TOPIC:

I'm going to be doing a Dirac trial soon and I'm trying to get organized. I have a couple of questions about positioning the mic when taking measurements:

1. I'm using a Umik-1 that is calibrated for 0 degrees so (as I understand it) the mic should be horizontal and point toward the speakers. My question is whether I should keep the mic perpendicular to the plane of the speakers or should I point the mic at a position mid way between the speakers? I'm not sure if it even matters but I'd like to know.

2. Based on reading the threads on CA and some reviews, I'm going to do 2 filters. One as "couch" and one as "chair." Doing the couch - literally - would mean that the mic position for some of the measurements would be directly opposite of one speaker (e.g. the mic would be directly across the room from the speaker). Is this a problem? Should I keep the mic inside the speaker position? My speakers are about 6 feet apart. If I went out a meter on either side of the sweet spot I'd actually be outside of the speaker.

Thanks in advance for assistance.

The software will guide you where to put the mic.
I did two calibration sessions.
One was a simple "get in touch with the product" during which I did not relly follow the instructions.
Correction was good but yet not perfect.
During the second session I did follow the instructions by the book and got what seems to be the best result possible in my setup.

I would strongly advise you to simply follow the Dirac calibration software's instructions.

I understand that you will take measurements with the mic in various positions, but I am not sure how to orient the mic itself.

I think I've usually left it pointed straight ahead, but it can't hurt to point it inbetween the speakers on the extreme left/right locations.

So it worked for you straight ahead? Thanks. How wide a span did you measure? The manual is a little vague on that.

The microphone is omnidirectional, so it doesn't matter if you're pointing straight ahead or a slightly angled in. Its frequency response will only start to change at extreme angles. Just don't worry about that..

As to where exactly put the mic for each measurement, I'm afraid there's no generic answer to that. I've had more success with the couch setting and relatively far apart spots, but every room and speaker combination is different and there's no way around simply trying out several scenarios and then both listen to the results and also measure them with something like REW.

This would actually be a feature request from my part, that Dirac would allow to make another measurement (or several at different locations) to check the results in reality and compare them to what Dirac predicted (which in my experience is FAR too optimistic).

Thanks Beauvis. I had planned on trying the couch option. I guess I will just have to try a few different positions.

Beauvais wrote:

This would actually be a feature request from my part, that Dirac would allow to make another measurement (or several at different locations) to check the results in reality and compare them to what Dirac predicted (which in my experience is FAR too optimistic).

Indeed, this would be a great feature !
Even Trinnov does not have that.

Beauvais wrote:

This would actually be a feature request from my part, that Dirac would allow to make another measurement (or several at different locations) to check the results in reality and compare them to what Dirac predicted (which in my experience is FAR too optimistic).

Hello Beauvais,
I understand why you think that the predicted responses are optimistic but they are actually quite precise if you will average the measurements in the different positions... i.e.see here:
www.whatsbestforum.com/showthread.php?15212-Dirac-validation
The validation of the predicted response is discussed here... point 3.6:
shop.dirac.se/topic/11-faq.aspx#Technical_questions_and_answers

Ciao, Flavio

I only took a quick glimpse at the first link and stopped reading immediately, there's so many false statements there, can't be bothered..

The second link is interesting, but that's exactly my point. Of course the predicted curve will match a real measurement at the exact same spot. But the whole point of Dirac is that it's NOT about a single point in space. I can make the FR ruler flat in one spot without Dirac myself easily, don't need a 900$ box for that..

I'm not putting Dirac down, I'm actually quite happy with it. But a ruler flat prediction for just one spot is pretty much useless when the whole idea is improve the FR in a larger area.

It is not possible to deal with a sweet spot smoothing if you cannot weight the measures taken far from the center of the sweet spot (first measure).

Audyssey uses a clustering method that will only weight accidents (it allows to determine accident probabilities, not enlarge and smooth a sweet spot)
Trinnov allows you to manualy weight any measure or cancel any which accident is singular. In that cas you can get rid of measurements that can add unwanted corrections and you can enlarge and smooth the sweet spot.
In all cases you can still manually weight an area by taking several measurements at a determined point.

I do not know how Dirac Live implements weighting.
I guess the measurment pattern give an ideal positioning.
I do wonder how the system work that out? Is the pattern an optimized theoritical mapping? Is it just a guidance in which weighting is applied?

In my experience the goal of enlarging and smoothing a sweet spot allows to correct a wider area without compromizing the correction at the first measure.
At least not to much.

Used in a wild acoustic environnement (no treaments) is it likely that FR and impulse won't measure the same by far at different points.
It is also true in a treated envirement but the sweet sport tends to be larger.

That means that the predicted corrected FR will likely be correct at center sweet spot (measure 1), but it will differ as we get far of it.

Beauvais wrote:

I only took a quick glimpse at the first link and stopped reading immediately, there's so many false statements there, can't be bothered..

The second link is interesting, but that's exactly my point. Of course the predicted curve will match a real measurement <strong>at the exact same spot</strong>. But the whole point of Dirac is that it's NOT about a single point in space. I can make the FR ruler flat in one spot without Dirac myself easily, don't need a 900$ box for that..

I'm not putting Dirac down, I'm actually quite happy with it. But a ruler flat prediction for just one spot is pretty much useless when the whole idea is improve the FR in a larger area.

I think that there is a misunderstanding...
as you correctly say the idea is to improve the FR (and impulse response) in a relatively large area.

So you don't want to take into account the frequency response (and again impulse response) in a single point in space only but you try to get the best possible average response (the nearest to the target curve) among the different positions in your listening area (nine points in our case)
That's what Dirac Live shows when predicting an average curve... and it happens that if you take the same nine frequency response measurements with another program like REW (or Fuzzmeasure as in the first link that I mentioned) and you will average them (as those programs are capable of doing) you will find that the predicted results are very close to the measured results (of course the same smoothing should be applied)

I'll not get into the weighting applied by the algoritm but in general the nine measurements allow to understand what has to be corrected, and what not, by taking into account what changes in the different positions.
You will find more details here: www.dirac.se/media/12044/on_room_correction.pdf

Flavio

I don't know what Dirac does, but if it does what you're saying, the guys at Dirac are f***ing idiots. What good is averaging they way you describe it? I couldn't care less about a theoretical perfect 'average' response.

Because my ears are not averaged out across the room, but rather always in a specific spot. The goal must be to get as good a FR in various different individual spots around the listening area, not some theoretical average you get when calculating all curves into one.

As an example: You filter the left speaker with a high pass filter at 1kHz and the right speaker with a low pass filter at 1kHz. What do you get? A perfect 'average' response but complete bollocks in real life. It's not about averaging! It's about finding the right balance of adjustments so you get an improved response at several different locations, about improving one spot without inducing collateral damage to another.

Again, the idea of an average response is plain wrong. There are only individual locations in space for which the FR can to be optimized by carefully adjusting individual responses to find a FR that works best for all locations.

Dirac Live doesn't average the left and right speakers. So your example makes no sense...

First of all, neither of us knows what exactly Dirac does and the devs are certainly not going to make their algorithms public.

But apart from that, I think you're right and Dirac does not do that. Yet the example very much makes sense just to show that simply averaging the FR can lead to absurd results.

And I wrote more than just an example on which you might disagree on and I think my reasoning is sound. You're welcome to criticize it, but just saying it makes no sense is not a very profound argument.