In this All Things Cognition podcast, I interview Matt Evans and Nicolas Davidenko about their recent paper on the pitch of earworms. Let’s get right into it!

Interview

Transcript

Lai: You’re listening to All Things Cognition, a Psychonomic Society podcast. Now here’s your host, Laura Mickes.

Mickes: Having a song stuck in your head is probably a nearly universal experience. These are called earworms, or more formally, involuntary musical imagery. Related to this is a paper recently published in the Psychonomic Society Journal, Attention Perception & Psychophysics, called Absolute Pitch and Involuntary Musical Imagery. And I’m meeting with two of the authors of that paper, Matt Evans and Nicolas Davidenko. [Pictured below with their coauthor, Pablo Gaeta.] Thanks so much for agreeing to meet with me to talk about your paper.

Evans: Yeah,

Davidenko: Thanks for having us.

Mickes: Great. I have no musical ability whatsoever. Friends might even say that I have amusia <laugh>. When I saw this paper, I was, I was pretty excited about what you found and I wanted to learn more about it, so I’m really excited to hear about your research.

Mickes: I guess we’ll just start with the usual questions. What’s the background? What, what got you into this research?

Evans: Years ago. Years and years ago, a friend bought me as a birthday present, bought me a book called, this Is Your Brain on Music, which was written by Dr. Daniel Levitin, who’s a, a big name in, in the field of music cognition and perception. And it’s a great book, New York Times bestseller, very easy read. And it sort of introduced me to the field of, of music cognition. I’ve been a musician my whole life, but never really put in enough work <laugh> to be, to be able to make it my sort of primary pursuit. I also, you know, knew at the time that I was interested in psychology, and so it seems like this great intersection, right, where I could, I could pursue psychology and, and, and cognitive science and school, which I, which I was good at and still stay connected to music, which I had not <laugh>, like I said, had not put in the work to be good enough at, to make a career out of.

So, then years, years, years, years later, I just been accepted to the, the PhD program at University of California Santa Cruz. And, and Nick and I were chatting about potential research directions. We, we knew, you know, a couple different directions that we were interested in going. And Nick mentioned that he, he has pretty much near constant earworms, and so turns out there’s a term for that in the literature. It’s called the, the Perpetual Music Track. And there’s, there’s one study about it that is an N of one that is a self, sort of self-observational study. So anyway, it just opened the conversation about earworms and it turns out the literature is very exciting as a new sort of early career researcher to find this field that the literature really only goes back to around 2006, 2008 in the English language literature.

There’s, there’s some other writing prior to that in, in German. But, but yeah, so it, it was, it was exciting to find this field where I was like, oh, I could actually reasonably read every paper that’s been written <laugh> on this topic. So I started and, and going back to that book that I mentioned, that Daniel Levitin, he talks about this study that they did that I, I didn’t know at the time, but is a, a super well-known study in, in the field of absolute pitch, which was, you know, you bring participants into the lab, you have them think of a song that they know very well, and imagine hearing it in their head and then sing it out loud. And it, it turns out that people have well above chance accuracy in terms of their recall of absolute pitches in these well-known voluntarily recalled songs. We realized like, oh, no one’s, no one’s done this with involuntarily recalled songs. And so it seemed like a natural sort of next step.

Mickes: Nick, you have earworms all the time. I feel like I do, too. Always. And it’s not just songs, it’s sometimes some phrase or something that somehow had some musicality, it seems. But yeah, that, that’s funny. Do you not have that, Matt, do you not have an earworm constantly <laugh>?

Evans: No, although the more I talk to people about this work, the more it comes up, the, the more I realize it’s I think probably fairly common. There’s tons of individual differences in all sorts of aspects of involuntary musical imagery.

Mickes: Okay. So that’s what got you interested in the topic. What were your research questions going in? I mean, I, I guess it follows from what you said, but I’ll just get you to say it explicitly.

Evans: Yeah, sure. So what we wanted to know was in layman’s terms, right, when a song is stuck in your head, is it stuck in your head in the correct key. And in more formalized language, what we wanted to know is so we have this existing evidence that it was done by Levitin in 1994 and then replicated on a very, on a large scale across six different international laboratories by Frieler And colleagues in 2013, I believe, that sort of replicated this effect of voluntarily recalled, well-known melodies preserve absolute pitch information accurately.

Evans: The main question was, does it require conscious effort and voluntary recall, or is it revealed in involuntarily recalled melodies that sort of pop into your head throughout the day? This was not one of our main research questions, but it was something we were also interested in, was in those previous studies, people picked the song that they recalled and, and they were instructed to pick a song that they know well. And so one of the other questions we had was, does familiarity with the tune play a role? That was one of the questions we asked folks when they reported an earworm was how well do you know that song?

Mickes:

Okay, great. How did you go about answering these questions? It did not look easy, as I read through your methods section …

Evans: <Laugh>, my favorite, my favorite story about this project is it was my first year project as a PhD student. And at some point Nick and I were having our one-on-one meeting, and he used the phrase, oh, this is a very ambitious project, which I think was a nice way of saying like, we took on maybe more than we normally would have for a first year project, but no, it was great. He is, he is very supportive. Yeah. So we used a, the experience sampling method. So we, we recruited it was actually sort of perfect timing because this was in 2020, so it was sort of peak covid and, and everything was remote, so it was perfect timing to, to use a methodology like this. We had 30 participants who we would meet with over Zoom to do an onboarding meeting, and effectively they gave us their cell phone numbers.

So Pablo, one of our co-authors, they were a computer science and engineering undergraduate student,

Mickes: Handy.

Evans: Working, yeah, <laugh> working as a research assistant in the lab. And they were just incredibly, incredibly vital to this project. They wrote us a web app and built a backend and set up our automated text messaging and all that. So our participants for a period of two weeks would receive six text messages a day at sort of pseudo random times, and each text message had a link to this web app survey that Pablo had set up.

Mickes: Cool.

Evans: And the first question every time was, at the time you received the text message, was there music playing in your head? And if they answered yes, they would click a button in the web app and it would turn on their phone’s microphone and start recording and we would have them sing or whistle or hum whatever the music that was playing in their head into the phone. And that would capture that and upload it to, to a cloud where we could, we could receive it later. And, and then there was a follow-up survey with 20 something questions.

Mickes: Like the familiarity to the music like you mentioned,

Evans: Such as the familiarity of the music and questions about their state of mind, what they were doing at the time, things like that.

Mickes: Okay, great. It sounds like a really fun experiment to take part in. I know it’s a, it’s, it’s two weeks, six messages a day is a lot, but it does sound fun. I don’t think you would have people complaining to you, but …

Evans: Yeah, I, I think people enjoyed it. The reason we did two weeks was we wanted to make sure that the process of asking people to sing and record their earworms didn’t have an effect on their awareness of their earworms. And so we had one sort of record week and one non-record week where during the non record week they would still get text messages and answer the survey questions, but they didn’t do any singing. And then we counterbalance that between participants just to make sure that we weren’t having some sort of impact on people’s ear worms just by asking them to sing and record, and we didn’t see any effect there.

The most common response I get from people is, oh, it’s a good thing I wasn’t a participant in your study. Because I would’ve been. Right, similar to what you said at the beginning.

Mickes: Yeah, exactly.

Evans: Sort of, everyone sort of has that same response of like, oh, I would’ve been a bad participant.

Mickes: Yeah,

Evans: And I mean, spoiler alert a little bit, right, for the results, but we, we don’t see any connection between singing ability and, and anything like that.

Mickes: That is incredible. But yeah, you, you beat me to the question, what, what did you find.

Evans: <Laugh>.

Mickes: So yeah, what did you find?

Evans: So there were 30 participants. We had to be able to match their recording to the original song. So that meant we had to be able to identify what song their earworm was from. And in some cases, their earworm was from a song that has been recorded in multiple different keys. And so we would have to toss those out. So we, we ended up with a total of 132 recordings that were able to be analyzed for pitch.

Mickes: That’s good.

Evans: And we compared those to the original segments. And so across those 132 recordings close to 50% I don’t remember the exact number off the top of my head, but close to 50% were in a, what we defined as a pitch error of zero semitones, meaning effectively they were in the same key as the original. We corrected for octave because people have different vocal ranges and we were really just interested in what we call the pitch color, which is, you know, if the song starts on an A, were they also starting on an A?

Mickes: And so if they didn’t, that would be an error.

Evans: Correct. So if the song that they were singing was supposed to start on an A, but they actually started on an A sharp, then we would code that as a pitch error of plus one semitones.

Mickes: Okay.

Evans: If the song was supposed to start on an A and they actually started on an A flat or G sharp, we would code that as a pitch error of minus one semitones. I see. And so because we’re correcting for octave, that gives you this range, right? Where an error, a pitch error of zero semitones is exactly the same key. And then you can be plus or minus one plus or minus two plus or minus three plus or minus four plus or minus five. And then plus six or minus six ends up being the same thing because we corrected for octave. So anyway, we ended up with this very sort of like peaked normal-ish looking curve with about 50% of responses with a pitch error of zero semitones. And then in absolute pitch or perfect pitch research, you often will allow for an error of up to plus or minus one semitone [see figure below]. And so if you allow for that error, then we ended up with just under 69% of responses were within plus or minus one semitone.

Mickes: That is incredible. Based on the previous research, maybe you expected that, but it seems like such a high number. What did you think about that?

Evans: On the one hand, you kind of expect it, but on the other hand, there, there wasn’t really necessarily any reason to assume that these involuntarily recalled melodies that would not necessarily include only songs that are well known. We didn’t see any connection between familiarity and pitch accuracy.

Davidenko: Yeah, I think that’s correct. I don’t think there was a relationship between people’s ratings of how familiar that particular song was and how close in pitch their recording ended up being. And many recordings participants didn’t know the lyrics, and so they would hum or get the wrong lyrics, and they were still often in the right key. So that was definitely a surprising aspect of our data that didn’t seem to require intimate knowledge of the song.

Mickes: And still had the pitch right. That’s incredible.

Evans: Yeah. One of the most interesting questions, I, because again, this was my, my first year project, right? And in our department, the first year PhD students give a talk on what they’ve been doing the year, and there’s a faculty member in our department Hannah Hausman, who’s a, a memory researcher who asked me a great question, which was, you know, normally in memory research we think of people as remembering the gist of a memory. And you might expect <laugh> that if that the gist of a song wouldn’t necessarily have to include the absolute pitch information.

Mickes: And so is she thinking that, I’m sorry to interrupt, but is she…

Evans: No, please.

Mickes: That it’s, more verbatim type memory for the song, so they nail it.

Evans: That’s, that’s one way to think of it, right? Is that maybe at least this aspect of musical memory appears to, you know, appears to be preserved vertically. And that it’s maybe other aspects of, of the melody, you know, like Nick was mentioning, that they don’t necessarily know the lyrics and in some cases they don’t necessarily even nail the sort of melodic contour. They insert extra notes or they remove a note or two. But the, the overall gist of the melody is still there, it’s still recognizable. It just happens to be at the correct pitch height, right? The absolute pitches is veridical.

Mickes: There are different … If you are thinking along the lines as she was thinking, their memory for the tune itself is gist, but for the pitch it’s verbatim. That’s …

Evans: Right. So it just, it opens this door into questions about like, okay, which, which aspects like what different musical features are represented vertically versus in a, in a gist form.

Mickes: So how prevalent is absolute pitch in society?

Evans: It depends on what society <laugh>.

Mickes: Really?

Evans: So yeah, so in western cultures where the native language is not a tonal language, so for example, America, right? The prevalence rate is estimated at about one in 10,000. And that one in 10,000 is specifically referring to the sort of <laugh>, I, I don’t mean this dismissively, but it’s sort of the, the way I’ve been referring to it, the sort of “party trick” version of perfect pitch, right? Where you can, someone can play a note on the piano and, and someone across the room could say, oh, that’s a, that’s the G sharp above middle C.

I was watching a TikTok recently where a friend was finding random different objects around the house, like a coffee cup and clicking it. And his, his friend was at the piano telling him, oh, that’s there’s actually two notes there that’s an F sharp and an A flat.

But that sort of like, party trick version of perfect pitch, right, of like, oh, I can hear a note. So that’s one in 10,000. And that is made up of some theory suggests that that’s made up of two different aspects, right? There’s the pitch memory, but then there’s also the pitch labeling ability, this study and, and some other studies on absolute pitch that suggests that it’s that pitch labeling ability that’s sort of critical to the party trick quote unquote version of absolute pitch.

Now, in, in cultures where the native language is a tonal language, so for example, Mandarin Chinese, when people learn a tonal language as their first language and they receive formal music training during the like critical language acquisition period, the prevalence rate goes up to around one in a hundred.

Mickes: Oh, that’s such a big difference.

Evans: Yeah. It’s, it’s a huge difference.

Mickes: That’s amazing. So that’s, this is why it surprises me because I thought it was very low, but I think I only thought of it as the party trick way. If we get back to your findings, what do you think about them, like what does it all mean?

Evans: First and foremost, our findings suggest that memory for absolute pitches is more accurate than commonly thought. And importantly, and novelly that accurate recall of absolute pitches happens automatically, doesn’t require conscious effort and doesn’t require close intimate knowledge of the melody being recalled.

Also, we didn’t find any evidence of a connection between participants singing ability or general musical sophistication, their accuracy.

Mickes: You would’ve asked them how much musical experience you’ve had or what do you, are you skilled as a musician or something? You would’ve asked those questions and, and didn’t find any difference between those with expertise or some sort of experience versus …

Evans: Correct. Yes. We gave a validated survey called the Goldsmith Musical Sophistication Index [GMSI].

Mickes: Oh.

Evans: Which is a measure that’s designed to be sensitive to differences both amongst musicians and non-musicians. And it’s divided up into several subscales. So one of the subscales is general sophistication, and one of the subscales is perceptual ability, singing ability, musical training. And one of the subscales is called active engagement. Active engagement is things like, how often do you go to concerts? How much money do you spend on music? Right? Like sort of actively engaging with and consuming music.

Mickes: Right.

Evans: Out of all the subscales on the, the GMSI in our sample active engagement was the only subscale that was predictive of people’s accuracy. Singing ability was not, musical training was not, perceptual ability was not.

Mickes: Wow.

Evans: Now we had 24 participants that we were able to analyze for pitch, because as I was mentioning earlier, some, some had to be dropped for a variety of reasons, but, but still, it’s an interesting result.

Mickes: Yeah. And it’s not that you got 24 people with tons of musical experience and it turns out that doesn’t even matter, it seems.

Evans: Exactly, yeah. A lot of our participants had a year of musical training that’s fairly common in America, right? People get <laugh> a year of school band

Mickes: Wait, not just playing the recorder?

Evans: Yeah, I mean they, that might have been what they were counting. I’m not sure, but yeah. Nick, did you want to add anything?

Davidenko: Well, I wanted to connect to a, a related point, I guess, which is regarding this engagement with music. So it could be that, you know, people who listen to a lot of music may have recent memories of the songs that they are having as an earworm. Maybe they heard those songs earlier that day. So one interesting analysis we, we did in the paper is because we asked, after each participant recorded their earworm, we asked several questions about the potential source of that earworm. And in particular we asked, did you hear this song recently or not? And it turned out that about 50% of responses said yes, and 50% said no to that question. So we were nicely able to separate the data into recently heard versus not recently heard songs. And it turns out there’s no difference in terms of the pitch accuracy of those two sets of responses. So whether you heard that song an hour ago or whether you heard it two months ago, whatever reason it popped into your head seems to give you the same probability that it’s going to be in the correct pitch.

Mickes: That’s incredible. Is it possible that they just didn’t remember hearing the song? Because I’ll be, I’ll be having an earworm and think, what, where’d that song come from? But then it was from a commercial or something. Ae they just misremembering?

Evans: I think it’s possible, at least for some of them. What’s interesting about that, right, is if it’s not consciously available as the source, then the fact that it still somehow is operating under the surface and, and maintaining that, that accuracy.

Mickes: Yeah, that is really incredible. This ambitious project! Now that you’re done with it, was it an understatement?

Davidenko: No, not at all. <Laugh>. It was extremely ambitious, more than I thought, probably at first because

Evans: Mm-Hmm. <Affirmative>.

Davidenko: Not only the parts that our co-author Pablo was responsible for doing all the kind of behind-the-scenes backend of this project, but also it required a lot of work from undergrad research assistants. The process for capturing the pitch of these sung earworm fragments and then finding the corresponding song, the corresponding section of the song, and then comparing those that took many, many hours of work from different research assistants and double coding so we could make sure that there wasn’t any bias. That was a huge component of the, of the project. I guess. I, I anticipated that that would be part of it, but.

Evans: Mm-hmm, <affirmative>.

Davidenko: It, it was a lot and it took a while to, to get all the recordings coded.

Mickes: Yeah. So it was ambitious to say the least, I guess you could put it.

Evans: Yeah. It, it was extremely rewarding and I’m, I’m really glad to have had the opportunity to do it and, you know, have made several connections with colleagues at different conferences and have ideas about how to do it easier, faster, more efficiently, more accurate, you know in the future. So I don’t think we’re done with this topic yet.

Mickes: I was going to ask what’s next? Can you tell us?

Evans: Yes, of course. So there, there’s several things we’re interested in, and a lot of it has to do with the nature of how we were analyzing pitch in this study. So one of the things we noticed when we were doing our pitch analysis is in not a huge number of cases, but in some cases, participants’ earworms were rapping. So rap songs or some other kind of like talk singing. And those stimuli certainly contain pitch information that can be analyzed, but it’s much more difficult by ear to match the pitch of someone talking versus someone singing. And there’s a lot of reasons for that, you know, we don’t necessarily have to get into here, but suffice it to say our solution was to not include. Our coders would try to code them, but very often what would happen is we’d have our first pass coder and our second pass coder, and then there would be disagreement. And so we would give it to a third pass coder and there would still be disagreement. And if we had disagreement between all three coders, then we just would exclude that stimulus. But that means we’re sort of systematically not analyzing a specific genre of music and, and that is a problem that we would like to address moving forward,

Mickes: And you figured out how to do that.

Evans: There are several options. We haven’t solved it yet. There are some obvious solutions, it’s just a matter of like what will actually be the most efficient. But like Praat, For example, is a linguistic analysis software that can find the average fundamental frequency of a syllable over time. I just haven’t played with it enough yet to know. I was just recently at the Society for Music Perception and Cognition conference up in Banff, Alberta this last summer and saw, you know, several people who were using Praat for pitch analysis in similar ways that we might be able to borrow. So

Mickes: Great. So you’ll explore it.

Evans: That. Yeah. So that’s, yeah, that’s one thing we want to do. We just want to make sure that we’re, you know, because we realize, oh, we are, not only are we systematically excluding a specific type of data, but a specific type of music that is mostly generated by people of color. And so we want to make sure that we’re not excluding that from our analysis. So that’s a gap we want to make sure we fill moving forward.

Nick mentioned this analysis of recency, and so we think we can get even more fine grained with that analysis. The analysis that we included in this study was a forced choice did you hear the song recently or not? But we are interested to see whether or not there is an effect of a more fine-grained effect of, you know, if it’s an hour ago versus six hours ago versus a day ago versus a week ago, right.

Are we going to be able to see an effect? And one of the ways we might be able to see that effect is the third thing that comes next, which is in this study we discretized participants productions to the nearest semitone. So in other words, we rounded a semitone, just to, to clarify is if you picture a piano keyboard, the distance from one key to the key that’s right next to it. So if it’s a white key to the black key right next to it, or if it’s a black key to the white key right next to it, that is a semitone.

Mickes: Okay.

Evans: And we, we discretized our participants recordings to the nearest semitone, but our participants were not perfect singers and they were in some cases slightly out of tune. And so that’s again why we used our multiple coders and we looked for disagreement.

However, if we did switch to an analysis software such as Pratt, we could get more accurate. Rather than saying this person is singing in a, we would be able to say this person is singing a tone with a fundamental frequency of 442.6 hertz.

Mickes: Right.

Evans: Which, if a research assistant heard that they would disco discretized that to just an a.

Mickes: Yeah.

Evans: But the software can get more accurate than that. Yeah.

Mickes: Oh, cool.

Evans: So we can just do a much more finely grain analysis in the future.

There’s another project we’ve been working on where we’re interested in effectively from a more top down, broader perspective. We’re interested in this idea of, and this is going all the way back to the, the beginning where we’re talking about people who sort of have constant earworms, but it, it’s also relevant to sort of across the spectrum of how prevalent or consistent a person’s earworms are.

We’re interested in this idea of what happens to the earworm when you lose conscious awareness of it. So in other words, if Nick who has constant earworms is, he’s fixing some of my coding mistakes in matlab, right? So he’s, he’s focused, he’s very focused and he is typing, typing away. He has noticed that his earworm can get pushed to the background.

Mickes: Oh.

Evans: And for me, who does not have constant earworms, if I am working on applying the edits that Nick has commented on my MATLAB code and I’m trying to like focus and type, type type, type type, if I had a song stuck in my head, it might go away. And so there’s this this idea of like, to what extent does attentional control of cognitive load impact what happens to an earworm? And so we have a project that we’ve been working on that the next thing that we’re going to be writing up and, and submitting. So stay tuned.

Mickes: Oh, excellent. I definitely will.

Davidenko: One of the few moments when my earworms go away is when I’m speaking, but not necessarily when I’m listening. So I could be listening to someone speaking to me and it’s still happening in the background. Or if I’m focusing on coding, it’s still happening in the background, but there are moments where maybe like a very focused or, or you know, doing something or I’m speaking where then it feels like, okay, that earworm is not there anymore. But then when it comes back, I’m, I’m curious to know, did it continue all the way along and now it comes back where it would’ve been?

Mickes: Right!

Davidenko: Or did it pause and then start back there or just start over again? Or sort of what are those temporal dynamics of what happens to this mental soundtrack when you become unaware of it based on what you’re doing or thinking?

Mickes: Right. That’s fascinating.

Is there anything we missed or anything else you want to talk about?

Davidenko: We were interested in, because when we look at our results compared to the result of Daniel Leviton from the nineties and the Frieler From the 2013, our data looks much more spiked. In other words, our results show a bigger effect, like more proportion of, of recordings that are on the right key. And so we don’t have, within our own study, we didn’t have the voluntary condition. We didn’t ask people, okay, think of a song you know really well and try to produce it and then compare that to when, when that happens spontaneously. So one of the potential future follow up studies is to kind of repeat our experiment, but then bring people into the lab based on the songs that they experienced as earworms during the two week period. Present those, you know, the names of those songs as a cue, have them think of those songs and see – now do you get the same kind of data or, or is it different if it’s voluntary versus involuntary?

Mickes: Oh, that is a neat study.

Is Pablo still working in the lab?

Evans: Pablo got a very great job…

Mickes: High paying job. <Laugh>

Evans: Understand, understandably. Yeah. Pablo is working for a major tech company across the country. So it was actually one of my, one of my favorite moments was towards the end of the year, I, I took my research assistants out for, for pizza and Pablo especially. I was like, man, you know, I, I really hope you were able to, they had, they had just gotten this job offer at, you know, X big tech company, but you can’t use x as a fill in phrase anymore because X is actually a big tech company. It’s not at x <laugh>, not Twitter. <Laugh>.

Mickes: Okay.

Evans: So Pablo had just gotten this job offer at this large tech company and I said, you know, I really hope you were able to, you know, utilize some of the, the work here for as part of part of your job application. And, and they said, this project was my job application, <laugh>.

Mickes: Oh, that’s awesome.

Evans: And so I said, they are well deserved working, but we are, we are hoping that we might be able to contract them at ….

Mickes: Oh, good.

Evans: a rate. We’re hoping they’ll give us the friends and family rate because we certainly can’t afford to pay them what they’re making now,

Mickes: <Laugh>.

Evans: But to, to help, help stand this back up in the future.

Mickes: That’d be great.

I don’t have any more questions, so I, I think this might be a good time to wrap it up and say goodbye. Thank you so much for talking about your research. I obviously think it’s super cool and can’t wait to see what else comes, comes from you both. Thank you.

Evans: Thank you so much.

Davidenko: Thanks so much.

Lai: Thank you for listening to All Things Cognition, a Psychonomic Society podcast. If you liked this episode, please consider following the podcast and leaving us a review. Reviews help us grow our audience so that we can reach more people with the latest research about cognition and the psychological sciences.

By the way – we run a blog, too! You can find it online at featuredcontent.psychonomic.org. New articles are published regularly, covering the latest research from all seven of our scientific journals.

We’ll catch you next time on All Things Cognition.

Featured Psychonomic Society article

Evans, M.G., Gaeta, P. & Davidenko, N. Absolute pitch in involuntary musical imagery.Atten Percept Psychophys 86, 2124–2135 (2024). https://doi.org/10.3758/s13414-024-02936-0