Author: geoff

Bang & Olufsen H6 headphone reviews

 

I was part of the development team, and one of the two persons who decided on the final sound design (aka tonal balance) of the B&O H6 headphones. So, I’m happy to share some of the blame for some of the comments (at least on the sound quality) from the reviews.

 

 

 

Audio.de reviewed the H6 paired with an Astell & Kern portable player. They said:

“Tatsächlich aber ertönte der H6 mit dem AK Junior erstaunlich präzise, extrem sauber und stabil und vor allem ungeahnt luftig. Das Zusammenspiel des speziell angefertigten Treibers mit dem ausgeklügelten Bassport stellte nicht etwa – wie oft üblich bei geschlossenen Hörern – die tiefen Frequenzen wummernd und brummend in den Vordergrund. Nein, besonders Stimmen und feine Details profitierten vom knochentrockenen und nicht zu gewaltigen Bass.

“Susanne Sundfos glockenhelle Stimme beispielsweise stand fest gemeißelt im erstaunlich großen Raum, umgeben von jederzeit verfolgbaren Bassdrum-Beats, echten Streichern und Synthesizer-Harmonien. Der kurze Probelauf mit H6 und AK Junior wurde zur ausführlichen Hörsession, die erste Begeisterung zur echten Liebe. Eben true love.”

Correction: That review stated that I said that the H6’s were tuned using the Grado’s as the reference. This is not really true. While we were tuning them, we listened to many different headphones. The Grado’s are one of the many hanging in the listening room…

 

Gramophone Magazine reviewed the H6 in the April 2014 edition. They said some very nice things about the headphones:

“…excellent clarity and weight, well-defined bass and a sense of openness and space unusual in closed-back headphones. The sound is rich, attractive and ever-so-easy to enjoy.”

and

“… by no means are these headphones designed only for those wanting a pounding bass-line and an exciting overall balance: as already mentioned the bass extension is impressive, but it’s matched with low-end definition and control that’s just as striking, while a smooth midband and airy, but sweet, treble complete the sonic picture.”

and

“As I may have made clear in the past, I haven’t been the greatest fan of headphone listening, much preferring a pair of small speakers on the desk. But with the arrival of fine headphones such as the BeoPlay H6, I’m having to do some re-thinking.”

 

Bobby Solomon wrote this review at thefoxisblack.com.

“The sound is refined, with the midrange coming through clearly, and the bass and treble are balanced perfectly.”

 

Kenneth Roberts wrote this review at head-if.org.

“I’d describe B&O’s “house sound” as natural and neutral, with a brilliant, feathery-light high-end that resolves a lot of detail. This describes the H6’s sound perfectly. It delivers a staggering amount of detail in its price-class. In fact, the H6 delivers an impressive amount of detail when compared to headphones well above its price-class! Cymbals, triangles, snares, and hi-hats all sound crisp and light, with nary a hint of sibilance or stridence. I’m guessing this deftly executed high-end lends much to the headphone’s spectacular imaging, which I’ll describe later.”

 

T3’s website has a review here

“Which rather handily leads us onto how they actually sound. The answer can be summed up with the word ‘balanced’. By this we mean that almost all ranges perform excellently, but never does one take precedence over the other.”

 

pocketlint.com has their review here

from the middle of the text: “As you might expect, audio quality is top-notch. While we can’t say it competes with some of the other stay-at-home audiophile grade kit, for a set of headphones you can listen to daily, they certainly deliver. Sound is nicely balanced with plenty of detail and not an overly punchy bass. The set of 40mm drivers and the internal bass port just keep everything as clean and simple as possible.”

 

Tim Gideon and PCMag.com has their review here

The concluding words state: “In the age of big, booming bass, it’s doing its own thing. This is by no means an anemic-sounding headphone pair, it just favors lows, mids, and highs over a wildly boosted sub-bass range. If a more refined, crisp mids-focused sound is what you seek, the H6 will not disappoint, and it’s refreshing to see such a unique sound signature in this field.”

 

“Bang & Olufsen Finally Got It Right” at head-fi.org

“The sound is surprisingly flat, but a little bit on the warm side. The sound is not as airy as the open back cans, but the soundstage is very good for a portable closed back.”

 

whathifi.com has a thorough review here

“Based on my favorable impressions of the H6 with the first series of tracks which are heavy on electronic sounds, and also on the second group which feature more conventional bands and acoustic sounds, I’d say the H6 bridges these different genres very well. There are very few headphones that have a decent deep bass response and reasonable impact but don’t have any upper bass emphasis or bloat, and the B&O H6 is one of those few.”

 

soundandvision.com has a review here

“This is no head-banger headphone, but the bass goes low and it’s nicely articulated. The frequency response is remarkably smooth, without a hint of the rolled highs and boosted bottom common to more mainstream ’phones; the downside to this clarity is that the headphone won’t do anything to blunt the harshness of overcompressed MP3s.”

 

beforeitsnews.com has a review here

“The sound, however, is near flat and crisp. You’ll be able to hear the fine details of a song or a recording. There’s also no distortion at top volumes, while the midrange is great. Bass is balanced rather than strong, preferring accuracy and refinement to simple power. Think of these headphones as fine wine and other headphones as beers. When you start appreciating its beauty and fine sound, oh boy, where have you been all my life?”

 

 

Customer Comments at beoworld.org

There is a long user discussion on the forum here. Persons looking for real-world opinions will certainly get their fill at that site.

“I did not really like the H6 when it was released, a bit thin on an iPhone and a big headphone. And expensive. So I bought the B&W P5 instead. Quite good comfort and good sound. But.. I just had to give the H6 a try so I bought a pair a few days ago (the black model – after a lot of pondering). I now have a completely different feeling about the H6. They sound very natural and good. I like how detailed and precise the sound is.  You can hear so many details very well and how acoustic noises changes in frequencies in a wonderful way.”

“My wife has a pair of H6’s in tan, and I am consistently impressed by them.”

“Soundwise, also nice to hear your opinion (and Chris´s). I can’t remember hearing a pair of headphones with a more true sound.”

“I’ve now got a tan pair of my own (240£ for a new pair via eBay) and use them every day.  I’ve been listening to some of my favourite music which I’ve listened to for over 40 years on CD, vinyl, digital via an ipod and A8’s and can now pick out musical detail that up to now has eluded me.  The separation of instruments on some of my live recordings is incredible, so much so that I’ve found that even Steve Hackett playing live does make some mistakes!  None of the music is muddied and I have to say that I don’t find the volume when using an ipod too low.  If I want loud music I’ll play it at home on a bigger amp set-up with some bigger speakers.  When using the headphones I want the isolation of me and my music.”

… but what do they sound like out in the kitchen?

Typically, when you read a review of a loudspeaker, you’ll often see a graph that shows a measurement of its “frequency response”. This is a measurement of how loud the loudspeaker is at different frequencies at one position, directly in front of it, if you feed the same signal level into the input of the loudspeaker, and you are not in a room with reflecting surfaces. Usually a measurement like this is done by placing a microphone 1 m in front of the tweeter and putting some special signal (like a sine wave with a changing frequency or something called an MLS signal) into it. In (some persons’ ) theory, the goal of a loudspeaker is to have exactly the same output level at all frequencies (assuming that all those frequencies went into it at the same input level). In other words, output equals input.

However, that’s only a very small view of reality. For starters, this measurement is only done at one signal level. There is no guarantee that the loudspeaker will behave the same way if you measured it with a louder (or a quieter) signal. However, for the purposes of this discussion, we’ll take issue with another point. One big problem with this measurement is that it only tells you how the loudspeaker behaves at one point in space – and this is simply not enough information.

In reality, sound does not beam out of the front of a loudspeaker like a laser beam. The truth is that sound comes out of the loudspeaker and heads in all directions – left, right, up, and down. So, one question to ask is “what’s the difference in the sound that goes out the front, and the sound that goes out the side or the back?” Well, generally speaking – and this is VERY general – there is less and less energy in the high frequencies as you come around to the back of the loudspeaker. There are physical reasons for this that we’ll talk about later (or you could go look it up now, if you prefer) but we won’t get into it here.

So, let’s take a very simplified example:

front

 

Let’s say that the plot shown above is a frequency response measurement of a loudspeaker done directly in front of it, “on-axis” to the tweeter, 1 m away. As you can see, it has an unbelievably flat frequency response (I’m faking it…) with a roll-off in the very low end (50 Hz) and the high end (18 kHz). How would the same loudspeaker measure if we were to put the microphone at the same distance, but directly to one side,  90° off-axis? Well, it might look  something like this:

 

side

 

You’ll note here that the low end roll-off hasn’t changed – we have just lost high frequencies. Now let’s measure again, but this time, we’ll put the microphone directly behind the loudspeaker. In this case, we might see a frequency response measurement that look something like this:

 

back

 

So you can see there that we have the same effect – just more of it.

So, the first moral of the story here can be read two different ways:

Option 1: The further around the back of the speaker you go, the less high frequency information you’ll get – or at least, the quieter the high frequencies will be.

Option 2: Bass goes everywhere equally, but high frequencies tend to “beam” forwards.

 

But there is a different moral to be learned here. Usually, when you buy a lamp to hang on the ceiling to light up a room, you don’t think about how much light is beaming straight out of it, down towards one point on the floor or the wall. Usually, you think about how much light goes out in all directions at the same time – how much it lights up the room (instead of just one location in the room). The same is true for a loudspeaker. We can think about how much energy is coming out of the loudspeaker in all directions at the same time – in other words, how much energy is going out into the room, and not just what’s headed toward your left ear.

In order to consider this, we have to add up the frequency responses of the loudspeaker going out in all directions at the same time. For the purposes of this discussion, we’ll pretend to do only 3 measurements – for the front, side, and back of the loudspeaker – but let’s say that’s enough for now. If we add up the energy in the three frequency responses we saw above, we’ll get something like the one below:

 

total

 

On that plot shown above, you can see the three original measurements, and the result when you add the three of them together (the red curve). In our simplified little world here, what this shows is that, if you consider the sound coming out of the loudspeaker in all directions at the same time (the red curve) you can see that there is more energy in the low frequencies than the high frequencies. So, if we normalised the two measurements (in other words, make them comparably loud (in other words, align them vertically on the graph)) and compare them directly, we see the plot show below, which shows that the power response is generally more bass-y than the frequency response.

freq_vs_power_response

So what? Well, if you turn on the music in the living room, and you head into the kitchen to make dinner, you aren’t on-axis to the loudspeakers. So the sound that you’re hearing has very-little-to-nothing to do with the black curve. What’s actually happening is that the sound radiates out of the loudspeaker in all directions at the same time, bounces around the living room and leaks into the kitchen. So, what you’re listening to has much more to do with the red curve than the black curve. If we were being more geeky, we would say that you’re listening to the loudspeaker’s “power response” (because we’re talking about the sum of the total acoustic power put into the living room) instead of its “frequency response”.

That doesn’t necessarily mean that you should try to build a loudspeaker  with a flat power response – that’ll sound really bright. However, it might mean that the frequency response curve you see in the magazine review isn’t the only thing that you should worry about… It also means that if you’re building a speaker for people  who have more than one chair in the house (or people who have friends), you might want to worry about something more than just the frequency response.

And, just in case you think that I’m oversimplifying too much here, let me prove to you that I’m not. The plots above were built on fake curves, showing what happens when I add only three measurements of a loudspeaker, one in the front, one in the side and one in the back of the device. However, take a look at the curves below. These are two real measurements of a real loudspeaker. The blue curve is the on-axis frequency response measurement (note that this is an active loudspeaker that has been equalised to have a flat-ish on-axis frequency response). The red curve is the measured power response of the same loudspeaker which was found by making a LOT of frequency response measurements around the loudspeaker and summing the results all together to get an idea of what the device was doing in all three dimensions. Looks pretty similar to the fake plot above, doesn’t it?

 

2013-08-27 at 11-22-49

How to mis-represent your data OR Why I hate spider plots.

Introduction

Before I start my little rant, let’s do some basic grade-school geometry.

The area of a rectangle is equal to its length multiplied by its width. For example, if you have rectangle with sides of lengths 2 cm and 4 cm, then its area is 8 square centimeters (2 * 4 = 8). This means that if you have a square, then its area is the length of its side squared. For example, if you have a square with sides of a length 3 m, then its area is 9 square metres (3 * 3 = 3^2 = 9). In other words, for rectangles,
A = L * W
where A is Area, L is Length and W is Width. The only thing that makes a square special is that L = W.

The area of a circle is equal to π multiplied by its radius (the distance from the centre of the circle to its edge) squared. In other words, A = π * r^2 (say “pie are squared”) where A is area, π is roughly equal to 3.14 and r is the radius. So, if you have a circle with a radius of 4 mm then its area is approximately 28.27 square millimeters.

In the case of the square and the circle, if you double the width (or diameter, or radius), you quadruple the area. If you increase the width by a factor of 3, you increase the area by a factor of 9 (3 x 3). Stated generally, the area is proportional to the square of the width.

Representing your data with a bar graph

Now, let’s pretend that we have some data to show to people. We’ll start with something simple – we’ll display the total annual sales of widgets over 3 years. Let’s say that, in the first year, you sold 10 widgets; 20 widgets the second year and 30 widgets the third year. Your competitor, by comparison, only sold 10 widgets in each of the three years.

How do we plot these data? Of course, there are lots of ways, but one way that makes sense is to use a bar graph. A bar graph shows a single bar for each value (in our case, widget sales in each year), side by side, where all of the bars have the same width. The value is represented by the height of the bar. An example of a bar graph of our widget sales is shown below.

simple_linear_bar

For the purposes of my little rant, there’s something important in that last paragraph. I said that the height of the bars shows the data, but the width of all of the bars are the same. This means that the data are not only shown by the heights of the bars, but also their areas.

Mis-representing your data with a weird bar graph

What if we were to get creative and say that the data are not only represented by the height of the bars, but also their widths? On the one hand, you could make the argument that this is fair, since you could look at either the relative heights OR the widths to see the data comparions. However, if you take a look at the example shown below where I’ve plotted such a graph, you’ll see that this might not be a fair representation. Why not? Well, if your eyes are like my eyes, you don’t see the heights of the bars, you see the areas of the bars. And, since I’ve doubled the height and the width of the bar with double the value, the area is 4 times. In other words, I’m exaggerating the difference in the values by doing this (to be precise, I’m squaring the difference).

simple_area_bar

But I hear you cry, “Of course no one would ever do this! I’ve never see such a plot! Or, at least, I can’t make one in Excel…(although I can in Tableau…)” Read on!

Getting to the point… bit by bit

Let’s say that we did some kind of test where we want to represent a bunch of data points for various categories. For example, a listening test comparing four loudspeakers, where each loudspeaker was rated on 15 attributes. We’ll assume for the purposes of this discussion that the test was designed and conducted correctly, and that we can trust the data. We’ll also assume that the test subjects that produced the data (our listening panel) are experts and can rate things perfectly linearly. In other words, for a given attribute, if the listening panel says that one loudspeaker gets a score of 30 and the other one gets a score of 60, then the second one is twice as good as the former. We’ll also say that, for the purposes of this test, each attribute was scored on a range from 0 to 100. Finally, we’ll assume (for the purposes of keeping this discussion clear) that a higher score for any attribute means “better”.

So, we did our test and we got some strange results. (Note that these data are not from a real listening test. I made up everything to illustrate my point.) One loudspeaker got a score of 100 (out of 100) on every attribute. Another loudspeaker got a score of 50 on every attribute. The other two loudspeakers were a little more realistic (but still faked, don’t forget…)

stupid_bar

So, as you can see in the above bar graph, one loudspeaker got a score that was only half as good as the other in all categories. This is easily seen in the bar graph. If you squint just right, you can also imagine two rectangles, one big black one and one big red one. Since those two rectangles have the same width, their areas also represent the data accurately.

But, what would happen if we plotted exactly the same data using a spider plot? That would look like the figure below.

stupid_spider

Notice that the same data is plotted as before, but the message your eyes see is slightly different. You see the black circle and how it compares to the red circle. Since the red circle has twice the radius of the black circle, it has four times the area. If you’re like me, you see the comparison of the areas of the circles – not their radii. So, if you don’t force your brain to do a little sqare-rooting on the fly, the plot appears to say that the second loudspeaker is four times better overall, which is not what the data says. This is basically where I’m headed…

Now let’s look at the “results” for the other two loudspeakers, whose data was a little more varied. These are displayed in the bar graph and spider plots below.

big_bar
spider

This time, at first glance, things look a little more “normal” but look carefully, particularly at the results for categories 9 and 10. The same problem (of a spider plot showing the square of the results) happens again here. The big area between the red points and the black points for 9 and 10 exaggerates the difference in the actual data, which are better displayed in the bar graph. One way to think of this is that the “slice of the pie” gets bigger in area as you go further out  from the centre of the circle. The figure below shows the way my brain interprets the plot.

spider_area

Of course, if the data were for the same for all attributes for both speakers except for one attribute where one loudspeaker got a score of 100 and the other got a 50 (so you would just see a spike at one angle on the plot), then the spider plot would come very close to representing the data correctly. But this is because, with those weird collection of numbers, you come close to eliminating area in the plot – and it just becomes what most people call a “compass plot” which is something different.

The conclusion

As the title says, my conclusion is that a spider plot mis-represents differences in data because they show us something that it more like the square of the difference rather than the difference itself. To be fair, its representation approaches the square of the difference as all of the values for a given product become more equal (as I showed in the first spider plot with the two circles, above).

Personally, I prefer to have graphs that show me the results – not a weird scaling of the results – which is why I hate looking at spider plots.

Mostly, however, it’s because I’m too lazy to do square roots in my head.