(upbeat music) - The Data Skeptic Podcast is a weekly show featuring conversations about skepticism, critical thinking, and data science. - So welcome to another episode of the Data Skeptic Podcast. I'm joined this week by Ryan McGuire. How are you, Ryan? - Doing very well. - So I recently discovered your work that I thought was really cool and we're gonna talk about one particular aspect of that today. But I kind of want to just share a bit of my perspective on it before I ask you your background. You know, I learned about this interesting project you did related to the song Thomas Diner. And at first I said, oh, this guy's a musician. He's doing a little bit of dabbling in computer science, but then I saw you're writing Java, Python. You've even done some Lisp. So clearly, you know, you're also a computer scientist and I saw you linking to Lily Pond, which is this great project that I love and I wish more people knew about. So you're obviously, in my opinion, both a musician and a computer scientist. - Well, that's very kind of you. - Yeah, how do you describe yourself to other people? - I would say I'm an artist and a technologist and I don't know where the threshold is for being a computer scientist. And you know, probably in some definitions I'm not, but my undergraduate degree is in physics actually. And then I've sort of picked up the computer science, the computer programming along the way. So I'm, you know, I'm a programmer and a researcher and an artist and I kind of try and juggle those hats. - Yeah, tell me a little bit more about your background and your academic interest. - Let's see. As I said, I guess I started studying physics when I was in college, but I've been a musician my whole life. I think it was my sort of senior year of college during my physics degree that I discovered Philip Glass and Steve Reich and sort of 20th century experimental music. And I discovered that composers were still people that existed, which is something I didn't know. (laughs) And so that sort of precipitated this sea change in my life and I ventured out into the unknown world of music composition. And so I went back to school, I moved out to Boston and went to the New England Conservatory and studied music composition there. From there, I was trying to find ways to bridge my interest in math and science and programming with my music background. I discovered this really amazing master's degree program at Dartmouth College called the Digital Musics Program. And at the time, there was a composer, Larry Polanski there, who was doing a lot of really interesting algorithmic composition work. And then another researcher named Michael Casey, who is just a really, really human being doing work at the intersection of music and technology. And so I went up there and I studied with them for two years and got a whole crash course in all the different ways that people are dealing with this intersection between music and technology and a lot of different approaches. From there, I decided that I kind of enjoyed being in academia and that it might be nice to become a professor at some point. And so I started looking for PhD programs and found this one that I'm in right now at the University of Virginia, which is a PhD in composition and computer technologies. It's one of the only interdisciplinary PhD programs in this field that really has strength in both of those areas. And so that's sort of my academic background. And I think I just have a general interest in music and technology and different ways that those two fields are able to interact. And the new insights that you can glean from the intersection of those two areas and the new sounds that are possible and all of that good stuff. - You have any thoughts on the work of Alan Lomax by chance? - Oh, yeah, he's, I mean, I think really a pioneer of using technology and getting out there and recording. And I think what he did was a really great service. I have a lot of colleagues that probably that are in, we have another program here that's critical in comparative studies and they might have some more critical things to say about some of that. But I don't know, that's a little beyond my research. - Yeah, maybe I should have led with more of an explanation. Alan Lomax, I guess, and you might define this better than I would, but an archivist in certain ways tried to capture a lot of the early music's cultural and non-western music's that weren't being captured at the same rate as technology developed. Would you say it's a fair characteristic? - Yeah, absolutely. And I knew about him because of his connection with Carl Sagan's and his work on Cantometrics, which I've always wanted to do a project on. So I just, I'm always curious to hear what people closer to the music side of things think of his work. - Oh, interesting. I didn't know that he had a connection with Carl Sagan. That's very, it's really interesting. - He, basically, you know, the Voyager record? - Okay, yeah, uh-huh, he was on the spacecraft. - Yeah, he was one of the main people that helped pick out the, you know, and basically Carl asked him to help to focus on how much diversity they could get on the record. - Sure. - 'Cause, of course, Bach belongs on there, but he's like, so does, you know, these Tibet and chanting groups and things like that. - Right. - So we're gonna talk about one of your specific creations, but before I got into that, I thought some good background might be to cover a little bit about MP3s. So I think listeners all vaguely know what an MP3 is and that it's a compression, that there's some loss involved that makes the audio file smaller, but I think that's actually where a lot of people's knowledge ends. - Right. - So I don't know if you could share some more technical description on what exactly is happening when we go from an unfiltered, uncompressed version to an MP3 version. - The human auditory system has some really interesting properties which are exploited by MP3s. And MP3s are one of a whole class of coders known as perceptual audio codecs. Codec is short for coder, decoder. Basically, we have this dynamic frequency range from about 20 hertz to 20,000 hertz. And so we hear sounds from very, very low sort of elephant sounds all the way up to these piercing dog whistle sounds. And across all of those frequencies, we can hear sounds over a varying field of magnitude. So we can hear very soft sounds and very loud sounds. So it seems like the human hearing system is really powerful and wide ranging, but that's actually not entirely accurate. There's a lot more sound vibration happening beyond what we can hear. Additionally, the hearing system is really adaptive. And so the things that we actually end up hearing and perceiving are really based on what kind of audio environment that we're in. There's this really famous thing called the cocktail party problem, which is how when you're in a noisy room and there are a lot of people talking, you're not completely overwhelmed by the din of noise taking place, you're able to focus on one person and you're able to separate out sources, different sound sources in this noisy environment. And that's kind of an amazing feat that our perceptual systems are able to do. There's also limitations to that. So in the presence of white noise, for example, when you're in a noisy environment, weaker sounds get masked. So that's why, you know, when you're walking down the street, talking to somebody and like a truck comes by, making all this noise, it's hard to hear them because the frequency content from that noise covers up the sound content that you're trying to hear. There's different characteristics of masking and some of it's due to like your physical ear and some masking characteristics are due to processing that happens in the brain. And so MP3 is really take advantage of this whole idea of masking. I'd say the simplest property to exploit is called the absolute threshold of hearing. What that means is that we just, we can't hear sounds that are too weak, that are too quiet. And the absolute threshold of hearing, it sort of varies with frequency. You know, curves have been approximated from different experiments that have been run on a bunch of subjects. Another really easily coded and exploited masking property is simultaneous masking. And that's based on critical bands. Critical bands are, you're hearing, it works sort of like a non-uniform filter bank. Critical bands can be said to approximate the characteristics of those filters. It's like you have a bunch of resonant filters in your ear, and when you hear sounds, it makes these different filters ring. And the bands don't really have on and off frequencies, but rather they have a width that's a function of frequency, and so those are called critical bandwidths. Simultaneous masking occurs when there's two sounds, or more sounds, I guess, that are in the same critical bandwidth occurring at the same time. And so when you have two sounds that are in the same, that would occupy the same bandwidth, the quieter sound sort of ends up just vanishing, really, you don't really hear it. And sometimes this effect can also spread to neighboring bands, and I think that gets exploited in some audio coders as well. In perceptual audio coders, the most important masking properties of the ear are they're modeled and combined to produce these masking threshold curves. According to this model noise that's under the threshold, this completely should be completely inaudible to listeners. So you have these curves, and a sound is analyzed, and compared to these masking threshold curves, sounds that would be underneath the masking threshold, you can reduce the number of bits that are used to represent that information, or even in some cases, like deleted entirely. Ideally, most listeners won't notice that that information has been reduced or deleted at all. - I hear a lot of audio files kind of complaining about what poor quality MP3 is, and how distorting it is of the original audio signals. - Sure. - On the flip side of that, I'm probably listening on my phone, on a busy street with some embarrassingly cheap headphones in my ear. - Yeah. - Where do you fall in on how much of a sin it is to have our music store as an MP3? - Well, you know, I'm a sinner, so I listen to MP3's on an almost daily basis with my cell phone, and my earbuds, as I'm walking to class, or walking to work, or whatever. It's funny, we oftentimes, when we're listening to music, in a way, we don't really even, we're not really listening, we sort of just wanna be reminded of a song. It's, you know, you listen to a familiar song, and you're walking, and you're doing other stuff, you're driving, and you hear the song, and you're hearing parts of it, but more, the thing I think that's more happening is that it's just, they're reminding you of the song, and then it brings all these pleasant associations, and you have this sort of physical memory of the song, and so I think that that happens a lot, and in those situations, MP3's are great, and they're totally fine to remind you of something, and to give you that pleasant experience of like hearing a song again. And even hearing music for the first time, too, MP3's do a really good job. I mean, that's why so many people listen to them, is that they're pretty darn good. It's sort of an amazing feat of engineering. - Yeah. - I think maybe the sin would be to not be aware, but I don't know, maybe ignorance is bliss, or something, I don't know. And there is a lot, there is a lot of information that's lost when you compress an MP3 file, and it is audible if you really sit down and listen, if you were to sit down in the recording studio with the recording engineer that recorded your favorite song and listened to it in the studio, on the monitors there with full fidelity, and then you went out and listened to it on your earbud headphones on the subway, you would really, I mean, the difference would be really considerable between those two experiences. And so I think more I just would feel bad that people don't have that experience as often anymore. I think that there was more of a culture of sitting down and listening to records, you know, earlier, like maybe in the 70s, and that's sort of gone away a little bit, and I guess it's making a bit of a revival, which is promising, but it's unfortunate that a lot of people don't ever end up experiencing like the really amazing recordings that have been made. We have these incredible, there's a lot of incredible engineers and artists making really amazing digital music. A lot of people don't get to really experience it, which is too bad. - One of the really interesting things I learned reading your paper, and I'll link to it in the show notes, was that the MP3 compression is actually biased to European men or men of European descent at least. Could you tell me a little bit about why that is? - Yeah, it's sort of an interesting thought that that's how it is. The MP3 format, when it was created, it was made by all these European audio engineers in Germany and Austria. They tried to be rigorous about how they tested the music, but didn't totally succeed. When the MP3 was being developed, they used a lot of listening tests to determine whether the format was doing a good job of compressing music and having it still sound good on the other side. You know, as we'll talk about more later, I'm sure as Tom's Diner was sort of the famous first song encoded as an MP3, and it's known as the mother of the MP3, and that was just sort of arbitrarily chosen by the engineers, you know, they heard it on the radio one day and thought that it would be an interesting challenge to encode, and you know, sort of the rest of the music that was used, it's all stuff that was pretty well recorded. There's part of a hiding trumpet concerto and a Tracy Chapman song. It's all stuff that really conforms to, I think like, sort of rigorous Western European and American standards of audio production, but you don't have any, you know, there's no Alan Lomax recordings, for example, and there's no, they didn't test it with, you know, with noise music or like punk rock or drumming music from Africa or these different kinds of music, and so it ends up that MP3s do a really fine job in coding really well recorded music, really pristinely recorded music, but they don't do such a good job of encoding things that are recorded less well or that maybe have different aesthetic values that, you know, when somebody is trying to make like noise music or something like that, that stuff really gets mangled with MP3 files, and so there's this sort of implied aesthetic that comes along with sort of this baggage of MP3 files that comes along with it that only certain kinds of music really fit with that format, and if it doesn't really fit that template, then it's just not gonna sound good. - So maybe we could move on and talk about the work you created that highlights some of the differences between the original recording and the compressed version. I've seen your work referred to as the ghost in the MP3, but also as modernist, or maybe it's modernist, if I'm reading the caps correctly. First of all, what's the right title, and then can you tell a little bit about it? - The ghost in the MP3 is the name for just like this whole, the larger project of just this sort of line of questioning and line of research, this specific Thomas Diner track that's made out of the material lost from MP3 compression from the song, "Tom's Diner." That one is the one that I've been calling it modernist. That's actually, it's an anagram of "Tom's Diner," so if you rearrange the letters. And I just thought it was kind of a fun title. - Yeah, I've disappointed myself for not seeing that. That's really cute. (laughs) - Yeah, most people didn't notice it, and so that's what's up with the funny capitalization. That's my little like embedded Easter egg, little clue that like there's something funny going on. - Yeah, so I'm gonna play just a snippet of that right here, and then I'll play maybe the full version on the way out. - Sounds good. (upbeat music) (upbeat music) - My show gets distributed as an MP3, so now it's an MP3 of the compliment of a recording of an MP3. How is this going to turn out to my listeners ears? - It should sound pretty good, actually still. - All right, well, I'll encourage them to check out GhostInTheMP3.com, which I'll link to in the show notes as well, or as in "The Ghost In The MP3." - "The Ghost In The MP3.com," and I also bought GhostInTheMP3.com, but I just haven't set that up yet, to link to the regular one, but it's yes, "The Ghost In The MP3." - So they can enjoy the full version there, as well as the interesting video that complements it. - Yep, and the lossless version is there to download too for free. - So I was describing it as the compliment, or what's left over from the compression, could you give your take on exactly the process for creating the work? - That's a really good description of it, I would say. What I did is took the CD quality audio of Tom's Diner. I analyzed that file, looking at all the frequency information over time. And then I made an MP3 version, and analyzed that one as well, using the same kind of analysis, to sort of learn what all the frequency content is in the file. And then I compared the two files, and you just go bit by bit, and compare what frequencies are present at any given moment, and you look for differences. So 400 Hertz is happening at 60 dB at five seconds, and five seconds into the file or something. You look at the MP3 file, and say, "How loud is 400 Hertz there?" And then you just compare, and so I found all the places where the MP3 is different from the original file, where it changed, and stored that information in an array, and then re-synthesized that into audio, so that you can then hear what was lost when you compressed it into the MP3. And there's actually a couple of different ways that you can get at that information, and I tried a whole bunch of different possible approaches, because for me, my goal was to then make music out of this. That was sort of my end goal from the beginning, it was to make music, and so I tried a lot of different approaches, and varied my parameters for how I was analyzing things, and re-synthesizing, and try as many different approaches as I could, to just generate material. Once I had all this material, I set about trying to re-compose, make this sort of artful reconstruction of the original Tom Steiner, but just only using the material that was thrown out during MP3 compression. - I think most people are familiar with waveforms, that is the sort of time display and amplitude of sounds, but not everyone knows the frequency domain. Could you describe sort of what that looks like, and I'll encourage people to look at your papers, but as best we can over an audio podcast, how that's different from the more traditional way people are used to seeing sound displayed visually? - Yeah, so normally when you look at a waveform, what you're seeing is it's a two-dimensional graph, and you have on the x-axis, you have time, and on the y-axis, you have amplitude, and what that's showing is it's a voltage variation, really, but which is the analog of an air pressure variation that's happening in the air, so when you have a sound wave, there's little displacements happening of pressure above or below the standard atmospheric pressure that's in the room at the time. So you end up with these voltages that are going up and down between one and negative one, get this sort of waveform outline that you see when you see a waveform. That's like a really direct display of what's the physical phenomenon of what's happening in the room. That's not really very relevant perceptually. What our ears do is we, our ears take this continuous signal of pressure variations, and we sort of compress all that information to use maybe a problematic word, a loaded word. We compress all that information into hearing frequencies, so if there's a waveform that we see patterns, basically, so if there's a waveform that's moving up and down between one and negative one at some rate, maybe 400 times a second, what we hear perceptually is we just hear a continuous tone at a particular frequency. And we would say, oh, if it's like 440 hertz, that would be A, so we'd say, oh, that's an A, and it's just a continuous sound. So we don't hear the, we don't really hear the variations in the pressure directly our brains abstracted and we hear a tone. We just hear a continuous tone. There's this mathematical technique, and there's like a whole bunch of variations on this too, but a really common one is the Fourier transform. What that does is it allows you to do this, simulate this process that our ears do in a computer, basically, or just you can do it on paper too, it just takes forever. And you can analyze a time domain signal, that would be these fluctuations that are happening in amplitude and time, and discover what the frequency content is of that. So if there were, you know, what frequencies are present, and so that, the end result of that is a different 2D graph that's frequency along the y-axis and time along the x-axis still. So that's called the frequency domain representation of an audio signal. And that's a little closer to what we really hear, as opposed to the time domain representation. - So then, tell me if my understanding is correct, that the way you created modernist is taking those two spectral frequency images essentially, and subtracting one from the other, and then creating something with the leftovers, if you will. Is that a fair description? - Yeah, absolutely, yep. And it's, yeah. So I ended up with like a third frequency time representation that was the difference between the original two, and then I did an inverse Fourier transform to that, and then you end up back with a waveform with an amplitude time waveform, and then you can play that out of your speakers, and that's like a voltage variation that gets translated into air pressure fluctuations, and then you hear the sound. - So I learned a new piece of vocabulary from reading your paper, it's the term plundrophonics. - Oh, cool. - You could give a definition, and tell me if you think your work qualifies as a work of plundrophonics. - Well, it's definitely inspired by plundrophonics. Plundrophonics comes from this composer, John Oswald, who wrote this great essay in the '80s, called Plundrophonics, or audio piracy as a compositional prerogative, and basically his idea is to make new music out of taking existing audio recordings and altering them in different ways to make new compositions, and so it's sort of like a form of sound collage, and he got into a lot of trouble for this when he started doing this in the '80s. I guess it's related to sampling and stuff like that. I would say, yeah, my piece sort of is a form of plundrophonics, although maybe the way my method or my technique is a little bit different than John Oswald's. - Could you tell me a little bit about the analysis tools you use on a software or library level? - Sure. - What do you do to create the work? - I did most of my project in Python. I used a few different libraries in Python. There's a sound processing library called PYO that some folks up in Montreal have been working on. It's really great. I don't think a lot of people know about it. It's pretty new, but it's nice. The nice thing about working in Python is that you're able to do audio work, audio manipulation, but then you're in a general purpose programming language, and so you can sort of visualize your data and manipulate the data using sort of MATLAB type functions and stuff like that, and so it makes it really easy to sort of handle the data and see what you're really doing. I used PYO, and then I also used the Bragman Toolkit, which is made by Michael Casey at Dartmouth College. That's also really handy sort of Python library for doing audio analysis and synthesis. It has a lot of actually different functionality that I only use some of. Let's see. And then I also, some of the final track, and this is more on the post-processing side of things and like the compositional end, I use this library called Headspace, which is a Python library that allows me to do head-related transfer functions, and so that's like a 3D audio trick. So if you listen really closely to the high fidelity version of the track that I made, and you put some headphones on or you sit there with speakers, there's sort of the simulation going on of 3D space that happens with just stereo sound, and so it makes it sound like the sound sort of moves above you and behind you and around you. So that was sort of a, that was just for, I think, more of like sound design, compositional purposes that I ended up using that. There didn't really exist any tools to do this. I searched long and hard for an easy way out, but there was not one. So it ended up being sort of a trial and error process on my part. - So you can definitely hear elements of Suzanne Vega and the band in Modernist. It's recognizable as Tom's Diner. I feel like if I played it for someone, didn't tell them what it was and ask them to guess the song, there's a high likelihood they'd get it right. - Yeah. - And part of my point in bringing that up is to say that there are things that are lost that are intrinsic to the sound. This isn't things that are inaudible completely or white noise or pink noise whether it's being taken out. This is something that's generally being lost from that track. So I'm curious, I wanna ask you to speak for all musicians, but at least yourself as a musician and a composer and a performer, how do you feel knowing that most people will enjoy your work in a format where you've partially lost control of how it's presented to them? - Well, I would say that MP3s do, they do a pretty good job of deleting material that in most circumstances are in ideal listening circumstances, I guess, aren't noticed. So it's that contextual aspect of auditory processing. The sounds that get deleted when you isolate them, it's like, oh man, that's a really clear part of the sound, but then when you put it back into its original context, it's something that you maybe wouldn't have noticed because it was being masked by something louder that's in the texture, that's in the overall texture. Ideally, the sounds that get thrown out are ones that are super fluidist or something, that are just not, they're just auditory cheesecake or something that is not necessary really to get the impression of the song, but it doesn't really work perfectly and it's an approximation and there are better formats and we've learned more about the human auditory processing system since the MP3 was developed and there are more modern formats that can achieve similar levels of compression in more sophisticated ways. And so, the MP3 is, it's a really good, but still sort of rough tool that does end up cutting out little bits of things that I think would be noticeable, that are noticeable in people, and people do complain about things such as pre-echo, how MP3's smear, they sort of end up smearing like really sharp transient sounds. So like a hi-hat or like a sound, which a lot of vocal sounds and different consonants, those sounds get sort of smeared and so they're not as crisp and the articulation isn't as clear in an MP3 as it would be in an uncompressed wave file. So I would say, I don't know, I've read interviews with audio engineers and I think like one sort of dictum that gets passed around among audio engineers is like, if it's recorded well enough, if you just record it like at high enough quality, and if it's like a really immaculate recording, then even with like it being an MP3, it'll still sound good. So like the idea is just to like record it as well as, as well as you can. And the things that get compressed into MP3's well are really good recordings. So like a great recording is still gonna sound pretty good on an MP3. But you know, not everybody has that capability, there's a lot of people that are doing stuff in their home studios now and that don't have access to like ideal equipment and pristine preamps and recording spaces and all that. And that stuff does get affected. And so I think it's, I think a lot of people get frustrated by it. I don't know. It's sort of a, it's like a necessary evil at the moment though, because I think it also, it has, it's a trade off right now. It's a compromise at the moment that hopefully we won't have to make for long, but it's like a funny compromise of this era that we're in right now that we need to make because our storage space on our cell phones and like mobile computing and bandwidth, 3G, 4G bandwidth, all that stuff, it's clear like everybody wants mobile computing and it's just like that. The storage space and bandwidth for that hasn't really caught up with the fidelity that's possible with digital audio. And so hopefully, I hope that in five or 10 years, we'll look back at this period and kind of be like, oh man, that was a weird phase. We had to compromise our audio quality so much instead of just having like really high fidelity. It's sort of this like, it's this like sideways technological movement that we've made and instead of, you know, normally you think of like technological progress is always moving towards like better and better higher fidelity things. We've sort of moved sideways and like, you know, MP3s don't sound better than CDs. - Yeah, it's interesting. It's an interesting time right now. - Absolutely. So I would say that as a whole, the Ghost in the MP3 project really stands on its own as art. These are really interesting things to go listen to. But in addition to that, do you hope that there's a takeaway or a message that people will get from the project as a whole? - Yeah, I think I would hope that people just become curious about digital audio and about sound, the sound that they're listening to and that it causes, you know, people to ask questions that maybe hadn't occurred to them. Because I think it's a really salient way of experiencing something. It's interesting, it's really interesting to hear the sounds that get deleted from MP3 compression. That's something that wasn't possible that, you know, most people had never done before, have never heard before. And so it's like, it's a new experience. And I think it raises a lot of interesting questions if it just causes people that maybe had never thought about these issues before to become interested in exploring different audio formats or different, you know, audio quality and to maybe sit down and listen to some, their favorite recording on a CD or something, have a musical experience and that would be awesome. - Absolutely. So what's next for you if people wanted to follow your work? - There's a Facebook page for the project. People can follow, it's the ghost on the MP3, it's on Facebook and I'll be posting updates there. There's the website, obviously. I think the Facebook page is good if they wanna follow updates. There's a sound cloud, I have a sound cloud account too. And right now I'm working on, so there was a whole series, there's about like eight or so songs that were used in the original MP3 listening tests besides Tom Steiner. I mentioned some of them before, like Fast Car by Tracy Chapman and this Hide in Trumpet Concerto. So right now I'm working on reconstructions of those things, these sort of, so I wanna make a whole series of pieces and so that's gonna be coming out in the next couple of months. Tracy Chapman one is nearing completion and then we'll see where it goes from there. - So I like to wind up all the shows by asking my guests for two recommendations. The first I call the benevolent recommendation, something you'd like to give a nod to but aren't necessarily affiliated with and then the self-serving recommendation, something you ideally get some direct benefit out of from the exposure here. - Let me think here for a second. - Okay, benevolent recommendation that I wanna recommend is this group called The Happy Valley Band. You can find out about their stuff on indexical.org, it's I-N-D-E-X-I-C-A-L.org and they're playing a series of shows out in California, sometimes soon, they do this really amazing project where they cover songs from the Great American Songbook but through machine listening. So they do like these machine listening transcriptions of songs from the Great American Songbook and then they attempt to perform with live instruments. These like really overly complex and ridiculous machine transcriptions that have been made of the songs. It's really entertaining and kind of funny and interesting. So they're really great and so I encourage people to check them out and I think they have a CD coming out soon. - And let's see, self-serving recommendation. Yeah, go to the Facebook page for the custom MP3, I guess. (laughs) - And if everyone likes that page, hopefully it'll pop up on more feeds. - Yeah, yeah, that'd be cool. - Cool, well Ryan, this has been really great. I'm really glad you took the time to do this. I think the listeners are gonna really enjoy hearing about your project. - Thanks a lot, I really appreciate it. Great speaking with you and it's really my pleasure. - And we're gonna go out with modernists today so people enjoy the full work. And thanks again for joining me. (upbeat music) (upbeat music) (upbeat music) (upbeat music) (upbeat music) (upbeat music) (upbeat music) (upbeat music) (upbeat music) (upbeat music) (upbeat music) (upbeat music) (upbeat music) (upbeat music) (upbeat music) (upbeat music) (upbeat music) (upbeat music)
