Hi everyone,
after opening a new thread on the German hifi-forum, I've decided to translate my findings into English and share my findings with you
Prologue
First things first: a while ago I had this idea of doing what had been on my mind for a while, which was making discussions about headphones a little more objective. There are several scattered sources of information all across the web with a few websites providing frequency response graphs for all sorts of headphones. These measurements were often conducted using totally different methods of measurement which logically leads to different looking graphs for the same headphone. Production variations on a wide range of headphones don't add to the credibility and comparability of measurements. I think it's understandable that not many people put much faith in interpretations of frequency response graphs.
So I wondered how accurate those graphs were and how exactly they would compare to my own measurements. I did some research on the web for affordable and simple gear to help me do my own measurements. I stumbled upon a few stupidly expensive professional solutions until I came up with the Soundman OKM II Pop/AV Electret Earbud-Microphones. At ~$150 these are relatively affordable and promise to be usable as microphones for binaural recordings as well. And these things deliver, I can tell you that in advance. I've created some of my own binaural recordings which sound pretty convincing. But let's not get distracted, this is about frequency fesponse measurements
First of all, these things are not In-Ear microphones, so they'd disregard any sort of effect the inner ear has on the actual sound that arrives at the eardrum.
I'm not entirely sure whether this was taken into account by Soundman when developing these microphones, I just know that the recordings done by these are pretty good and allow the brain to reproduce spacial information quite accurately. These microphones however do not have a totally flat Frequency Response as indicated by the following pictures:
The measurements shown here were conducted by the Technical University of Berlin and should at least have some representative value. Sadly, I do not know which of those graphs ultimately belongs my particular model, since they're declared as "OKM II/K" and "OKM II/R", none of which seems familiar to me.
The actual measurements
In order to take those measurements, I put the mics into my ears and albeit feeling a little uncomfortable with the spiky construction underneath the cushion, I was able to get a half-way decent fit. I then hooked the mics to the microphone input of my soundcard and fired up two instances of Cool Edit. One of those instances was used as playback, the other was used as the recorder. Playback was done via my soundcard's optical out, my HA25D as DAC feeding the Corda Aria as the testing amp.
Firstly, I generated a few seconds of white noise since white noise seemed the most appropriate signal to do measurements with. It is a noise that incorporates all the frequencies from 20 Hz to 20 Khz across the spectrum in equal amounts.
I then played back that noise in an endless loop using the mics in my ears and the tested headphones on top of the mics. Not one of the most sensual experiences in my life, I can assure you, but sometimes uneasy things just have to be done for the greater good
For each headphone I first did several runs to adjust the volume to a uniform sound pressure level. I started off by recording a total of 10 seconds of white noise so that the average RMS (Root Mean Square) input came out at roughly -15 db. If the input level wasn't quite there, I repeated the process until I had a signal at roughly -15 db. This means that in contrast to the method Headroom uses for their FR measurements, the input level here won't be matched at 1 KHz only, but at an average sound pressure level of -15 db over the entire spectrum. I figured that this would make the graphs more comparable than the 1 KHz method. After all, headphones with a dip at 1 KHz would have a relatively higher FR curve in bass and treble (U shape) that mostly exceeds the curve of headphones with a bump in that area.
I then created a 30 seconds recording of white noise with the leveled volume. Afterwards, I normalized the recorded waveform to exactly -15 db RMS digitally. I saved the waveforms created like this and used Rightmark Audio Analyzer (RMAA) to generate the frequency response graphs.
Keep in mind that minor changes in fit and position change the graphs quite significantly. Especially the treble area beyond 4 KHz seems very sensitive to such changes.
The following headphones were tested using this method:
New: I've also done a few more measurements on the different variations of the MS-Pro and MS-Ultimate. Check the gallery under "Special: MS-Pro". There's also a special gallery on the Ultrasone PRO-900 burn-in-process and the effects of its placement on the head.
Long story short, the results are here:
Headphone Frequency Response Measurements
I've divided the gallery into three categories.
1.) "Headphones" has measurements of headphones the way I regularly use them (pad mod on all AT, keep that in mind!) and all the other headphones I had with me a while for taking measurements.
2.) "Amps" has measurements comparing different amps when using one particular headphone model
3.) "Mods" shows the differences a mod does to a particular headphone.
Things I see confirmed by those measurements:
- There's certain similarities among AT woodies, though there's still plenty of differences between the woodies and the open models to claim that there's no such thing as a general AT house sound
- BUT: most AT headphones being mids-centric headphones totally goes to show with these graphs (which is what I love them for)
- Solid state headphone amps only have absolutely minor influence on the sound signature
- The Pad Mod I did to my AT headphones definitely has the effect I felt it had
- Dry and boring headphones like the K701 and HD540 measure pretty flat with a slight emphasis on treble
- The HD600 and AD2000 sound and measure pretty similarly (but of course I still prefer the AD2000 :P)
- The AD2000 is pretty neutral and totally different from any other AT in the field
- The HFI-680 has a bathtub-signature
- ... and the PRO900 even more so
- The JVC HA-DX1000 has pretty honky mids due to a massive bump at 350-700 Hz
- The W2002 has more extended bass than the L3000
- The W5000 and W2002 being closely related in the AT ancestory goes to show in that they sound rather similar, while I still prefer the W2002 for its euphonic signature.
- The midbass-centric nature of the HD650 is nicely reflected by the graph. This is not so much the case with the HD600 which is pretty neutral, but the lack of deep bass is surely obvious
- Most AT woodies (the L3000 also technically being a woodie) have a very large dip at 6 KHz. This is audible in a sine sweep, but I'm not sure how much the proximity of the microphones to the driver emphasizes this effect.
- The ATH-EM9r sounds metallic, harsh and bass-shy
- The MS-Ultimate effectively tames the harshness of the stock MS-Pro
Things I didn't see confirmed:
- The L3000 sounds more bassy than the graph suggests
- The K271S doesn't sound as AT-like as the graph suggests
- The MS-Ultimate and K701 don't sound as much alike as the two graphs suggest. This can probably explained by the differences in soundstage.
Lessons Learned
It would seem that comparing graphs of different headphones isn't actually as easy as I thought. Psychoacoustics also play a big role and one shouldn't forget about the inaccuracies of the method of measurement.
Although if one knows what to look for, comparing different parts of a single graph and putting it in relation to other parts of the spectrum can tell you quite a lot about a particular headphone. The fact that I can find a lot of my subjective observations confirmed and seeing how similar those graphs are compared to graphs others have done, I'm led to assume that these measurements can't be too far away from truth.
I also feel confirmed in not paying too much attention to the amp story. A simple roll of cotton stuffed under the pads of my AT headphones has several times the effect a different amp has.
Also, don't be fooled by the scale of the graphs. My graphs feature a dynamic range of about 30 db as compared to graphs elsewhere. Its simply easier to see the relevant dips and bumps that way.
I've also learned to appreciate that crooked graphs mustn't sound crooked. And headphones measuring flat don't necessarily sound better. To my preference anyway.
So, I hope you enjoy comparing the graphs and I look forward to discussing with you about the specifics of the method and the actual results.
Benjamin
after opening a new thread on the German hifi-forum, I've decided to translate my findings into English and share my findings with you
Prologue
First things first: a while ago I had this idea of doing what had been on my mind for a while, which was making discussions about headphones a little more objective. There are several scattered sources of information all across the web with a few websites providing frequency response graphs for all sorts of headphones. These measurements were often conducted using totally different methods of measurement which logically leads to different looking graphs for the same headphone. Production variations on a wide range of headphones don't add to the credibility and comparability of measurements. I think it's understandable that not many people put much faith in interpretations of frequency response graphs.
So I wondered how accurate those graphs were and how exactly they would compare to my own measurements. I did some research on the web for affordable and simple gear to help me do my own measurements. I stumbled upon a few stupidly expensive professional solutions until I came up with the Soundman OKM II Pop/AV Electret Earbud-Microphones. At ~$150 these are relatively affordable and promise to be usable as microphones for binaural recordings as well. And these things deliver, I can tell you that in advance. I've created some of my own binaural recordings which sound pretty convincing. But let's not get distracted, this is about frequency fesponse measurements
First of all, these things are not In-Ear microphones, so they'd disregard any sort of effect the inner ear has on the actual sound that arrives at the eardrum.
I'm not entirely sure whether this was taken into account by Soundman when developing these microphones, I just know that the recordings done by these are pretty good and allow the brain to reproduce spacial information quite accurately. These microphones however do not have a totally flat Frequency Response as indicated by the following pictures:
The measurements shown here were conducted by the Technical University of Berlin and should at least have some representative value. Sadly, I do not know which of those graphs ultimately belongs my particular model, since they're declared as "OKM II/K" and "OKM II/R", none of which seems familiar to me.
The actual measurements
In order to take those measurements, I put the mics into my ears and albeit feeling a little uncomfortable with the spiky construction underneath the cushion, I was able to get a half-way decent fit. I then hooked the mics to the microphone input of my soundcard and fired up two instances of Cool Edit. One of those instances was used as playback, the other was used as the recorder. Playback was done via my soundcard's optical out, my HA25D as DAC feeding the Corda Aria as the testing amp.
Firstly, I generated a few seconds of white noise since white noise seemed the most appropriate signal to do measurements with. It is a noise that incorporates all the frequencies from 20 Hz to 20 Khz across the spectrum in equal amounts.
I then played back that noise in an endless loop using the mics in my ears and the tested headphones on top of the mics. Not one of the most sensual experiences in my life, I can assure you, but sometimes uneasy things just have to be done for the greater good
For each headphone I first did several runs to adjust the volume to a uniform sound pressure level. I started off by recording a total of 10 seconds of white noise so that the average RMS (Root Mean Square) input came out at roughly -15 db. If the input level wasn't quite there, I repeated the process until I had a signal at roughly -15 db. This means that in contrast to the method Headroom uses for their FR measurements, the input level here won't be matched at 1 KHz only, but at an average sound pressure level of -15 db over the entire spectrum. I figured that this would make the graphs more comparable than the 1 KHz method. After all, headphones with a dip at 1 KHz would have a relatively higher FR curve in bass and treble (U shape) that mostly exceeds the curve of headphones with a bump in that area.
I then created a 30 seconds recording of white noise with the leveled volume. Afterwards, I normalized the recorded waveform to exactly -15 db RMS digitally. I saved the waveforms created like this and used Rightmark Audio Analyzer (RMAA) to generate the frequency response graphs.
Keep in mind that minor changes in fit and position change the graphs quite significantly. Especially the treble area beyond 4 KHz seems very sensitive to such changes.
The following headphones were tested using this method:
- ATH-L3000
- ATH-A2000X
- ATH-W5000
- ATH-W2002
- ATH-W11JPN
- ATH-W1000
- ATH-AD2000
(all AT up to here pad modded) - Precide Ergo Model 2 (closed)
- ATH-EM9r
- JVC HA-DX1000
- Sennheiser HD540 Reference Gold
- AKG K701
- AKG K271S with DT770 Pads
- Sennheiser HD650
- Sennheiser HD600
- Alessandro MS-Ultimate
- Alessandro MS-Pro
- Grado SR-60
- Ultrasone HFI-680
- Ultrasone PRO 900
New: I've also done a few more measurements on the different variations of the MS-Pro and MS-Ultimate. Check the gallery under "Special: MS-Pro". There's also a special gallery on the Ultrasone PRO-900 burn-in-process and the effects of its placement on the head.
Long story short, the results are here:
Headphone Frequency Response Measurements
I've divided the gallery into three categories.
1.) "Headphones" has measurements of headphones the way I regularly use them (pad mod on all AT, keep that in mind!) and all the other headphones I had with me a while for taking measurements.
2.) "Amps" has measurements comparing different amps when using one particular headphone model
3.) "Mods" shows the differences a mod does to a particular headphone.
Things I see confirmed by those measurements:
- There's certain similarities among AT woodies, though there's still plenty of differences between the woodies and the open models to claim that there's no such thing as a general AT house sound
- BUT: most AT headphones being mids-centric headphones totally goes to show with these graphs (which is what I love them for
)- Solid state headphone amps only have absolutely minor influence on the sound signature
- The Pad Mod I did to my AT headphones definitely has the effect I felt it had
- Dry and boring headphones like the K701 and HD540 measure pretty flat with a slight emphasis on treble
- The HD600 and AD2000 sound and measure pretty similarly (but of course I still prefer the AD2000 :P)
- The AD2000 is pretty neutral and totally different from any other AT in the field
- The HFI-680 has a bathtub-signature
- ... and the PRO900 even more so
- The JVC HA-DX1000 has pretty honky mids due to a massive bump at 350-700 Hz
- The W2002 has more extended bass than the L3000
- The W5000 and W2002 being closely related in the AT ancestory goes to show in that they sound rather similar, while I still prefer the W2002 for its euphonic signature.
- The midbass-centric nature of the HD650 is nicely reflected by the graph. This is not so much the case with the HD600 which is pretty neutral, but the lack of deep bass is surely obvious
- Most AT woodies (the L3000 also technically being a woodie) have a very large dip at 6 KHz. This is audible in a sine sweep, but I'm not sure how much the proximity of the microphones to the driver emphasizes this effect.
- The ATH-EM9r sounds metallic, harsh and bass-shy
- The MS-Ultimate effectively tames the harshness of the stock MS-Pro
Things I didn't see confirmed:
- The L3000 sounds more bassy than the graph suggests
- The K271S doesn't sound as AT-like as the graph suggests
- The MS-Ultimate and K701 don't sound as much alike as the two graphs suggest. This can probably explained by the differences in soundstage.
Lessons Learned
It would seem that comparing graphs of different headphones isn't actually as easy as I thought. Psychoacoustics also play a big role and one shouldn't forget about the inaccuracies of the method of measurement.
Although if one knows what to look for, comparing different parts of a single graph and putting it in relation to other parts of the spectrum can tell you quite a lot about a particular headphone. The fact that I can find a lot of my subjective observations confirmed and seeing how similar those graphs are compared to graphs others have done, I'm led to assume that these measurements can't be too far away from truth.
I also feel confirmed in not paying too much attention to the amp story. A simple roll of cotton stuffed under the pads of my AT headphones has several times the effect a different amp has.
Also, don't be fooled by the scale of the graphs. My graphs feature a dynamic range of about 30 db as compared to graphs elsewhere. Its simply easier to see the relevant dips and bumps that way.
I've also learned to appreciate that crooked graphs mustn't sound crooked. And headphones measuring flat don't necessarily sound better. To my preference anyway.
So, I hope you enjoy comparing the graphs and I look forward to discussing with you about the specifics of the method and the actual results.
Benjamin
Anyway, Big thanks for sharing. 
