Christer Volk - Participation is Key
Welcome to "AigoraCast", conversations with industry experts on how new technologies are transforming sensory and consumer science!
Dr. Christer Volk is a full-time specialist at SenseLab with a PhD from Aalborg University in Denmark with a thesis entitled “Prediction of Perceptual Audio Reproduction Characteristics.”
Christer’s tasks include: helping clients formulate test requirements, designing listening tests for perceptual audio evaluation, analysing data, improving SenseLab’s proprietary software SenseLabOnline with special focus on the statistics module, contributing to Audio Engineering Society conferences, and research activities in government-funded projects.
Christer is the current coordinator of the Psychoacoustics group in the Danish Acoustical Society and he is also a reviewer for both the Journal of the Acoustical Society of America and Acta Acustica.
Transcript (Semi-automated, forgive typos!)
John: So, Christer, welcome to the show.
Christer: Thank you very much for having me, John.
John: Oh, thank you for being here. So something that you and I were talking about before the call here is that in sensory, I think that audio sound, you know, audition, however you want to refer to it, is underappreciated, like it's a very important modality and very little. I think it kind of typical sensory science isn't thinking much about sound. There's this huge bias, I think, towards visual perception if you're going to pick between visual and audio perception. And I think it would be really helpful to start the call if you could just bring our listeners up to speed on what you think sensory scientists should know about sound and about audition.
Christer: Yes. Yeah, I think luckily sound is coming a little bit back because of all the new tests with virtual reality, so it's becoming I think the sensory scientists are becoming more aware that this is something that you can include. And if you do more realistic tests in a context, then sound can actually had quite a lot to that. So I think the trend is going to a point where when sound is becoming important again. So that's very nice. I do, however, still think that this may be a chance to improve a little bit on how sound is included. So, of course, this is a new area for many sensory scientists so it's a learning process. But there is yeah, it's still a very casual way of using sound in most of the studies I've seen so far. So first of all, it's important to note the level of audio and this is a small thing, but since our auditory system is so non-linear in every possible way you can think of. I mean, it's difficult. And that means that you could get different results if you play at a low level and at a high level. And if you want people to be able to reproduce your results, you need to report the level.
John: Right. And for people who know nothing about sound really like me, so level it simply means volume? That's the volume at which the sound is being played.
Christer: Yeah. And this is a bit of a requirement, actually, because you can just use your phone to measure this. You need a special sound level meter to do so. And it is possible to get an app that can measure it. But usually those apps are quite inaccurate. So at the very least, you need to talk to someone who can help you calibrate your phone once in a while. So if you are within, let's say, plus minus three the best, then probably you are fine with using something that's not too expensive. But having the sound level is important.
John: Okay, so what would be the other topic? So there's kind of two pieces to this, right? Because you have controlled environments like a virtual environment or maybe some augmented environment where we control the sound levels, right?
John: But then there's also there's this push towards ecologically valid testing, right? Where you try to do testing in certain environments. So, your research, is it focused more on controlled environments or are you doing research that's in more of these kind of natural environments where you're taking measurements?
Christer: Well, traditionally, it's been very much controlled environments. So a good example of where we are trying to, you know, still have this ecological validity but doing everything is in the lab is that we are testing a lot of headphones with the noise cancellation that's becoming very popular in a lot of headphones and headsets right now. And we do that by having the headsets mounted for recording on head and torso dummy in our listening room. And then we have 30 loudspeakers positioned on a sphere around that subject and playing an environment. So this could be from a train station or restaurant or a busy road or concert, something where you kind of get this real feeling of being in a natural environment. And sometimes we want to hear that environment if we're testing hearing aids. But if we're testing headphones with noise cancellation, then we want to test whether the device is able to suppress that background sound or noise and maybe also whether the sound quality is affected by this algorithm that's suppressing the noise, which often happened. So, you can say, the first step in getting to a more natural situation. And furthermore, we also do sound works where we go out and we go to a place where something new is about to happen that will affect what we call the soundscape. So that would be that there's a new road being built or a new building or rail station or something that will have an impact on the environment, the audio environment. And we use local people to rate, how is this environment right now? I mean, both the positive, can I hear the birds and maybe the negative? How loud is the road? And could also be how lively is the atmosphere and how noisy is the children playing on the playground next to it? So in this case, it's a completely natural environment. And the only thing we try to control is that we know town which time of day is you could say representative for the situation. And that's something where it also helps a lot to use local residents because they can say, oh, this state is actually very quiet. We need to go a different day or now, this is completely as expected. And they will also have an expectation for how the environment should be. So if they're used to this noisy highway next to them, then maybe that's not a big focus for them. It's more this new thing that's coming that they are not too happy about.
John: Interesting. So, is this the Danish government that brings you involved? I mean, who would be the kind of client? The people building the building or the local government?
Christer: Yeah. That would be the municipalities who are interested in understanding the soundscape and understanding how we can make this change into either something positive, at least something that's not negative. And these days, it's actually becoming a viable thing to do because there's a lot of building happening to handle all these big amounts of water and all these environmental changes that we are seeing. So we're getting flooded streets and we are getting a new sewer systems for that. And that's quite a lot of changes these days to have kind of climate safe streets. And when you are taking up everything, then there's money to actually do something while you're doing it and change the environment. And I think also the impact on animals is becoming more of a thing when they're constructing roads or I mean, it needs to be an environment that works for not just people, but animals as well.
John: Fascinating. So this is really consumer research involving sound to kind of help ensure quality of life as these kinds of changes are happening.
John: Well, yeah, that's fantastic. So what are some of the other areas of research that you're involved in? What are the things that are kind of the most interesting that you feel like, you're working on these days?
Christer: Yeah, I think smart products is something that we have a big focus on. So, smart products is the Google loudspeaker or the Amazon speakers that are everywhere these days. And they are really everywhere these days and that will continue for sure. And they are actually quite a challenge to test because they are so adaptive. So usually when we are testing a device, we know, okay, we are testing at this volume. We are testing with these soundtracks and we are testing with these settings. And that means that you can make a very neat design of experiment and have a nice test and you can do statistics where it's easy to understand what's going on. And you kind of know that the results that you get is representative for using this product. And that's something that's a challenge for us because everything with audio is in direct evaluation when we are talking about product from audio, from products reproduced audio, because you can't test the loudspeaker. You have to test it with some music or some sound or so you can't, it's not like food science where you can taste the food directly. Here, there’s a big layer of having good samples, good music to test your system with. And if that is not representative, then your data and your results will not be representative. And that's a challenge in itself. But with the introduction of smart products, it's becoming an even bigger challenge because they will adapt to the environment. They will adapt to the position. So if they are near the wall, then they will adapt the output. And a lot of them will adapt even on every what volume setting you set, it will have a different frequency response or different response in some sense. So a good example is that you have a small speaker that wants to play a lot of bass. But if you turn it up to a very high volume, then it will start to have a lot of distortion. So it's common for these products that they turn down the bass a little bit if you turn the volume up. So when that happens automatically and it's not telling you what it's doing then how do you know that you're testing in which situation? I mean, sometimes you can't even report which situation you actually did your tests.
John: Right. You have the same problem with headphones, right? You have adjusting headphones that don't report what they're doing.
Christer: Yes and especially with the wireless technology. I mean, they have all this communication going on between them about what's the sound quality that we will use for playback. And all of that can change quite a bit what the user hears. And that's not reported in any way for, you could say the scientist to see what was actually agreed on. What was the technology that's used while we are recording or evaluating. So that's a big challenge for us. And in many cases, we can get a lot of help from our client if they are the ones who developed the product. But if we are doing bench-marking where we are testing a client project, a product against a competitor's product, then we can't know anything about the competitor. That's a challenge, especially if you want to make sure that we are setting up a test in a way where it's good both for the client project, but also for the competitors. I mean, we don't want to give any unfair advantage to any product. And that can be a challenge when you don't know exactly how every product works.
John: Right. I suppose that is a problem we do have in food science where you don't know the formulation of the competitors’ products but you do know the formulation of your products that maybe analogous there.
John: Very interesting. Okay, so maybe we could just take a step back here just for the people who don't know you to talk a little bit about your background because it's always interesting to hear how people who are in sensory got into sensory. I mean, it's a field that always seems like, it is very rare to find two sensory scientists who have exactly the same background. We all end up here through different twists and turns. So can you take us through the history of how you ended up in sensory?
Christer: Yes. Yeah. So I did a bachelor in electronics and computer science. And for my bachelor product project, I wanted to go out in a company and kind of do a collaboration. And I ended up in the Danish hearing aid company, Oticon. And that's kind of how I got into sound and understood that was something I wanted to continue working on. And that got me to the Technical University of Denmark, where I did my master's, and that's where I really got into acoustics. And they have because there's three very large hearing aid companies in Denmark and they are funding this department quite a lot. They have a big focus on the perception of sound. This is touching a little bit on the sensory side, but we're still not really there. And while I was a student, I started here in SenseLab as an expert listener. So that means I was part of the train panel where we went to extensive training to understand test methods and understand attributes. It's not really as I would say, as far ahead as in the food sensory science because we are still a very small business. But the methods are very, very similar and all of them are, I would almost say, stolen from the foods science.
John: Inspired. Inspired by food science.
Christer: For sure, we owe the food science a lot. But after starting as an expert listener, first of all, I realized that this was a very interesting area and once you dig into it to this experience where it's something you can't really measure, then you realize how many aspects just a couple of headphones could have. I mean, how many things can be different about them. And finding out how to describe that is, I think, so interesting. So this is where I understood that the sensory science was something that was very, very interesting. Unfortunately, there's no education to become a sensory scientist within audio, at least not in Denmark. And I'm not sure if there is anywhere. So, I kind of learned by becoming more and more a part of SenseLab. I think I'm very lucky that we have some good people in SenseLab. One of them is Nick Zacharov, who wrote the book Perceptual Audio Evaluation, which is kind of, we call it the Bible, but maybe we are biased by being in the same department. So for us, that's where you go if you want to get started in sensory science within the audio world. So I've had some very good teachers and that really helped a lot. He taught me basic statistics and then I took a single PhD calls in sensory statistics as well. So I got a little bit of the statistics part which is also I think the most interesting but really it's learning by doing and I think that there is a strong sensory science network in Denmark. And so I've become part of that. And I think I'm learning a lot from seeing what they do in the food science, which is very big in Denmark. I think with the many universities working in this area and also big companies like Arla Foods.
John: Right. And then Aalborg University, of course, is a sensory research.
Christer: Yes. Also Copenhagen University.
John: Right. And so then while you're at SenseLab, then you finished your PhD and that you were working concurrently at SenseLab when that happened?
Christer: Yes. So we have industrial PhD's in Denmark which means that I didn't have to teach, but I had to make a project that was very much in sync with both what they wanted in SenseLab, but also what they wanted at the department I was in at Aalborg University. So that's actually quite tough because the company wants something that they can sell tomorrow. And the university, of course, wants something that's actually new research. So that's a tough one. But at least it's only three years in Denmark. I know it's usually a lot more in the US.
John: Yeah. Mine was six including the masters, but I did surf a lot. I probably could have done it faster. Anyway, you know, so much said for surfing. So anyway, well, that's fascinating. So then that was four years ago, you finished your PhD? So we've talked a little bit about some of your research interests. What are some of the other research interests that you've been focusing on the last few years?
Christer: Yes, we've moved into machine learning recently, and that's kind of a good topic for me because it's a continuation of what I did in my PhD. I did the perceptual modeling of audio attributes. So to see if I could predict the perceived base strength of a loudspeaker or the perceived treble strings, some kind of simple attributes, and see how they would correlate to any measurements we could do. And I think that's quite an interesting area, because what we see is that the companies we help are very, very late to getting in touch with us. So they kind of have a final product and they want to make sure that it's good enough. And that means that we were unable to help them early on in the process. I mean, it's rarely that we can help them in the development process because everything is so secret. So they don't want to share it.
Christer: So that's why we are going into machine learning, because if we can make a small tool that can you know, maybe not perform to the level of listening tests or real something customized for exactly what they need to know, but still can kind of point in one way or the other. Then that’s something that could be very, very interesting. And I think a special area where that's interesting is that we also do tests with the hearing impaired subjects and we see that this is a major problem in the hearing aid companies that so many of these algorithms are being tuned by normal hearing people. I mean, there's not enough people with a hearing loss out there to hire us as engineers for developing. And even if you're one person, the hearing loss that you have will probably not be representative. So this is actually it's very difficult to know. Should I go this way or the other way? So there are so many choices and if we can make a machine learning tool that can kind of help point in one way or the other, then that could be helpful tool. And I think as an academic area, it's also very interesting. It seems like all the scientific rigor that have kind of been described over such a long period in research is kind of all of it is coming together. I mean, I love that machine learning conference is so open that you publish on the same data set. How good a model could you do? How good was the performance? And there's a ranking in a school board and all the code is open source. All the papers are open and it's completely transforming how fast the scientific community can actually make progress. And that's kind of also this, I think, great idea that, okay now there's a new problem and we want to solve that. And it really doesn't matter who solves the problem, we just want to solve it. So let's have a conference where everyone is competing about solving this problem and then move on to the next. So that's great.
John: Right. The open source ethos.
Christer: Yes. So unfortunately, it's also very, very difficult area to get into, I think.
John: Well, yeah. I mean, I would say that the tools for entry level machine learning are becoming more and more accessible. I mean, you and I both program a lot in R. You've got tidy models package which I think makes a lot of data processing and a lot of the kind of hyper parameter like tuning. Yeah, a lot easier. So I think that part is becoming more accessible, which you're right. The high end stuff is, I think getting harder and harder. And I mean, if you're talking about models, neural networks that are 50 layers like Google or Microsoft have, I mean, there's no you know, there's only a few companies in the world that have enough data and enough processing power to train those networks. So, yeah, it's an interesting situation in that to some extent it's been democratized. And on the other end, it's you know, there are some capabilities that are only really possible for a few select companies.
Christer: Yes. And the advantages are so big. I mean, we have a model called Peak that's used quite a lot in testing, for instance, headsets. So this is a model developed many years ago and it's trained for predicting telecommunication devices and the quality of those. And it was the state of the art model when it was made and it was tested on what was at that time. A big headset. Before real machine learning, I would say that was a big data set that it was trained on. But I mean, these days it's being used for so many other stuff than what it was trained for. And it's kind of a quality stamp to get a certain rating from this model, even if that rating isn't really true. And that’s becoming part of getting products, maybe not this model, but others like it are used for kind of getting the stamp that this headset is good for Zoom or this headset is a team's validated headset. And if those quality measures aren't really true to the perceived quality then everyone will build towards the target that's not the optimum. So I think that's a really big benefit from getting more advanced models and also may be getting models that are easier to train for new purpose. But, yeah, getting the data to do that is a big problem I think. We can't do like the big players are doing better where they're just spending endless streams of data. I mean, conducting an experiment even for, yeah, whether it's audio or food or chemicals. I mean, it takes time and you can't get millions of data points easy.
Christer: And even though we have millions of data points, it's in so many different types of experiments and different scales and different questions. And, you know, you have talked about using historical data and they are still valid when you try to use them. So, I mean, you can't spend too much time collecting data either. So it's a challenge, but a fun one, I think.
John: Yeah. Hearing a lot of my favorite topic here, knowledge management is of course a big topic, right? Because like you said, we have a lot of information, but it's in all these different forms. How do you bring it together into some comprehensive, useful form? That's a big, big problem. And it's not one, honestly, that, you know, the Google and Microsoft, they're collecting a lot of data that's in digital form to begin with. And it's coming in a very consistent way, you know. I mean, I'm sure that they are solving very hard problems also in knowledge management, but they have an advantage that, you know, the Google search information is, you know, how many searches are happening every hour? You know, millions, if not billions of searches and they're all coming in more or less in the same format. So, I mean, they're getting a lot of data consistently formatted all the time. And we have to do a lot of work to set up the systems for our data to be formatted consistently. Well, Christer, this is fascinating. You and I could easily talk for another half an hour and probably another several hours but we do need to wrap it up. So I always like to conclude by asking people for advice. What advice do you have for young sensory researchers? What do you think that people should be thinking about the next few years here?
Christer: Well, I think starting with the audio people, I definitely think that they should start to take some classic sensory science courses. And it's not something that is easy to even hear that this exists. So that's the first problem. But really, I've read lots of books from the food sensory scientist and they are accessible to me. So you don't need to be in this field to really, really benefit. And so I think that's fun and a good way to do it. And then I think I've learned so much from just signing up for every experiment I could participate in. I mean, you get to test so many different technologies and the types of methods and I mean, that's so helpful. And if it's an area you're in, then you can easily ask about the technical stuff once you've participated. And everyone love to have people volunteer for the experiment and lots of students are not able to pay for it. So they are so happy when you do it and you learn so much from doing it. I think that’s a main advice. And I think for the food scientist, I think I am getting a bit more into the audio and how to play it back. What kind of environment you need is the normal booth with the low enough noise and also I think,yeah, really, really listening and making sure that it sounds good to you. I mean, if you can validate by measure so much but with sound. It should feel natural when you're listening to it yourself.
John: That's fascinating. Yeah, that actually takes us into something which I think there's a big area as augmented reality and 5G become more and more commonplace. I see a world where, you know, when you open a brand name drink that the speakers that are in your ears full time will start playing some music or something that some background, you know, sound that will help you to enjoy the drink more. I think that we're going to have those kinds of things happening where, yeah, brands will be able to offer a multi-modal benefit that will not be restricted just to the product, but the product will be complemented by some sound experience that helps you to appreciate the product. It's really exciting, actually. I'm really glad to see that research happen.
Christer: Yeah. Absolutely.
John: Okay, Christer, this has been fantastic. So thank you so much. How can people get in touch with you? Are you on LinkedIn?
Christer: Yes. I'm on LinkedIn and I have a very unique name so easy to find.
John: Okay, we'll put the link in the show notes so people will be able to find you. Yeah, well, hopefully, yeah, some people will reach out to you, I mean if a student wanted to collaborate with SenseLab, do you have internships?
Christer: Yes, we have. Mostly we do help with the projects so people can get in touch for bachelor projects, master projects. We have a PhD student right now. And even if people are brought someplace, if what they do is interesting to us, we could also offer to have a listening test with that material and that set up their methods, everything. So we are eager to collaborate if we can learn in the process.
John: Okay, great. And your clients are typically headphone companies, people make loudspeakers, hearing aids, this kind of thing?
John: I see. Okay. Excellent. Okay, Christer, thank you so much. This has been a really interesting show for me and I appreciate you being here.
Christer: Thank you.
John: Okay, that's it. Hope you enjoyed this conversation. If you did, please help us grow our audience by telling your friend about AigoraCast and leaving us a positive review on iTunes. Thanks.
That's it for now. If you'd like to receive email updates from Aigora, including weekly video recaps of our blog activity, click on the button below to join our email list. Thanks for stopping by!