[earfonia]

Tonality or frequency response is probably the first aspect that come to our mind when we evaluate audio gears such as earphones, headphones, and speakers. To accurately measure frequency response we need some skills, knowledge, and proper measurement tools. In most cases not a practical approach to many. My objective with this 'by ears' FR evaluation is basically to develop a practical approach to 'estimate' the perceived frequency response of audio transducers (IEMs, headphones, speakers) by our hearing. Something that can be done by most audio enthusiasts.

For easy referencing, I coined a term for it: EFRE (Earfonia Frequency Response Evaluation)
EFRE is basically a method to estimate the perceived frequency response or a certain audio transducer by ears. So EFRE is a subjective evaluation as people might have a different hearing response.
EFRE is a set of audio tracks, to compare by hearing the loudness of different tones to a 1kHz tone. So anyone can try it using their own audio system, to roughly estimate the perceived frequency response of their IEMs, headphones, and speakers, according to their hearing response.

This is the download link for the EFRE tracks:

EFRE Tracks

[COLOR=000000]Instructions:[/COLOR]

1. Track 1 is to set volume level of the audio system. Track 1 is composed of the following:
2 secs of 0 dB 1 kHz (0.25 dBFS is set as reference for 0 dB loudness)
2 secs of +12 dB 1 kHz
2 secs of 0 dB 1 kHz
2 secs of -12 dB 1 kHz
The set above repeated 3 times. Set the volume until we hear the +12 dB part sounds a little too loud. On my SPL meter, the 0 dB is usually set at around 85 dB(A).

2. Tracks 2 to 9 are reference tracks of 1 kHz tone, for our ears to hear the difference of loudness between the 0 dB and +3 dB, -3dB, and so on until +/- 12 dB.
I usually keep going back to tracks 2-9 during the evaluation, to estimate the loudness differences.

3. Tracks 10 to 45 are the EFRE tracks, from 20 Hz to 20 kHz, all set at the same 0 dB loudness with the 1 kHz. Each track will have 2 secs of 1 kHz tone and 2 secs of other tone, played in 3 repetition.

4. Excel spreadsheet to build the graph is included in the EFRE folder. Use the spreadsheet to take note of the FR evaluation.



Having tried the EFRE for a few headphones, I observe the following:
1. It is difficult to compare loudness of different tones. Some practice definitely required.
2. People might have different hearing response, therefore various results from the same gear are expected.


This method is under development, so to be improved by inputs and suggestions from users. Comments are welcome!

 

[earfonia]

Reserved for EFRE results.

Notes:
My hearing is limited to 15 kHz, therefore I cannot observe higher frequency tones.

Otherwise specified, I use Questyle CMA600i (DAC + Amp) for below measurement.

For Hifiman HE6, I used Matrix HPA-3B headphone amplifier to drive it.


[COLOR=FF4400]
[/COLOR]

 

[yuriv]

This is similar: http://newt.phys.unsw.edu.au/jw/hearing.html (needs Flash player)
 
They talk about it in this thread: http://www.head-fi.org/t/661886/equal-loudness-curve-testing
 
You can get more resolution than the Flash tool on that site by using something like Sinegen, an SPL meter, and a parametric equalizer. Also, there are a few how-to-EQ guides on this site that play on a variation on that theme.

About your test tones--why is the octave between 10 kHz and 20 kHz being treated so exquisitely (with tones at 10k,  11k, 12k, ..., 19k, 20k), but there are no tones in the octave between 1 kHz - 2 kHz or 100 Hz - 200 Hz? Yes, the frequency response gets more jagged the higher you go up in frequency, but one can argue that the response between 1k and 2k, where the ear is more sensitive, is more important. To a lesser extent, the same goes for 100 Hz - 200 Hz.

 

[earfonia]

  This is similar: http://newt.phys.unsw.edu.au/jw/hearing.html (needs Flash player)
 
They talk about it in this thread: http://www.head-fi.org/t/661886/equal-loudness-curve-testing
 
You can get more resolution than the Flash tool on that site by using something like Sinegen, an SPL meter, and a parametric equalizer. Also, there are a few how-to-EQ guides on this site that play on a variation on that theme.

About your test tones--why is the octave between 10 kHz and 20 kHz being treated so exquisitely (with tones at 10k,  11k, 12k, ..., 19k, 20k), but there are no tones in the octave between 1 kHz - 2 kHz or 100 Hz - 200 Hz? Yes, the frequency response gets more jagged the higher you go up in frequency, but one can argue that the response between 1k and 2k, where the ear is more sensitive, is more important. To a lesser extent, the same goes for 100 Hz - 200 Hz.

 
Thanks for the info! Interesting topic indeed!
Similar method but with a different goal. The goal of that thread is to plot Equal Loudness Contour of our hearing, while my goal for EFRE is to plot estimated FR response of a particular headphone or IEM.
 
The tones are to plot the frequency response logarithmic scale on the FR graph as shown above. It is a good idea to add more granularity in the area that our hearing is sensitive as you mentioned above. I will consider that. As for the tones above 10 kHz, my intention is to actually check the upper limit of our hearing.

 

[VNandor]

3. Tracks 10 to 45 are the EFRE tracks, from 20 Hz to 20 kHz, all set at the same 0 dB loudness with the 1 kHz. Each track will have 2 secs of 1 kHz tone and 2 secs of other tone, played in 3 repetition.

4. Excel spreadsheet to build the graph is included in the EFRE folder. Use the spreadsheet to take note of the FR evaluation.



Having tried the EFRE for a few headphones, I observe the following:
1. It is difficult to compare loudness of different tones. Some practice definitely required.
2. People might have different hearing response, therefore various results from the same gear are expected.


This method is under development, so to be improved by inputs and suggestions from users. Comments are welcome!

If I understand it correctly when you listen to the different pitches you EQ it until both tones sound equally loud? Because if you do that, that's a bad thing as everyone's hearing starts to roll off below ~200Hz and above ~5kHz. So even if you listened to a headphone with a perfectly flat frequency response, your curve would not look flat and would show a roll of below 200Hz and above 5kHz.
 
Even if you simply tried to estimate the loudness differences between different pitches (without the use of EQ) you would run into the same problem.

 

[earfonia]

  If I understand it correctly when you listen to the different pitches you EQ it until both tones sound equally loud? Because if you do that, that's a bad thing as everyone's hearing starts to roll off below ~200Hz and above ~5kHz. So even if you listened to a headphone with a perfectly flat frequency response, your curve would not look flat and would show a roll of below 200Hz and above 5kHz.
 
Even if you simply tried to estimate the loudness differences between different pitches (without the use of EQ) you would run into the same problem.

 
Yes you're right about the roll off around the bass and treble area! But the objective of EFRE is to observe how a headphone frequency response is heard by our ears on normal music listening level. The objective is different than plotting equal loudness contour. So besides the estimated FR plot of a certain headphone using our own hearing, EFRE objective is to see how bad is the roll off around the bass and treble area, so we can see how much bass and treble we can hear from a particular headphone or IEM, on average listening level. We can hear more bass and upper treble when we increase the loudness, but that's not the point here, because we don't listen to music at that volume level. So there is no EQ or loudness adjustment during the evaluation. Volume is set once, before the evaluation.
 
I do find that EFRE helps me to a certain extend to spot annoying peaks and dips on headphone FR. For example, I could easily spot the 7 kHz peak of Sennheiser HD800 using EFRE. While I don't hear that 7 kHz peak on few other headphones that I've tested.

 

[VNandor]

If you want to plot a headphone's estimated frequency response by boosting/reducing frequencies until they sound equally loud (essentially using your ears), you will have to compensate for your equal loudness curve.
 
It is because when a headphone's frequency response is a flat line (which is ideal more or less) you won't perceive it as flat. If you had a headphone that had a flat frequency repsonse and you played back different sine waves with the same amplitudes but different pitches, you would notice roll of at both ends and a slight peak at ~1kHz. But this is how it should be, you would perceive a natural spectral balance. If you had a system where you played back sine waves at the same amplitudes but different pitches and you perceived all the pitches equally loud, that system would be bad. It would have too much bass and treble.
 
If you want to plot how we hear a headphone's frequency response it's fine but but keep in mind that an ideal headphone's (with flat frequency response) plot should look like an inverse of the equal loudness curve and NOT a flat line. So if you did this plot and you saw a roll off in the bass it wouldn't necessarily mean it's bass light; in fact if the roll of wasn't as steep as suggested by the invert of the ELC it would be bass-heavy.
 
So if you want to plot a headphone's frequency response by using your ears, there is no way you can avoid either applying a generic equal loudness (and hoping that yours match up to it) curve or somehow measuring you own ELC.
And if you want to plot how you hear a headphone's frequency response well you don't need your ELC for it but that graph would be pretty much useless without something to compare to. And again that target is not supposed to be a straight line, since that's not how we hear. The target is supposed to be the invert of the ELC.
 
It has nothing to do with the fact our ELC changes at different SPLs... That only would be relevant if you measured your ELC at let's say 115dB SPL @ 1kHz (ouch) but when you made the plot for the headphone's FR you listened at let's say 85dB SPL @ 1kHz. In that case you would add a slightly different compensation curve compared to how your hearing actually skewed the perceived FR.
 
So no matter what you do, as long as you use your hearing to plot frequency responses, you'll need to take the ELC into account. That's why you see it frequently being mentioned on the linked thread by the way.

   
I do find that EFRE helps me to a certain extend to spot annoying peaks and dips on headphone FR. For example, I could easily spot the 7 kHz peak of Sennheiser HD800 using EFRE. While I don't hear that 7 kHz peak on few other headphones that I've tested.

Your method is actually quite good at spotting narrow peak and dips but it won't let you know the overall spectral balance of the headphones. Even just simply listening to some music would be better for that. (As long as you choose not to compensate for your ELC.)

 

[earfonia]

  If you want to plot how we hear a headphone's frequency response it's fine but but keep in mind that an ideal headphone's (with flat frequency response) plot should look like an inverse of the equal loudness curve and NOT a flat line. So if you did this plot and you saw a roll off in the bass it wouldn't necessarily mean it's bass light; in fact if the roll of wasn't as steep as suggested by the invert of the ELC it would be bass-heavy.
 
So if you want to plot a headphone's frequency response by using your ears, there is no way you can avoid either applying a generic equal loudness (and hoping that yours match up to it) curve or somehow measuring you own ELC.
And if you want to plot how you hear a headphone's frequency response well you don't need your ELC for it but that graph would be pretty much useless without something to compare to. And again that target is not supposed to be a straight line, since that's not how we hear. The target is supposed to be the invert of the ELC.
 
It has nothing to do with the fact our ELC changes at different SPLs... That only would be relevant if you measured your ELC at let's say 115dB SPL @ 1kHz (ouch) but when you made the plot for the headphone's FR you listened at let's say 85dB SPL @ 1kHz. In that case you would add a slightly different compensation curve compared to how your hearing actually skewed the perceived FR.
 
So no matter what you do, as long as you use your hearing to plot frequency responses, you'll need to take the ELC into account. That's why you see it frequently being mentioned on the linked thread by the way.
Your method is actually quite good at spotting narrow peak and dips but it won't let you know the overall spectral balance of the headphones. Even just simply listening to some music would be better for that. (As long as you choose not to compensate for your ELC.)

 
I agree with you if the goal is a straight line of frequency response, or something close to that. If something close to a straight line is the objective, ELC (either generic or my own) is required to compensate what I (or other person) actually hear. But my objective with EFRE is not a straight line FR curve. My main objectives are to estimate subjective FR response by our hearing, and to spot annoying peaks and dips on the FR, using a method that is as simple as possible. By simply listening to music, for example HD800, we can hear that it sounds bright, definitely there is some emphasize around the treble, but at what kHz ? Trained ears can guess pretty accurately, but not many people could guess it accurately. By using EFRE tracks many could guess pretty close to the real treble peak. Therefore we can communicate more objectively, like by saying "I hear a treble peak at around 7 kHz on that particular headphone".
 
Another reason why I try to avoid loudness adjustment and to use ELC compensation on EFRE, is because the process will be more complicated and would require an app to simplify the process. I would like to make EFRE a simple process and can be loaded into DAPs or smartphones and we can easily play it when testing a headphone or earphone in an audio store, to roughly estimate if there is any abnormal peaks or dips. Practically a simple set of tracks that could be easily played back on any audio system.
 
I will consider adding more tones, without making it too many tracks. Probably add tone at 150 Hz and 1.5 kHz would be a good idea as suggested by @yuriv.
 
Btw, thanks a lot for your suggestion and arguments. EFRE is still under development anyway, and I'm taking any inputs and suggestions to improve it.

 

[VNandor]

   
I agree with you if the goal is a straight line of frequency response, or something close to that. If something close to a straight line is the objective, ELC (either generic or my own) is required to compensate what I (or other person) actually hear. But my objective with EFRE is not a straight line FR curve. My main objectives are to estimate subjective FR response by our hearing, and to spot annoying peaks and dips on the FR, using a method that is as simple as possible. By simply listening to music, for example HD800, we can hear that it sounds bright, definitely there is some emphasize around the treble, but at what kHz ? Trained ears can guess pretty accurately, but not many people could guess it accurately. By using EFRE tracks many could guess pretty close to the real treble peak. Therefore we can communicate more objectively, like by saying "I hear a treble peak at around 7 kHz on that particular headphone".

Okay then. Your method is perfectly fine for what you've described above. A slightly more accurate approach would be to listen to a slow logarithmic sine sweep while looking at a real time spectrum analyzer (which is not hard to get by the way) but that would make the process more complicated with little additional benefits.
 
What confused me is your original post where you said:

My objective with this 'by ears' FR evaluation is basically to develop a practical approach to 'estimate' frequency response of audio transducers (IEMs, headphones, speakers) by our hearing. Something that can be done by most audio enthusiasts.

which is fundamentally different from what you've said later and I thought you may not understand the difference.

 

[earfonia]

  What confused me is your original post where you said:
which is fundamentally different from what you've said later and I thought you may not understand the difference.

 
Oh yes, you're right! My mistake, my bad. I should have said 'to estimate the perceived frequency response of audio transducers by our hearing'. My previous statement leads to the 'close to linear' FR that will required ELC in the process. Thanks for pointing out! I will correct it.