The truth about "fast" headphone drivers

追加: 2026-05-24

[00:00:00]Thank God there's hope. Hey folks, I'm Mark Ryan.

[00:00:07]This is Super Review. And last week, I posted this survey to my YouTube audience, asking them which of these three IM represented by these three different graphs has the faster driver.

[00:00:18]And the people who ended up responding, "The hell you talking about, bro?" They were probably the ones that were right.

[00:00:24]But let me back up. So, every once in a while, I get a question or a comment on my YouTube videos kind of like this that is asking me or talking about this concept of driver speed as if there's like this metric or quality of a driver that is its speed that's different between different headphones and it has some influence on the sound quality of a headphone. And there's a lot of like generalizations that people make about driver speed. Like things like planarss and electrostats are known for being fast drivers, whereas I guess basically everything else would be considered a slow driver. And again, these have some implications on the sound quality, but also potentially some limitations on how well you can EQ them. So, personally, I'm not really engaged with this concept of driver speed too much because one, I'm not a headphone designer or a manufacturer, right? I'm not really dealing with just drivers by themselves, but also it seems like most of the time when driver speed or CSD graphs come up, sounds a little bit on the pseudociency side, right? So those graphs that I showed earlier in my survey, those those are called CSD graphs or cumulative spectral decay graphs. And a lot of the people that seem to be bought into this concept of driver speed seem to think that CSD graphs are kind of the missing link between their listening impressions, things that sound fast, and some evidence showing that yes, indeed, they are fast. And I wanted to dig into it. And as it turns out, I actually have a lot of CSD data. Basically, as a byproduct of having measured hundreds of headphones and IM, I've got a ton of data. In fact, I made some new data doing some double-checking and I even talked with a researcher at UC Irvine trying to understand CSD graphs, driver speed, what do people mean by driver speed? Are CSD graphs useful? Are square waves useful? And maybe importantly, we're also going to talk about what was the result of that survey and why the people who were the most confused were probably the ones who are right. This episode of WaveGuide is sponsored by Juier. That's right, it's pronounced Juier. Now you know. And I wanted to start just by talking about what I think people mean when they're talking about fast sound. Right? If you've spent any amount of time in audio file circles, you've probably encountered people talking about fast sounds, maybe slow sounds. What do they actually mean?

[00:02:34]Well, like a lot of audio file terms, it's a little bit confusing, partly because everyone kind of uses these terms a little bit differently, but for the most part, what I think people are describing when they're talking about a fast sound is they're talking about sharp transients, right? And maybe on top of that kind of like just a a cleanliness, a cleanliness of the sound, like a a a strong delineation between instruments. So, in any given song, you might have a drummer, you might have a guitarist, you might have someone playing a trumpet and a vocalist and a basist kind of all playing on top of each other at the same time. And if the sound kind of ends up being a little bit smeary, like it's not really clear where the beginning and the ends of those different instruments are, that would be a sound that people would describe as being slow. But when it's really clear how differentiated those different instruments are, people would describe that as a fast sound. And there's some things like planer IM and planer headphones that are very often described as having a fast sound. Now, these impressions of things sounding fast and slow are generally impressions that I share. Actually, I I might even describe things as sounding fast and slow. It's not really that controversial that people kind of agree that things will sound fast or slow. Even if they use different terms to describe it, people will kind of agree on these sound signatures that it is not that controversial. However, some people will kind of take that a step further, right?

[00:03:55]Rather than just saying that this this headphone sounds fast, they'll say that it sounds fast because the driver itself is fast. And that was what I was kind of interested to find out. Can I find any data supporting that? So to start with, I actually just surveyed my audience on YouTube asking them for examples of fast sounding headphones and IM. And just kind of give you an idea of highle results. Planer magnetic IM and headphones unsurprisingly were very often cited as being fast sounding. So that wasn't that interesting. This was kind of interesting is there was a very strong correlation between things sounding fast and them being expensive.

[00:04:31]Also, I don't have this one graphed here, but I also kind of anecdotally noticed that a lot of the things that were being described as standout fast were kind of older products, maybe legendary type products, that was interesting. But whether or not these things actually are fast, it should be measurable. So, are CSDs the way that we measure it? Okay, we're going to get into some data that is a little touchy and a whole lot complicated. And I don't want to overstate things. I am not a sound scientist. I'm not a sound engineer. I'm just a pragmatist and I like using data that's useful. And while I'm not a scientist, as it turns out, I actually have more data on this than kind of all but a handful of people in the world because I measure frequency response graphs. And frequency response measurement is basically the same data that we need to produce a CSD graph like the one that you see here. In fact, you can see the frequency response of this IM here in the background. So, just to kind of break down what we're looking at in this in this visual right here, the x-axis is just like a frequency response graph. It is your frequency from 20 hertz, the bass, the low frequencies, up to 20,000 hertz, the high frequencies.

[00:05:38]Uh here on the y-axis, you're looking at amplitude, right? How loud is that frequency or how quiet is that frequency? The thing that's different here on a CSD graph versus a typical frequency response graph is we now have a third dimension, which is the Z-axis, and this is time. So underlying this concept of sound speed is the idea of decay. A driver can make a sound but then it might kind of linger, right? The the driver might not recover quickly enough and so it ends up kind of making that sound a little bit longer than it intends to. And that's what you would maybe see with a visual like this where the sound happens at its peak. It's loudest and then it kind of quietly fades off. Whereas a faster sound would be something that happens and then very quickly decays off because ostensibly the driver has recovered from making that sound. And that's kind of the intuitive visual that we get from a graph like this. All right. Before publishing this video, I actually shared an early draft of it with the aforementioned researcher and professor at UC Irvine, Craig Stark. We'll talk more about him in a little bit, but he actually added some very helpful commentary that I'm going to sprinkle in throughout the video. And for starters, on this topic of driver decay, Craig mentioned a really useful comparison of being a bell, right? And and and imagine a bell that you can ding and it kind of resonates and that sound gets slowly quieter over time. That is kind of the decay and the resonance of that bell.

[00:06:57]This is something that does happen in headphone drivers and that's I think ostensibly what a lot of people are looking in these CSDs to try and figure out. But notice that a lot of the times or pretty much all the times when you see these ridges in uh the CSD graphs that they are also corresponding with peaks in the frequency response because these resonances also result in peaks in the frequency response. Now a lot of the time when I see people using CSD data to me it feels a little bit like palm reading like they're looking at this crease and saying that this explains you're going to live a long healthy life. It's like they're starting from conclusions and kind of sifting through the data to find what looks like evidence to support them, which it's not exactly how data should work. Data should work in that you have data, you have some sort of analysis that generates a prediction, which then you can go out into the real world and validate that prediction. But unfortunately, it feels like a lot of the times when CSD data is used, it's more used in like post hawk rationalization. And as I dug more into my own CSD data that I've got, I think I understand why. And for starters, there's just the introduction of time in this analysis, I think, complicates things a lot more than people intuitively understand, right? Because you look at a visual like this, we have this new Z-axis of time, and we have this concept that there is the moment that sound happens, and then there are milliseconds that happen after that, right? How long is that sound resonating beyond that moment that it happened? But when you understand what frequencies are, you understand that's actually kind of an incoherent concept because low frequencies are actually Okay. Well, all frequencies are are kind of a product of time. A low frequency tone at 20 hertz is a waveform that repeats 20 times per second. A 20 kHz tone is a waveform that repeats 20,000 times per second. So, when you're looking at a visual like this that goes, let's say, up to 10 10 milliseconds of s of time. By 10 milliseconds, this tone can't even be registered, right? 20 hertz takes 50 milliseconds to register because it can only happen 20 times per second. In that same 50 milliseconds that this can play, you could actually have a thousand moments that 20 kHz is able to play. So all that said, while this visual looks really intuitive and it looks really clear exactly what's happening, the math behind it is actually a whole lot more complicated.

[00:09:14]Another thing worth pointing out is the influence of frequency response and the decay patterns that we see here. So I already pointed out that you can see the frequency response of this IM in the background. But that kind of implies that well basically all the things that are contributing to the frequency response are going to contribute to the decay pattern that you see. So you're not just seeing the speed and the performance of the driver itself. You're seeing the result of the housing of the headphone, the pads, the crossover design of the IM or even EQ. So in this measurement, this is literally the exact same IM. So the exact same driver, but the decay patterns look slightly different because I've manipulated the frequency response with EQ, which is again just kind of pointing out that the patterns you see in these graphs is not just a reflection of the driver's speed.

[00:10:01]And then the last problem I want to talk about with CSDs is the windowing.

[00:10:05]Basically, these are parameters and how you visualize the data, which means there's kind of like an infinite number of ways that you can visualize the data.

[00:10:13]And what happens is that the conclusions that you're likely to intuitively come away with are going to vary depending on which parameters you choose, right? To wit like this is the exact same data that we were looking at before, but it probably looks a little bit different, right? You don't see those resonances and the treble frequencies. It looks like everything is decaying pretty quickly and we're just left with a little bit of bass slowness, which is maybe what we would expect. But just by taking that exact same data, literally the exact same measurement and changing the visualization parameters here, it looks like the base is a lot slower and there's like a secondary resonance that's happening. Which one of these is true? Or we can change the parameters here and it looks like everything is kind of decaying and falling off at exactly the same time. Again, all of this is exactly the same data, the same exact measurement. I'm just manipulating the windowing settings to change the visualization which is I think contributing to again this sense that a lot of the CSD reading sounds like or yeah just reads like palm reading to me where feels like people are starting with conclusions taking the data manipulating it visualizing it however they however they want to support that conclusion and that's not really exactly how data should be working right there should be some sort of standard around this is these are the parameters that we use for visualizing the data these are the patterns that we're looking for and you can start from the data, recognize those patterns, make predictions, and then go out into the real world and validate those predictions. I first heard about Juier the same way a lot of people hear about new brands in the IM space. Just some underground hype slowly building up on Reddit, Discord, and HeadFi. By the way, this is an ad.

[00:11:50]Unfortunately, a lot of the time, this kind of hype turns out to be a dud. But something about the Juicier 41T called out to me. Probably it was the clear resin shells. I'm kind of a sucker for that. So, I bought it and against all odds, I actually really liked it. Now, years later, Juier is a wellestablished maker of IM with products from $50 and up, depending on your budget and sound preferences. They've got some of the most popular collabs like the Juier Cross Zeus Defiant and the Tuned with Squiglink Harrier. But, they've even got more IM that are aiming to balance performance, aesthetics, and price like a symphony in perfect harmony. That's their words. Check out Juier's full lineup linked in the description below, including frequency response graphs as measured on my rig. Okay, back to the show. Now, like I mentioned, until now, I've not really been engaging with CSD data. So, that level of analysis and prediction, I can tell you flatly I can't do. But that doesn't mean that nobody can. So, that was why honestly I surveyed my audience, right? I started by asking them what are the headphones and the IM's that people generally consider to be fast and then I surveyed them with this which is basically a poll with three different CSDs of three different IMs. One of these is a planar, one is a dynamic driver and one is an all balanced armature set. Basically trying to figure out can people look at these CSD graphs and predict the one that looks the fastest and would that correlate with the ones that people said sounded the fastest?

[00:13:15]And to be honest, I kind of just expected the votes to be completely random, right? Like basically 33% here, 33% here, and 33% here. But an interesting thing happened that there was a very strong winner. Option B, people agreed looks the fastest. So remember when I was talking about that first survey and the things the IM and the headphones that people recommended, people often cited planer magnetic IM and headphones as being fast sounding.

[00:13:40]In fact, specifically, the Lehore S12 was an IM that a lot of people called out as being a fast sounding IM. So, that was one of the IMs that was included in this measurement lineup. In fact, I'll just tell you right now, the option number one was the Planner. That is the Lit Shore S12. Option number three was an all balanced armature set.

[00:13:58]This is the RSV Mark II, which is actually a little bit expensive, around $500 or $600. So, one of the more expensive options, but that was not the one that won. In fact, the option that won was a dynamic driver set. But it wasn't just any dynamic driver set. It was in fact a $7 dynamic driver set.

[00:14:18]This is the JVC FX7 Gummy. And by far, to my surprise, this is the one that based on the CSD graphs, my audience thought looked the fastest. And if we go back to the survey where I asked people for recommendations for the IMS and the headphones that they think sound fast, like perceptually sound fast, it was mostly again planer magnetic drivers and it was mostly things that were kind of expensive. Neither of which are things that you could say about the FX7 Gummy.

[00:14:47]In fact, I can tell you right now from having listened to this IM, this is a sound that I don't think anybody would describe as sounding fast. And yet looking at the CSD graphs, this is the one that people kind of predicted would be the fast sounding one. So I don't know exactly what conclusion to draw from this. I mean maybe the the data about speed is just not really there. Or maybe there's just, you know, no agreed upon windowing settings, right? There's not a culture of analyzing this data.

[00:15:15]Maybe there are people out there who can look at CSD data and again make those predictions about things that will sound fast, go out and hear them, and then confirm or validate whether or not they sound fast. But again, I haven't seen it. Okay, so no luck with CSDs. But another way that I've seen people try to measure driver speed is with square waves. So if you've ever seen a sound, a sine wave, right? This is what sound is.

[00:15:36]All sound is these swoopy curly lines like this, squiggly lines. Um, but there is a concept of a square wave, which looks like this. And rather than it being kind of like curvy and bendy, it goes up really straight and fast and then it stops really suddenly, which seems pretty convenient, right? We're we're trying to measure how fast the driver can start and stop. And it kind of seems intuitive that a visual like this would help us do that. In fact, you can you can take measurements with a square wave. And you end up with results that kind of look like this, right? This is no longer square, which kind of indicates that this driver is struggling to produce a square wave, which is then suggesting that, you know, the speed of the driver is inadequate. So, I think that's kind of a intuitive understanding, but in the past, I've been told that that's wrong. I didn't really understand exactly why. So, I talked with Craig Stark, who is a university professor at the University of California in Irvine.

[00:16:29]>> He's got all of that experience. I don't have any of that. I do have though something he doesn't have, a whole bunch of detached human ears.

[00:16:39]And he's also a lifelong audio file. In fact, he has his own YouTube channel where he dives deep into some audio specific topics. I'll have linked I'll have him linked in the description down below if you want to check him out. But he basically told me that that was completely wrong. So I presented before a square wave as an alternative to a normal curly squiggly wave. But that is not at all what is going on here. So effectively a square wave is a bunch of waves kind of stacked on top of each other. Basically you you start with a fundamental frequency. In fact 300 hertz seems to be the common one used for headphone measurements. You start with a fundamental frequency and then you layer on top of it the harmonics of that frequency. And as you take this frequency plus this frequency, you end up with a resulting waveform that looks like this. And the more harmonics you add into it, the closer that this line starts to look to a straight line. And that's where you end up with something like this square wave. Basically, this is just a zoomed out view of those individual frequencies. In fact, if you zoom back into them, you would see that this line is in fact not actually square. it is. Yes, it rises quickly because it's going up in amplitude quickly, but that would also be true of really any frequency. Uh there's nothing really special about that in particular.

[00:17:51]What's really special is that this burst of sound is effectively all frequencies being played kind of at the same time.

[00:17:59]So effectively, when we're looking at a square wave measurement of naim or a headphone, we're just kind of looking at its frequency response, but in a more unintuitive way. And to make that point, look, I've got the JVC FX7 in the flesh here in my couple right now. We're going to show you what the uh the square wave result looks like in real time. So, this is a real-time measurement of the FX7 uh playing a square wave. And you can see this this is looking a little bit on the slow side, but all I have to do to correct it is apply an EQ. That's all that's going on here. And now you can see that that square wave suddenly looks a lot more square. But I didn't change anything about the driver, right? I didn't make the driver faster or slower.

[00:18:40]I just corrected the frequency response.

[00:18:41]So, Craig pointed out, there's kind of a missed opportunity here to demonstrate that a square wave really is just the sum of a bunch of harmonics. Uh, and what we can do is we can jump back over here to Rue. We'll go back to playing the square wave here. And, and we've got our our little analyzer here showing us what's going on in real time. And like I was saying before, I can apply an EQ here to make that sound make that more square. But what's interesting here is is this this low pass filter that I can also apply to it. So basically what this low pass filter is going to do is as I drag it to the left we are slowly cutting out the higher frequencies so that you're just left with lower frequencies. And you can see as I do that the waveform starts to look just like a traditional waveform. And you can see how it starts to build up and starts to look square the more waveforms kind of added into the mix, which is what happens when I'm dragging it to the right. But then I'm dragging this lowass filter to the left. And you can see that it starts resulting or just kind of reducing it to just one basic sine wave which kind of demonstrates what we were talking about that square wave is just a sum of a bunch of frequencies which is why frequency response of a headphone or an IM ends up being the the large determining factor in terms of what you see in the square wave response. So, as it turns out, both CSDs and square waves are effectively dominated by frequency response, which is kind of funny because it means that the people that are using this data, looking at this data, and trying to intuit it driver speed from it, are really just analyzing frequency response, but in a pretty cludgy and indirect way. It would be kind of like if I was trying to figure out if I'm allergic to a cake by looking at pictures of the cake and studying the lumps in the cake rather than just looking at the ingredients list, which would probably tell me whether or not I'm allergic to it. So, look, I don't know that frequency response is literally the only information that's presented in square waves and CSDs. I just know that it's the vast majority of it. It's possible that if you removed frequency response from the equation, there might be some interesting data there. I just haven't seen it. And from what I can tell, the smartest people I know also haven't. All right, Craig, why don't we add a little bit of a disclaimer here about this talk of CSDs being effectively just frequency response. That is true in the world of minimum phase devices, which is what headphones and IM typically are. Now, if the term minimum phase doesn't mean anything to you, join the club. It's a term I'm still struggling to fully understand, but maybe I'll do another video about minimum phase devices. But yeah, if you're you're wondering why are CSD graphs apparently useful in measuring speaker systems and I don't know cathedrals but not that useful in measuring headphones and IM, it's because of this concept of minimum phase and I guess non minimumphase devices. So the sensation of speed is real in the sense that like multiple people will report it like a lot of people will report this sensation of speed. It's just not literally the speed of the driver. It's just the frequency response that the driver is producing which isn't entirely disconnected from the sensation of speed or the concept of actual speed because if you want to be pedantic what actual speed of sound is is frequencies right how fast a driver is is determined by the maximum frequency that it can play because like I mentioned earlier frequency is time right 20,000 hertz is the driver moving 20,000 times in the span of a second if a driver can play 20 hertz It can also move pretty slow, just 20 times per second. And maybe that's kind of unsatisfying because like I mentioned, a lot of people, myself included, will have a perception, a sensation of sound being fast or sound being slow. But I think there's another there's another explanation and it's frequency response. So if we go back to that original data set where I asked people what are the headphones and the IM that people perceive as fast, right?

[00:22:32]There was some correlation with the type of driver. Planers were generally considered to be fast. There was also some correlation with price, right? Some of the most expensive products were the ones that people frequently cited as being fast. But the thing I didn't show you was that there's also a much stronger correlation with sound signature. Right? Things that sound bright, that sound lean, or that sound neutral are generally the products that people describe as sounding fast. And the things that sound warm and bassy are generally the things that people describe as sounding slow.

[00:23:04]And as it turns out, this is something this is exactly what you can predict with frequency response graphs, right?

[00:23:09]So you can take an IM like this, the JBC FX7 gummy. You can measure the frequency response. You can see there is an excess of midbase here. There is kind of a recession here in the upper mid-range.

[00:23:19]You could look at this and predict this will probably sound a little bit slow and it does. Whereas the S12, the Lit Shore S12, this is a planer IM that a lot of people recognize as sounding fast. You can look at the frequency response say, well, there's actually kind of a lot of excess trouble frequency here. This one might sound fast. And in fact, I think that's exactly what's going on. So, I guess the point I want to make is that I understand why people use these terms to describe sound. Why they describe sound in terms of speed, in terms of being fast, in terms of being slow. These are terms I might even use myself in the same way that I might use the word warm to describe sound. Right? It's not that I think the sound literally is warm to the touch. It just it sounds a way that makes me think more and the way that certain things sound fast. It's just not a result of them actually physically being any faster than other sounds.

[00:24:08]Because again, if it goes up to 20,000 hertz, it's basically as fast as any other headphone is and as fast as you can hear. um CSD graphs and square waves, they might actually have some useful diagnostic purposes, but in the way that they're commonly interpreted for determining driver speed. It's just not it. It really is kind of all about frequency response. And uh if you like that, or even if you didn't like it, if you like this video, hit the like button, subscribe to the channel, ding the YouTube bell, join me on Discord, and I'll catch you in the next super review. Cheers.

[00:24:44]Yeah.

[00:24:46]Yeah.

[00:24:48]Yeah. This review is super Yeah.

[00:24:54]Yeah.

[00:24:56]Yeah.

[00:25:13]Join with the force of reviews. We now have the tools to brave the misleading world of audio fools. Hey, and so are you. Grab your headphones, sniff a graph, and share your thoughts in this pursuit.