out there and it's being used really well we're also trying to get the New Zealand MOLRI version of the test released as well so in order to do that obviously we need to validate it and compare it against people with varying degrees of hearing loss and one of our students James Dawson is undertaking that process this year so here you can see some year 10 MOLRI students who are trying out that test but with all of these people taking the test we have had a chance to look at large numbers of data and what we find is that even though the test is is very short when you are testing both ears one at a time and particularly in the Triton version of the test they have to enter demographic information such as their name and their email address beforehand so they can get a copy of the results those times add up and what it means is that some people actually just give up during the course of the testing so they'll be in the middle of the test and maybe the their prescription will be ready from the pharmacist and they just stop and leave or they may get through one ear and then can't be bothered testing the other ear so there's this general feeling that it maybe even though it is short it's taking too long and some people aren't completing it so that's obviously a bad screening outcome if someone is giving up on the test and walking away that means that we lose them we don't have their test results if we can improve the their stickability if we can make sure that they hang around and complete the test then that actually will achieve a better screening outcome and the way in which we can do that is to make the test more rapid somehow how can we do this well this has been on my mind for some time and so in the literature there are a number of different approaches that have been put forward and which are being adopted in various places so one way you could do this is just to stop the test early so standard digit triplet tests present 27 sets of triplets and the reason there are so many is to achieve a high test retest reliability some of those have now experimented with using a slightly shorter number but whenever you reduce the number of trials it usually increases the amount of uncertainty around the result so having shorter number of trials means we're not as certain of the person's result as we may wish to be and that has impacts on the sensitivity and specificity of the test one example that I came across was the Royal National Institute for Deaf people in the UK and they had a version of the digit triplet test available as a screening tool on a website they've since replaced it but this particular version had an interesting quirk when you tried it out it wasn't a quirk it was just a method of giving you a shorter test result normally the test takes a number of averages so if you had poor hearing it may go out to the full 27 trials if you had a reasonable amount of hearing so if you were sort of on the borderline you might stop after 18 trials but if your hearing was clearly quite good it might stop after only 10 trials so it had a mechanism built in so this data you can see here is where I was trying to reverse engineer what was going on I recorded the speech samples and measured their levels and tried to calculate what algorithm was underlying the test because this algorithm wasn't published as far as I can tell but it is clear that you have different stopping points for different degrees of hearing loss so the rationale is well if you were down here at this particular signal to noise ratio you couldn't really achieve that if you had poor hearing therefore it's safe to stop the test so because again as it wasn't published the effect that this has on overall test accuracy is unclear but what is visible now if you go to the RNID website is they use a different digit triplet test so different recording and it uses the standard protocol of 27 trials so I'm not sure the reason for that but they have gone back to to a more traditional test format so one thing I mentioned with the matrix sentence test was the difference between the three word item three word sentences and the five word sentences and the fact that if we go to five words we increase the number of scoreable items and that is because there's a relationship between the number of trials that we take and the precision of the estimate and if we've got more scoreable items we have a more precise estimate of hearing threshold so typically the more trials you do in a test such as 27 trials or cutting it down to 18 or to 10 the more trials we have we have a greater degree of precision and the reason for that is that those test retest errors the way we measure that reliability they're inversely proportional to the square root of the number of trials and as I said the more scoreable items if we've got a more precise method of estimation that actually means we need fewer trials because we're arriving at a more precise estimate in a shorter space of time so examples for this are when we're doing CVC word testing a standard word list we choose phoneme scoring rather than word scoring because now we've got three scoreable items the consonant, the vowel and the consonant rather than just the word as a whole similarly if we're doing monosyllables we have we actually end up with more scoreable items if we use a sentence rather than just a single word in our testing so as I mentioned typically digit triplet tests are scored at the triplet level that means you've got to get the whole triplet correct you've got to get three three out of the three numbers correct in order for the result of that trial to be a correct so that's a binary classification it's a zero or a one correct or incorrect so even if you get two out of the three digits correct so you've done that job mainly right that's still an incorrect why do they do that? Well the advantage is that it increases the steepness of the psychometric function and there have been numerous studies that have demonstrated that the standard deviation of the SRT estimate is reduced when the slope of the psychometric function is steeper so that's the reason we typically use triplet scoring and that has largely been implemented across the board with digit triplet tests however periodically people come back and revisit that issue see if they can improve the functioning of digit triplet tests making them quicker by increasing the number of scoreable items through digit scoring rather than triplet scoring and the most recent example is this is a team from Belgium led by Sam Dennis who looked at digit scoring for the digit triplet test so if your score on digits instead of just a zero or a one your possible scores could be zero one out of the three two out of the three or three out of the three what they did was vary the step size of that adaptive procedure instead of just two decibel one up one down steps they adjusted it to match a specific target on the psychometric function that was similar to the triplet scoring method and what they found interesting was a higher significantly higher test retest reliability with this method and because of that because of those more scoreable items there was no need to change the SRT cutoffs but they were able to achieve a 37% reduction in the number of trials while achieving the same test retest error as the triplet scored version so instead of needing 27 trials to get a result they were able to stop the test after just 17 trials and so that's quite a big improvement there Another way in which you can improve the sensitivity of the digit triplet test is by avoiding the need to test both ears individually so of course you can just deliver the test binaurally to both ears the danger in those cases is that if you've got a sensory neural hearing impairment in one ear but very good hearing in the other ear you may still pass the test because it's your good ear that's doing all the work so to get around that a lot of digit triplet tests test the right ear and the left ear separately and of course performing the test twice takes twice as long because it's a speech in noise test it consists of a speech signal the digits presented to both ears and noise that is also presented to both ears so here you can see the noise which is at a constant level and we have the speech signal this one here says the digits and then three digits that you can see there and they are at a variable level and that's what we're referring to in that adaptive track so this signal is presented the same to both ears but there is another way in which you can improve the sensitivity of the digit triplet test and decrease the test duration and that is through the use of antiphasic stimuli so antiphasic means that the phase of the waveform is flipped in one ear than the other and that operates quite differently in the auditory nervous system and it causes it to stand out from things which are in phase so here you can see if we have a look at this time domain waveform here if you pay attention to the shape of these waves these are clearly in phase you can see this rising portion on that last digit there if we flip the phase of one of those channels they are now antiphase and a funny thing happens when you perform a digit triplet test where the noise is in phase but the signal is antiphase and that is displayed here so here is a standard a similar to the plot that I showed you for the New Zealand version before where we have normal hearing people are shown in black and then various types of hearing impairment are shown there with different colors so unilateral or symmetric sensorineural hearing impairment and also conductive hearing impairment and you get a certain spread and your task is to draw a line through through that data on the y-axis to find a cutoff which is going to give you a good sensitivity and specificity and that's shown here as that red dashed curve on the ROC curve that you see there but have a look at the advantage that is given to normal hearing listeners when you present antiphase stimuli so look at this this is the same y-axis we get a much bigger spread of data so instead of all being bunched up the normal hearing listeners come down and we get a spread of hearing impairment as it spreads down there and this increase that spreading out increases the separation of normal results from those with unilateral or symmetric sensorineural hearing impairment or indeed those with conductive hearing losses so when we look at the ROC curve for this antiphase stimuli you get the blue line that you see here and that is much closer to that top left corner it shows a much more much increased sensitivity and specificity and again because we're not testing each ear individually it has halved the test time we only have to perform this test once and we get excellent results in that way so that's one approach that is significantly improving the performance of the digit triplet test and it's being implemented in a number of different territories most notably South Africa All right so I want to talk to you about a relatively new published method which yields vast increases in test speed and that is to switch from an adaptive digit triplet test to one that utilizes a single fixed signal to noise ratio so as I mentioned digit triplet tests are adaptive speech tests they present a number of trials about 25 to 27 triplets and you calculate a signal to noise ratio in DBSNR that is for that individual that's their speech reception threshold for digits in noise so here you can see we've taken 27 trials to achieve the right amount of precision to say yes we're confident that that is their specific SRT and then we compare that SRT to the normative data to decide whether they have passed the test or whether it's a refer you can see here we get a pass result or a refer result depending on whether their SRT is above that signal to noise ratio criterion or below it so once you've gone through the trouble of figuring out what that cutoff should be once you've established those criterions it may be possible to improve the efficiency by switching to a fixed signal to noise ratio version of the test now Cas Smiths who was the person to popularize the digit triplet test and release a telephone version of the test in the Netherlands he released this paper a couple of years ago which proposed a completely different method of hearing screening using digits in noise so what you what you can see here is the number of trials that are being presented is on the x-axis on the left hand panel and that straight line is just a one for one line so this is five is five 10 is 10 15 is 15 and so on and on the right hand side you can see for different decibel signal to noise ratio thresholds so if you've got a particular SRT how many trials does it take you well no matter what your threshold it's going to take you 27 or in in this particular case 25 trials to arrive at a decision so no matter if you've got excellent hearing really poor hearing or something in between in the standard adaptive mode it takes the same number of trials 25 trials in this case but what you can also do is calculate the cumulative probability that someone is going to pass or going to fail when you present a signal a single signal to noise ratio repeatedly so what you do is you choose one particular signal to noise ratio that's very near that criterion threshold that we mentioned and then we keep going until we achieve until we're 95 percent certain that that person is going to fail or pass the test and so what you end up with is lines that look like this so as you're going along you present a certain number of trials and if for example when we get to five trials if you have scored all five of those five trials correctly then we're more than 95 percent certain that you will pass the test overall at that criterion signal to noise ratio and so we can stop the test and say you've passed similarly if we've done those five trials and you got none of them right then it's when 95 percent certain that you are going to fail the test you're going to get less than 50 percent at that criterion value and so we can stop the test at that point and say that you are a refer so that's a clear example there where if you're hearing is very poor we know after five trials that you're going to fail the test if you're hearing is very good we know after five trials that you're going to pass the test what about those people in the middle well they have to continue on and we deliver more stimuli so let's say we get up to 10 stimuli now after 10 trials if you've scored eight of those 10 trials correctly then again we can stop the test and you have passed if you've only scored two of them correctly then again we can stop the test because we're 95 percent certain that you would fail and we can follow the rest of those lines like that and if we look at the distribution of the number of trials that it takes to make that decision with the same amount of confidence what we see in modelling is a 67 percent reduction in the number of trials from 25 to around 8.3 on average so you can imagine that's a vast improvement in the speed at which that test was delivered but how does the sensitivity and specificity stack up? well actually the pass and refer rates for the fixed signal to noise procedure the very short one is nearly identical to the one that takes 25 trials no matter what your SRT so this shows a great speed advantage for testing the only downside is that we don't actually end up with that number which is the person's SRT so we can't tell by how much they've failed or passed the only information we may have is that they stopped after a short number of trials or a long number of trials but regardless of that if all we're interested in is saying do you need further diagnostic testing or not then this is a great way of going about that so yeah really an interesting method and I'd be interested to see if people adopt this on a mass screening scale we spoke about language and the importance of using a language that is correct for the person so that's why we've got digit triplet tests in all sorts of languages but another approach is to not use speech at all and I shouldn't really be talking about this in a speech audiometry talk but people have experimented using ecological sounds so one test that was implemented in the Netherlands was called the sound ear check and that's developed as a language independent automated self-test that's based on recognition of ecological sounds presented in noise so in the presence of background noise can you tell the difference between a bird chirp and a bell ringing and a telephone and a baby and a piano and a car horn and a cat for example how well did it perform? Well it was actually found to be less sensitive less specific and less reliable than the digit triplet test in its ability to detect mild hearing loss but if you have to test someone who doesn't speak a language that you have a test available in then this is maybe an option worth pursuing in the short term one thing I'd like to talk about now is the use of digit triplet tests with children so the DTT has been trialled for school age hearing screening in Flemish children so it's the Flemish speaking region of Belgium in children age nine to sixteen and those children were previously tested using pure tone hearing screening but they made the shift to test first with the digit triplet tests and if they fail that test if they get a refer then they receive an audiogram so those two different age groups the fifth grade elementary school and then the third grade secondary school the older children they had two different referral criteria two different SNR cutoffs if you like so in the Flemish speaking population of Belgium they actually have screening across a wide range of ages so not only neonatal and preschool but also that school age screening that uses the digit triplet test and also testing at entry to school five to six years of age so that's a very large number of test points for hearing screening they did try to use the digit triplet test for five to six year olds but it was found to be unreliable and so they still use pure tone screening for five to six year olds so why is that? Well the answer is that if you use a DTT on young children they have a limited attention span and the test durations actually work out to be quite long so that's eight minutes for a single ear in a six year old and so they just can't sit through that test so we could do other measures to reduce the number of trials for example but again that's going to reduce the precision of the single to noise ratio estimate but one approach that is proving quite successful is gamification of the digit triplet test to increase engagement so that team of researchers led by Astrid von Vieringen and Jan Voters have developed what's called the pirates game and this is a digit triplet test where six year old children open pirate chests by entering a three digit code so it's like they've got a padlock and they hear they get told what those magic numbers are for the padlock and instead of just pushing the buttons they have to choose the right numbers on the padlock and that could open up the treasure chest and the results have shown that it enhances the children's motivation and is able to sustain their attention such that they're able to complete the digit triplet test and get and get more reliable results another approach to gamification has been taken with the Sound Scouts app produced by Carolyn Me with the collaboration of the National Acoustic Labs in Australia so here's some screenshots that you can see there it's actually a composite of a number of different listening games they're calling them but some are speech in noise and others aren't as a variety of different tasks that they see there and the aim again is to differentiate normal hearing kids from kids with a hearing impairment and it's proven so successful that the Australian government has actually just paid for that test to be free for all Australian school age children between ages four and seventeen can download and use this app for free people outside their age range or people in other countries pay a small fee but it is free across the Australian school age population which is fantastic so again it's a successful method of testing young children through gamification so their aims were to produce a game that was engaging and fun for children to play that detected hearing problems and differentiated between sensorineural hearing loss conductive hearing loss and auditory processing disorders did they achieve those aims well yes it was fun and engaging they didn't measure that specifically but the anecdotal evidence was that it was fun and engaging for children to play did it detect hearing problems yes it did so if you combine age adjusted scores from the three tests that are part of there one of those tests is based in speech and quiet one is based on speech and noise and the other is based on tones and noise you can combine those scores to produce a composite metric and that reliably detected children with a four frequency average hearing loss that was greater than 30 dBHL in either ear did it differentiate between sensorineural conductive hearing losses and APD actually no it did not and there's a variety of reasons why that may be the case but further work is indeed in progress okay so what's next for our particular suite of speech tests well we're doing a number of things to ensure that those tests are rapid accessible and equitable so work is indeed continuing so in terms of improving the speed and sensitivity of our speech tests we're re-establishing the digit triplet test normative data using anti-phasic stimuli that should halve the duration of those tests how can we improve the accessibility well we're working with a programmer at our university Robert Fromont who is really fantastic at creating web interfaces for things and we've developed web-based delivery formats for both the digit triplet test and the matrix sentence test which is very exciting for enabling the release of that to clinicians who may be in hospitals or remote locations who may have a phone or a tablet or a PC you can use the same test on same test platform on any of those devices what about ensuring equity well again with tests in different languages where we're carrying out the validation of the tereo Maori digit triplet test this year working with indigenous communities to make sure that we're making a test which is suitable for their needs and we're also next year developing tereo Maori matrix sentence test both the standard five word version and the simplified version but I'm going to start with the simplified version I think just because we've shown that it is very successful in a range of circumstances and also as I will talk about tomorrow work with Seifle and other on the Malay pediatric matrix sentence test so I just want to briefly mention that this work isn't possible without the help of a very very large number of students and collaborators if you look down the bottom there you'll see an old photo of Seifle who worked really diligently on creating the Malay versions of the digit triplet test and the matrix sentence test so I'd just like to thank a large number of people particularly Seifle Azar Nerlin Sarah Rahmet all my collaborators in Malaysia I really enjoy working with you and and hope to do so more in the future so thank you very much