accurate bass doesn't really exist in nature, if it makes gals movin it's accurate enough.

 

Last edited: Mar 30, 2025 at 9:40 PM

There is a good reason audio SF has the best bass of any nightclub

 

You can't properly compare the accuracy of a measurement rig by looking at measurement made with different headphones/IEMs. Even if it's the same model, you will have unit-to-unit variation.

But doesn't it introduce a huge amount of uncertainty ? Even with Taguchi methods, you will have to run a lot of trials to get a consistent, repeatable result.

That's true, although why those two ? The Moondrop Variations are not acoustically optimal, its high-end is underdamped, with strong ringing, not fixable with EQ. And the LCD-5, as it is over-ear, is still subject to internal reflections. These are not minimum phase, so hard to deal with EQ. Sine sweeps for IEMs will help detect the ear canal impedance peak(s), which is minimum phase, and easy to fix with EQ. The 5128 has the best high-end performance for headphones, so corrections using such a rig will be better, in this frequency region. It has issues in the lower end of the spectrum though.

This is the FFT magnitude of pink noise (ignore the slight boost above 10k, uncalibrated analyzer), measured over 10 seconds :



And the one of a sine sweep :



The low-end is too chaotic on pink noise. Mathematically speaking, you have to wait an infinite amount of time to reach the magnitude precision of a sine sweep. So if you're specifically measuring under 500Hz, it can be a problem.

Because they are using FIR filters with phase response correction, and active room treatment. So they can EQ them to the same magnitude response.

 

I'll feed you 140dB SPL of critically damped subwoofers in a plain field with no distortion, and you'll tell me if it's not accurate

 

The issue is consistent across all IEMs, mainly being that the upper treble is more dipped in Crinacle 711 IEM measurements.

Well, that's why the ranking method for earphones requires 40 trials. However, the A/B slider only requires 1-3 trials for consistent, repeatable results in most cases for the reasons mentioned in my original post.

When measured on the BK5128 or Gras RA0042, the Moondrop Variations has perfectly smooth treble. When listening in tone generator at an optimal insertion depth, there are no peaks or dips and the treble is well extended.

The LCD-5 is carefully designed to minimize reflections, giving it low reseat variance and an extremely smooth response.


I agree that the 5128 has usefulness for IEM high frequency measurements, but RA004x + KB500x over-ear measurements have great high frequency accuracy. 5128 over-ear measurements consistently show an 8 kHz peak not found in humans.

Here is a loopback measurement with Blackhole 2ch which has some clock drift. Note the scale on the left: none of the errors exceed 0.1 dB.

Here is a measurement of an IEM using our periodic pink noise and RTA method vs a sweep in REW

 

Why does DIRAC EQ full range? When I tested DIRAC live full range with Infinity Primus 360, the higher frequencies were definitely worse sounding. Wouldn't the direct sound inherently be in phase if the loudspeakers are equidistant from the listening position?

Do you have any evidence that the summed response of channels calibrated with DIRAC is equal to the individual response (aside from loudness)? You could have both speakers perfectly in phase at the listening position, but a room mode at 70 Hz causes a deep null when both play together — not because of timing, but because of position-related wave interference.

 

Last edited: Mar 31, 2025 at 11:43 AM

From Dirac https://www.dirac.com/wp-content/up...rRlmwQ4ElJ8S9-DjmOxwPbD5w-Tx13xjT-kN4iX13J8mM

Only "Full Bass Optimization" with Bass Control of Dirac's options seems to factor in the summed response at all.

 

According to this it still only works on a per channel basis meaning there is no guarantee that when all channels are playing together, the response will match the target curve above the crossover frequency

 

Last edited: Mar 31, 2025 at 2:23 PM

There is no way to tell between these measurements if the deviations we see are from unit-to-unit differences of the IEMs, or differences between the measurement rigs. You are fighting against two variables here, so for the best results, you need to isolate one of them to eliminate guess work.

I ran your A/B slider 4 times with 9 trials each, using the default track, and these were the different corrections it gave. That's just me, but I'll be surprised if you didn't need way more trials to get consistent, repeatable results with most people.

For the Moondrop Variations, I wasn't talking about magnitude response, but about their lack of dampening in the high-end :



I also measured some cheap Sennheiser IEMs for comparison. Granted, this issue isn't too horrible, as it's rare to hear it when listening to music. But in some specific moments, it's audible. So that's why I wouldn't call them "acoustically optimal in higher frequencies".

The graph you shared shows some terrible uneven response above 8kHz.

Yeah but with speakers in a room, it's a different story. Mathematically, noise is never perfectly flat, unless you measure it over an infinite period of time. Even over a quite long 10.6s (512k FFT at 48kHz) measurement, you can still see the roughness from the noise under 100Hz in this :



Granted, depending on the smoothing that is required, that is a non-issue. But watch as the magnitude rises below 40kHz. That's because of a dynamic range issue. The RTA confuses the noise of the room with the pink noise of the generator. That to me is the more important reason to use sine sweeps, apart from IR windowing.

 

Well, I never said that Dirac does it correctly. I just said that they do it, no matter how well it's done

 

This doesn't really make any sense. Variations is minimum phase.

The measurement rig has its limits, but if you know what to look for on GRAS 43AG/45CA with RA004x coupler, this is very smooth. 5128 measurements of it are very smooth too, and out of all the headphones I have listened to in-person, LCD-5 has the smoothest and most well extended treble.

If you try the speaker measurement, you will notice that the loudness of the periodic pink noise is high for this very reason. Additionally, 1/16 smoothing is performed before EQ is applied.

 

Variations is unfortunately notorious for having very high unit variation, no pun intended

 

Well it can be minimum phase and still have some ringing. It's the difference between the forced and natural response of a system. But it's interesting that your spectrogram doesn't show any, I ruled out manufacturing defect as this happens on both my left and right unit. And I clearly hear it too, so it's not a measurement error. How did you make your measurement ?

You shared an average curve, it's always going to be smoother. There really isn't anything exceptionally smooth with the LCD-5's treble, the HD650 is just as smooth, and its design is more than 20 years old.

 

Last edited: Apr 1, 2025 at 3:58 PM

I measured with a sine sweep in REW using a clone 711 coupler

HD650 treble is pretty smooth, but the response varies a lot with pad wear, so you would need to continuously change your EQ to meet a target curve. I did hear a peak at around 9 kHz with hd650 which doesn't exist with LCD-5. LCD-5 is basically the headphone that is fully smooth in tone generator.

 

Hey boss I am really into this and have followed the process all the way to getting an outputted txt file to use.
I am on a mac computer using Ableton as my software. I would like to know what I can do on here with this file given that you haven't provided any pathways for mac users on your site.

I am also wondering if it is better to go with the avg performance rating or my own one that I did after 9 A/b tests.

Cheers!

 

Hey man, thanks so much for trying the software out! On mac I recommend using EQMac Pro for simplicity.

As for average preference rating vs your own, it's up to you and which you like more, but I personally use the average targets.

 

Although, I would more strongly recommend a dac/amp like Qudelix T71 or MiniDSP 2x4 Flex for lower latenxy