Measurements and Frequency Response - How to read headphone graphs

Introduction

Many of us entrenched in the audiophile world are familiar with frequency response measurements and graphs, but for those who are new to all of this, it’s worth understanding how measurements and frequency response correlates to sound quality. Reviewers like myself make use of graphs to try to communicate or give a visual indication of certain aspects of a headphone’s tonal balance. In theory, visual indicators like this are great and can help inform prospective buyers about whether or not they’ll enjoy the headphone in question. Unfortunately publishing graphs also comes with a certain responsibility, because these graphs and measurements are far too easily misinterpreted. 

You may have seen a graph of a headphone get posted somewhere only to have it be received with a certain amount of criticism, and while this kind of evaluation is fine, it’s important that we properly understand what’s going on with measurements - or at least understand the basics before making claims. For the current purposes of this article, I’ll be focusing on over-ear headphones and not in-ears.

As a general note, this subject goes very deep and there’s always more information becoming available and so the following will serve more as a general overview than as a comprehensive resource. I should also mention that the following article lists information that my experience in this subject has let me to understand so far - there's still much more to discover. Like with everything, the more you learn about something the more you realize how much more there is to know, and therefore I will also update this article in the future with new or revised information.

The Basics - What is Frequency Response?

Frequency response measurements are a visual representation of the headphone's sound pressure level (SPL) across the full frequency spectrum for human hearing (typically 20-20khz). In other words, these measurements represent the amount of energy (volume) each part of the full frequency range has. This allows us to get an idea of how the headphone may sound. If you see an emphasis below 200hz, it’ll likely have more bass, if it has an emphasis above 6khz it’ll likely have more treble energy, and so on. But it's also important to know that looking at measurements is not a substitute for actually listening to the headphone, and at best it's just an indicator or predictor for how it might sound.

Limitations

Graphs can tell you a lot about a headphone's overall tonal balance, and while these measurements contain much more information than "bright", "warm", "neutral" and so on, there are a number of limitations to graphs - at least as far as interpreting them goes. At the moment, graphs should not be used to indicate what many have called the technical aspects of headphones.

This includes many audiophile terms like detail retrieval, dynamics, soundstage, imaging, speed, timbre and so on. To be clear, I'm not suggesting this isn't captured by the frequency response somehow, merely that we're not yet at the point where we can reliably correlate those qualities as they're experienced with features of the frequency response. There have been some efforts to do so with soundstage and spaciousness, where a number of correlations have been drawn, however there are still far too many outliers and counterexamples for this to be relied upon. There's also a strong argument that suggests this only identifies areas where soundstage has been 'faked', like cutting certain areas of the upper midrange to enhance the perception of stage (tuning tricks), rather an actually indicating large and spacious presentation overall.

In any case, graphs are still valuable for providing some indication of tonal balance, and importantly when they're taken on industry standard rigs they provide a key data point for us to build EQ profiles from.

Raw vs Compensated Graphs

When looking at measurements, there are two types of graphs for frequency response:

Raw graphs - These graphs show how the headphone measures without any compensation or target in mind. Raw measurements for headphones should never be a flat line across - at least if we want the headphone to have a clear and balanced sound for most people. This may seem counter-intuitive, because we’re used to seeing measurements relative to a flat line in many places, but when it comes to headphones those are usually compensated measurements. 

Due to the various gain factors the human ear imparts to a headphone's frequency response, the general shape to a 'normal' frequency response will have a substantial elevation somewhere between 2-9khz. In other words, the physical shape to our ears amplifies certain frequencies. Unlike speakers, headphones are coupled to the side of the head (for over-ear headphones) and because of this we have to consider what those ear-related gain factors are - ideally achieving a frequency response that matches these gain factors to a certain degree, since this is what the brain expects to hear.

Here’s an example of a raw frequency response graph using the Sennheiser HD6XX:

When looking at measurements, make sure you know if it's a raw representation, and then also know what that raw graph should look like for the various sound signature targets you may be looking for.

Compensated graphs - For those new to the hobby, these graphs will be more useful. Compensated measurements show how significantly the frequency response deviates from a specific target. For this measurement, a straight line across is desirable, however it's important to understand which target is being used for the compensation

Here is an example of a compensated measurement using the Sennheiser HD6XX:

This is the same measurement as the previous raw graph, just using the 2018 Harman target as its compensation. If we want the headphone to perfectly match this target, then we want it to measure like a flat line straight across here.

Frequency response targets are usually devised in a way that imagines if a headphone's frequency response perfectly matches the target curve, it would sound 'neutral'. However because there's no current consensus on what neutral is - at least for headphones - we have a number of potential target curves to choose from. In other words, there are multiple different approaches to trying to find something like ‘neutral’.

If we understand what the compensation target is for the measurement we’re looking at (and what it might sound like), we can better understand what the deviations from that target mean. 

When a measurement deviates from a flat line on a compensated graph, we need to know what the compensation target is in order for it to mean anything. This is generally why raw measurements are more useful, because it will show us meaningful information regardless what people prefer to use as a reference or 'neutral' target.

Reference Target Curves

The most common reference target curve is the Harman target. This target was developed by Harman Research to best identify what kind of tuning in headphones people on average preferred. Compensations based on this target do happen to follow the general emphasis conferred by the gain factors for the human ear (with a rise starting around 1khz), but with extra bass emphasis and an overall balanced sound for a wide variety of genres. Note that this target doesn’t strictly match the ear-related amplifications we see between 2-9khz, but rather this is based on preference - what people actually wanted their headphones to sound like.

The 2018 Harman target looks like this:

There are a few things to notice about this target curve. The first is that there’s a substantial bass elevation below 150hz, followed by a somewhat contoured lower midrange and then an elevation around 1khz for what’s called ‘ear gain’ - what I’m referring to here as ear-related gain factors. This area is where these ear-related gain factors start to impact and amplify the sound before it reaches the ear drum, and this is why an elevation in this region is desirable. Some communities refer to this effect as 'pinna gain', however the pinna flange itself is only one part of what contributes to the overall amplification.

This elevation is correlated with what our brains expect to hear because we’ve lived our whole lives with ear-related amplification going on, and therefore this kind of elevation sounds normal to us. In other words, our brains normalize ear gain, and the absence of it in headphones will end up sounding muted and muffled by comparison.

What’s interesting about this target is that since it’s based on preference research, the contoured lower mids and substantial bass elevation aren’t actually correlated with any ear-related gain factors, since those are all at 1khz and above. This means that an elevated bass response isn’t necessarily ‘neutral’, ‘correct’ or even optimal for human hearing when considering the effects of human ear anatomy on sound, but rather it’s there because we like it. But does that mean it shouldn’t be there? There may also be a multitude of additional factors like the room, torso and head that contribute to certain expectations for what sounds 'normal' but this is up for debate - more on that later.

In any case, for those who don't like how significant the bass shelf is on the 2018 target, there's an earlier version of the Harman target published in 2013 has a more modest bass response, with slight differences to the upper mids and treble as well:

Notice how there’s a bit less bass and a bit more 3khz energy, as well as a slightly rolled-off treble response on the 2013 target. There are likely a number of factors that come into play that account for the differences, but in both cases, this is a fairly safe target, and headphones tuned to this target will make most music sound good to most people. Strangely, very few open-back headphones match this target in the bass, but many high end headphones are getting closer and closer to this tuning rather than the more traditional diffuse field target (discussed later on).

There are valid concerns about the information below 1khz with the consumer preference targets, since some may be looking for what I’m going to call ‘anatomical neutral’ - or in other words, a target that only elevates at the ear-related gain factors above 1khz. Remember that headphones are coupled to the sides of the head, and so while there may be additional gain factors to assume for what sounds most 'normal', the physical ear and ear canal don’t amplify the lower frequencies.

We may want to eliminate all the additional preference-based gain factors below 1khz for an alternative target that doesn't have the Harman bass shelf, but at the same time, this target’s upper mids and treble are also based on consumer preference research, and so they’re not strictly ‘anatomical neutral’ either. It’s understandable that we’d look for this kind of frequency response because at minimum the Harman target's preference-based elevation correlates to a certain extent with where our ears actually do amplify frequencies above 1khz.

Both targets sound fine in my opinion, but there are a number of key considerations to think about when looking at a graph that’s using the Harman target as its compensation.

Key Considerations:

1. Just because this is an average of what people prefer, that doesn’t mean it isn’t some form of neutral. This target curve has been criticized in some circles, largely for its bass emphasis, and I personally also do find it to be too much for what I enjoy (at least the 2018 bass shelf is too strong for me). 

Here’s where I’m going to be resistant to the notion that “the masses are wrong and I’m right”, or “the masses are all just bassheads who listen to terrible music”. Even if you don’t enjoy this target personally, it’s important to think about this curve as an average. In some ways, this could be thought of as ‘consumer preference neutral’, even if not strictly ‘anatomically neutral’ to the gain factors our brains expect to hear.

The Harman research was conducted across a number of demographics that had different preferences, and yours may simply not align with the standard curve shown above.

2. Many high end open-back headphones don’t have the bass shelf seen in the Harman target, and this is okay. Even for headphones that are well-extended in the bass all the way down to 20hz (like many planar magnetic headphones), most aren’t as elevated as the Harman bass shelf suggests - certainly not for the 2018 target. 

This may have to do with tuning parameters for open-back headphones, but what’s important to consider when evaluating frequency response is that for open-back headphones it’s quite common to see some deviation from the Harman target in the bass - even on great sounding headphones. This is generally not a problem because most tonal information in the bass is above 60hz anyway. It may be nice to have a sub-bass emphasis, but it's really not a problem if it doesn't match the target's suggested elevation.

3. The Harman Target does not reflect more granular deviations that should exist in a headphone’s frequency response. There are certain elements of the frequency response that should show a dip or elevation that will look like a deviation from the target, but the problem here is more to do with the target being strongly averaged than an issue with the headphone in question. 

There are a number of reasons for this, like certain behaviors of headphones on measurement rigs - different headphone build styles and so on. But there are also certain interactions with the human ear that the reference target doesn’t properly include. An example of this is that there should be a dip (deviation) somewhere between 9-10khz on most headphones due to specific parts of the ear (concha), and this isn’t indicated by the target curve. 

To the average reader, this dip may look like it’s a problem because it will deviate strongly from the target, and at the same time if there’s an elevation in this range that does match the target, it will look like it measures well. In both cases the reader will have come to the wrong conclusion about the headphones in question.

Some publications have addressed this problem by heavily smoothing the frequency response, but this is also misleading as it can make other aspects of a headphone's frequency response look better than it actually is. The solution in my opinion is to ensure the reader doesn't scrutinize the more granular aspects of a headphone's frequency response, but rather take a wider view of the whole thing - or at least stick to various frequency ranges than individual ones. This is a matter of educating readers and providing key context for the measurements being presented.

Other Targets

A diffuse field target aims at a frequency response that's meant to emulate the way flat-measuring speakers would sound in a somewhat 'lively' room (as opposed to an anechoic chamber). Unlike the Harman target, this doesn't base the result on consumer preference-based research. A diffuse field target ends up being slightly brighter than compensations based on the Harman target, and doesn't have as much bass emphasis, but it's still warmer and more realistic than a Free Field target developed in a completely 'dead' room. Diffuse field has been a common target for many years, however more recently headphones have tended to aim for the extra bass emphasis of the consumer preference curve instead.

Which Target is Best?

So which target should we prefer when considering a reference tuning? Or in other words, if we do want to represent the optimal frequency response as a flat line on a compensated graph, which compensation should we use?

My answer to this question is that for now we should probably be using some version of the Harman target - at least for the mids and treble - and not the traditional diffuse field or free field targets - and I think there are good reasons why the consumer preference targets have been used for evaluation purposes. 

There may be some resistance to this notion, partially due to the fact that these targets err more on the side of ‘safe’ rather than necessarily ‘optimal’, and perhaps people would rather attach the notion of ‘reference’ to something more concrete than an average consumer preference target to demarcate their own preferences from those of a more general consumer audience. But I don’t think any of this matters when you consider that the result ends up being the most agreeable for most people. For those of us who are outliers, it's our responsibility to understand where our preferences differ from the reference target, and then make conscious adjustments from there.

An important concern about these consumer preference targets, however, is whether or not the bass shelf found in both the 2013 and 2018 Harman targets is appropriate and whether it should be included as part of the reference target or not. Insofar as what the human ear does to the sound, the brain doesn’t have any expectation for the contoured lower mids and bass shelf found in both preference target curves. As mentioned earlier, there may be an expectation for this that comes from other places, but it's not likely to be part of the brain's normalizing of physiological gain factors - at least this is what my current investigation into this subject has indicated.

In my mind there are two ways of looking at the bass and lower mids issue. The first is to simply use what sounds best as the reference target. For me, this ends up being a combination of the 2013 Harman target for its bass response and the 2018 target for its upper mids and treble.

 

The problem with doing this is that it’s still anchored to subjective preference, and in particular there’s no correlation between that preference and any physical gain factors that our brains expect to hear below 1khz. As a matter of representing a reference target this way, it may be useful for providing a target to EQ to (to get it to what most people prefer), but perhaps less useful for showing how strong the bass is for a given headphone. Remember that on a compensated graph, the Harman target will cause most open-back headphones to look like they roll off in the bass, even though they're well extended all the way down to 20hz. In these cases, these headphones just don't have the bass shelf suggested by the target.

The second way to look at this is by saying we can ignore everything below the ear-related gain factors - so everything below 1khz. The main benefit to this approach is that when we see deviations in lower frequencies, we can assess whether that fits with our preferences. So a bass shelf that would normally fit with one of the Harman targets would actually look like a bass shelf relative to a flat line, and this means we’d be able to tell if the headphone is bassy or bass light and so on. The other advantage of representing it this way is that as mentioned above, many high end open-back headphones don’t follow the Harman bass shelf, and this means they won’t look like they have as strong of a roll-off in the bass.

The downside of doing it this way is that it makes comparative aspects to other measurements - even when taken on the same rigs less viable, and importantly for anyone wanting to EQ their headphones to this target, it won’t sound as good to most people (remember that the contour and bass shelf seen in the preference curve is what most people prefer). 

The bottom line when deciding which target curve is best is that it’s a complicated issue that introduces many questions about what’s most prudent for anyone looking at the graphs - for anyone trying to figure out which headphones they will end up enjoying the most, and that ultimately there is no strict ‘neutral’. But with that said, I find there to be significant value in using a target that can also be helpful for those of us who want to EQ our headphones and get them to an optimal tuning - for what most people prefer. This way these graphs can be suitable for anyone looking for a starting point. 

This is also why it’s important to at the very least include the raw frequency response measurements, and not just show compensated targets. It may be a bit more confusing initially and also require some explanation for how to read it, but the more widespread understanding there is of raw frequency response measurements and that headphones don’t actually measure flat, the better.

How to Read a Frequency Response Graph

When reading frequency response graphs it’s important to mainly consider the overall curve and frequency ranges rather than looking at individual frequencies. The reason for this is that we don’t actually hear individual frequencies very well within the context of the broader range - at least when it comes to strong dips in certain areas. So for example, if you see a sharp cut in one particular area of a headphone’s frequency response, this will not be as audible as the graph might make it out to be.

Sometimes there are also other reasons why a headphone will measure a certain way. In particular, many high end Audeze LCD headphones show a strong cut around 4khz. To my ear these headphones don't sound like they have as extreme of a dip as the measurement shows, and so this may just have to do with the headphone's design style and its interaction with measurement rigs. Moreover, it's also imprudent to key in on specific recessions that may be there due to interactions with the physical ear like the 9-10khz dip that should exist for most headphones mentioned earlier.

When a headphone’s frequency response shows a number of deviations that elevate relative to the target, this may be more of an issue, but it’s also important to once again consider the frequency ranges around these elevations. Are they also elevated to a similar level? If that’s the case, the peak in question won’t be as pronounced given that you’ll also be hearing everything else that’s at a similar volume level.

Let’s use the HiFiMAN Arya as an example:

If you look at the upper mids and treble, there are three notable elevations to consider here. One at 4khz, one at around 8khz, and one at 11khz (also note the dip around 9khz that should be there likely due to concha interaction). A person could be forgiven for thinking that this would sound sibilant or ‘peaky’, because the reference curve used here (Harman Combined) is quite a bit more smooth throughout this range.

But in reality, this headphone doesn’t have any issues with harshness in this range - at least not the current revision - in part because of how these elevations complement one another. This is helped by the fact that these elevations are suitably wide, and this will often yield a smoother sound than if those elevations were narrow peaks. Moreover, this frequency response may be a bit brighter relative to the target, but just because the frequency response doesn’t match the target perfectly, doesn’t mean it’s not still balanced.

Let’s look at the frequency response of the Focal Clear next:

For the most part the Clear has an agreeable frequency response, with perhaps a bit of weirdness in the treble. The first thing to notice is that it looks dipped around 4-6khz - at least relative to the target. This should be a bad thing, right? 

Leaving aside for a moment the question of why this looks a bit dipped, what we actually should consider here instead of just focusing on the dip is the relationship between 4-6khz and the lower range of 2-3khz, as well as its relationship to the range above it. Notice how even in the target 3khz is higher than 4-6khz as well - not to the same degree, but there is a general similarity there.

Then ask the question, how does this relationship in the frequency response impact the music I might be listening to? This should cause us to consider instrument tones that token these frequencies for both their fundamental and resonant harmonic/overtones. Maybe you’ll find that 3khz here is a bit too strong relative to the ranges around it, or maybe you’ll find it to be exactly where you want it. But the fact that it shows a deviation from the target curve here isn’t on its own a bad thing - it’s maybe not perfect, but it’s not really a significant problem either.

Now let’s look at the peak at 8khz. In my mind this is more of an issue, and I find this to be responsible for some of the occasional grittiness that shows up for consonant tones in vocals like ‘S’, ‘F’, and ‘T’ (you can confirm this by using EQ with a narrow filter to smooth out that peak and the grittiness goes away). This is because the relationship between the 8khz elevation and the ranges on either side of it is not as smooth as it ideally could be. This won’t show up on all recordings, really only some of them, because the precise range of those tones will vary from one recording to the next.

The type of elevation seen here at 8khz can have the effect of drawing more attention to certain elements in the mix as well, and it’s not as though this is particularly sibilant overall since this whole section is still a bit more relaxed. But because this elevation is so narrow at 8khz it can on some recordings emphasize the edges of certain tones above tones that token counterpart ranges on either side of them.

So to conclude, when evaluating frequency response, look first at the overall tilt relative to the target - not whether it matches the target perfectly or not. Then look at various frequency ranges and their relationship to the ranges around them. This will indicate tonal balance. And lastly, when looking at significant deviations from the target, consider how narrow those deviations are and the frequency ranges around them. More often than not, something that looks sub-optimal doesn’t actually sound that way.

Let’s Talk About That Bass

For some reason bass level and elevation seems to be a controversial subject. I think maybe because we all have such varied preferences for how we want our bass to sound. When looking at the Harman target, however, there’s a very good reason for why the bass elevates at the place that it does, namely below 150hz. 

The target indicates that people generally prefer a decent amount of bass presence, but importantly that we prefer bass that’s distinct from the midrange. This may be due to various assumed gain factors that are not part of the ear, maybe because we’re used to being in certain environments or hearing music in those environments, but also maybe just because we like it. Regardless, having the contoured lower mids and then distinct bass shelf below 150hz allows the bass to come through in a satisfying yet clear manner.

I’ve come across a number of good reasons why some may prefer to have the lower mid contoured section to be linear and filled in instead, and this preference may also align more closely with open-back headphones than closed-back headphones. In general it stems from the desire to have a more “full bodied” kind of sound - but then also because certain recording styles simply don’t token the lower bass frequencies, meaning in order to get those recordings to sound ‘full’, we’d need a headphone with a more filled in transition between bass and lower mids.

But in either case, we can get a sense of whether or not the bass of a headphone will be distinct by two factors. A) The overall volume level (sound pressure level) of the bass shelf, and B) how far up the frequency response the bass shelf stops. So for example, if the bass elevation continues up to 300hz and doesn’t come down until the lower mids, there’s a tendency for this to sound kind of muddy and thick, as it bleeds into the midrange a bit. By contrast, if a bass elevation stops around 150hz and comes back down to where the mids are (or even a dip in that area), the bass will sound more distinct - and potentially more emphasized. If there is no bass elevation and it's a flat line across, we have to look at where the level is in relation to the upper mids and treble. There’s a good chance this will also have the bass sounding distinct, but not quite as distinct as the bass shelf suggested by the reference target.

Whichever style is preferred will of course be a matter of preference, but when you see the kind of bass elevation like the one pictured below, you can assume that while present, it won’t be as distinct as it would be if the shelf dropped down again around 150hz:



The bottom line (see what I did there) for bass and lower mids is that you have to ask yourself what kind of bass response you want. If you want a more full bodied bass to lower mid transition, maybe you want a flat line all the way down. If you want a very distinct bass response, look for a shelf that follows the Harman elevation. And lastly, if for some reason you like boomy and overbearing bass (hey, some people do), then look for something a bit more like the graph of the Sony WH-1000XM4 pictured above.