Post-correction measurements

[Jbrunwa]

Inquiring mind wants to know why Dirac Live and Audyssey do not provide a built-in way to measure post-correction results?

 

[ppataki]

I believe this explains it:

The red curve is with Dirac, post-correction
The green one is again with Dirac post-correction but further corrected with a target curve in Dirac that will make the response actually flat in the listening position
The difference is huge, not just on the below graph but by listening to it too...

I love Dirac (could not survive without it) but I guess you need some manual tweaking if you really want a flat curve....
I have opened a ticket with their support team if they plan to include this in any future releases so no manual tweaking would be necessary - after all it is the flat curve that we shall be aiming for

 

[alex-z]

after all it is the flat curve that we shall be aiming for

Your in-room response should not be flat. It is well established that our ears prefer slight bass elevation and treble droop. Commonly referred to as the Harman or JBL Synthesis target curve. Speakers which measure flat anechoic will naturally have that downward tilt when placed in a room.

 

[ppataki]

By 'flat' I meant not that zigzaggy as the red curve but rather 'flat' like the green
Then it is up to one's taste to modify the tilt accordingly

 

[Frgirard]

Your in-room response should not be flat. It is well established that our ears prefer slight bass elevation and treble droop. Commonly referred to as the Harman or JBL Synthesis target curve. Speakers which measure flat anechoic will naturally have that downward tilt when placed in a room.

+1

A well designed speaker is flat in anechoic condition, not at the listening position.

 

[Absolute]

By 'flat' I meant not that zigzaggy as the red curve but rather 'flat' like the green
Then it is up to one's taste to modify the tilt accordingly

That green one is undoubtedly the result of massive over-correction, likely a single-point correction or very small measurement grid.
Unless it's a reflection-free room we don't want that kind of response in the listening position because we can only achieve a response like that by ignoring the direct sound in favor of steady-state.

Psychoacoustically that's a bad idea and in a reflective space the curve is only valid for a single point in space. Re-measure 50 cm in all directions and compare.

The red curve upwards tilt looks like something I would expect if the measurements in Dirac is done without a microphone calibration file loaded.

 

[Jbrunwa]

That green one is undoubtedly the result of massive over-correction, likely a single-point correction or very small measurement grid.
Unless it's a reflection-free room we don't want that kind of response in the listening position because we can only achieve a response like that by ignoring the direct sound in favor of steady-state.

Psychoacoustically that's a bad idea and in a reflective space the curve is only valid for a single point in space. Re-measure 50 cm in all directions and compare.

The red curve upwards tilt looks like something I would expect if the measurements in Dirac is done without a microphone calibration file loaded.

Regardless of the target curve, am I the only one who wishes that these products would allow users to actually measure the post correction result? I mean, you have to use other tools like REW, and there is no good way to use the same microphone positions. So you are stuck using REW and then you have to go back into Dirac to make adjustments. Very unwieldy. Surely these companies have a way to measure post-correction results as part of their internal testing process. Or are they worried that showing the actual results would create too many questions about their products?

 

[Absolute]

Well, I think Dirac (and other companies) knows that most users can't use any after correction measurement for anything meaningful. Those algorithms are created based on psychoacoustics and not for the purpose of generating pretty graphs after correction, but many people believe that nice graphs equals nice sound.

Since that's not true, it makes sense to skip a feature that'll certainly confuse users who are under that illusion.

Used correctly Dirac will/should correct the modal area in a way that makes psychoacoustically sense and limit itself to correcting the speaker only above around 1 kHz. How many users out there can we assume would even know the difference between a single-point measurement and spatially averaged multi-point measurement? And how many can we assume would go through the pain of measuring the anechoic response of the speaker before and after Dirac to make an educated evaluation of how well Dirac does?

I think Dirac made the logical choice by making a software that's easy to use and gives you the results you want without providing confusing information most people can't use.

 

[Jbrunwa]

Well, I think Dirac (and other companies) knows that most users can't use any after correction measurement for anything meaningful. Those algorithms are created based on psychoacoustics and not for the purpose of generating pretty graphs after correction, but many people believe that nice graphs equals nice sound.

Since that's not true, it makes sense to skip a feature that'll certainly confuse users who are under that illusion.

Used correctly Dirac will/should correct the modal area in a way that makes psychoacoustically sense and limit itself to correcting the speaker only above around 1 kHz. How many users out there can we assume would even know the difference between a single-point measurement and spatially averaged multi-point measurement? And how many can we assume would go through the pain of measuring the anechoic response of the speaker before and after Dirac to make an educated evaluation of how well Dirac does?

I think Dirac made the logical choice by making a software that's easy to use and gives you the results you want without providing confusing information most people can't use.

I'm not sure I understand, but I'm still new to this. Using the same logic, why bother showing the initial measurements to begin with, and give customers the ability to make changes, if it is all magic?

 

[Kal Rubinson]

Regardless of the target curve, am I the only one who wishes that these products would allow users to actually measure the post correction result? I mean, you have to use other tools like REW, and there is no good way to use the same microphone positions.

So how would Dirac (or others) do any better? Their measurements and corrections are based on multiple mic positions and, unless you have 9+ microphones in fixed positions, there's no way to replicate the original.

 

[abdo123]

I'm not sure I understand, but I'm still new to this. Using the same logic, why bother showing the initial measurements to begin with, and give customers the ability to make changes, if it is all magic?

to pick their desired target curve?

 

[Jbrunwa]

So how would Dirac (or others) do any better? Their measurements and corrections are based on multiple mic positions and, unless you have 9+ microphones in fixed positions, there's no way to replicate the original.

OK, since uber knowledgeable people are saying there would be no benefit to performing post-correction measurements, there must something really basic that I don't understand, and I apologize in advance if this is the case.

Let's say using Dirac Live (or Audyssey app, for that matter), the initial measured response in a room shows a peak or a dip in a bass or midrange region when you run Dirac Live. When it defines the filters, it shows you the target curve, and for the purpose of argument, you like the Harman curve or whatever curve Dirac uses as a default. You then send the generated Dirac filter to the AVR. How do you know that the peak or dip has been adequately corrected in the room unless you do some type of post-correction measurement?

 

[Jbrunwa]

to pick their desired target curve?

You could pick a target curve without knowing anything about the initial measurements.

 

[abdo123]

You could pick a target curve without knowing anything about the initial measurements.

Not in the seven hells i would.

I would never let an automatic room correction just ‘do whatever’.

Low frequency and high frequency roll offs need to be respected, maximum boost amount needs to be well thought out.

Otherwise it will do more harm than good.

 

[GDK]

You can save the REW sweeps (with the acoustic timing reference) on your system. I have them included in my Roon library, although I assume the same approach would work the same with other playback systems. Play the file back with Dirac enabled and measure with REW.

 

[dualazmak]

Hello friends,

Just for your reference,,,

Although my audio system and approaches are considerably different from your AV systems, I also always would like to measure and confirm/understand the frequency responses in reliable and reproducible manner at each level/step of the audio system;

1. in digital level after the EQ-DSP processing,
2. in (multichannel) DAC's analog output level,
3. in amplifiers' (I mean each of the multiple amplifiers) SP output level,
4. in amplifiers' SP output level after protection capacitors if used,
and of course,
5. in actual room sound at listening position by using measurement microphone

Since Fq response measurement is based on the statistical FFT analysis, I believe that we should be careful enough on the sample size (total energy of the sound signals for analysis), method by rapid sine-wave sweep, by pink noise or by white noise, selection of FFT size, and other smoothing algorithms (like REW's psychoacoustic smoothing) if applied.

For the common and reliable method in all the way through 1. to 5., I use rather primitive "cumulative (recorded) white noise averaging" for which I already could have confirmed and validated the usefulness. If you would be interested, please visit my recent posts here, here and here.

 

[Weeb Labs]

I have found ARC's predicted responses to be surprisingly accurate, despite the absence of a post-correction measurement function.

Here are the uncorrected (green) and predicted (orange) responses of my desktop setup, consisting of a pair of T5Vs.

And here are the uncorrected (green) and post-correction (orange) measurements, taken via REW

ARC's predicted response is rather close to the actual post-correction response.

 

[Kal Rubinson]

How do you know that the peak or dip has been adequately corrected in the room unless you do some type of post-correction measurement?

REW. It will not be exact but if you compare the REW pre and REW post, you should be able to appreciate some of what has been accomplished.

 

[Jbrunwa]

REW. It will not be exact but if you compare the REW pre and REW post, you should be able to appreciate some of what has been accomplished.

Thanks. So assuming top of line system components but not using separate systems for stereo and multichannel, about the best one can do then is then is use REW to try to optimize speaker locations, room treatments, and listening positions for smoothest response below Schroder frequency and optimal listening window for high frequencies. Run Dirac and cycle through all listening positions, finishing at prime listening location, save several filter options, then without changing mike position, REW with and without Dirac, match speaker levels, for each of the chosen filter configurations. Do this for stereo, stereo with subs, and multichannel presets. Or am I over thinking it?

 

[Jbrunwa]

I have found ARC's predicted responses to be surprisingly accurate, despite the absence of a post-correction measurement function.

Here are the uncorrected (green) and predicted (orange) responses of my desktop setup, consisting of a pair of T5Vs.

View attachment 137405

And here are the uncorrected (green) and post-correction (orange) measurements, taken via REW

View attachment 137406

ARC's predicted response is rather close to the actual post-correction response.

I see an overall similarity. I guess that ARC over smooths the graphs, or maybe their averaging of listening positions smooths some of the dips, unless I read the scales wrong, the REW peaks to valleys are greater than ARC reports.