Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Flak]

I'm pleased to see that Uli himself has stepped into this thread.

I think that there is agreement on what multiple measurements do when only frequency response is taken into account.
How much gain/attenuation limitation can result from averaging depends on the number of measurements.
As a result averaging this way works like a measurement based limitation on how much gain/attenuation there will be for each frequency.

As Uli says designing a filter this way is probably not a good idea to start with, or at least not the best idea.
A lot of information is discarded (the phase) so any solution hoping to correct the phase cannot throw this information away, and Dirac Live does this by addressing the problem from the time domain.

But in the time domain you still have the same issues... using one impulse response gives less information than using two.
Can two impulse responses being just 30 cm apart be all that different?
It depends on the speaker and the room.
Of course it is always possible to be conservative, but that will affect your end result because using multiple measurements gives more information that can be used to determine what can be compensated for and what should be left alone.

This is a very interesting subject and it could be discussed forever but the paper that I mentioned before explains these concepts with more available space: http://www.dirac.se/media/12044/on_room_correction.pdf
of course it does not go into details on how we do it, but after all Uli is reading our thread  
Among other things it explores what we can measure versus what we hear (i.e. from the middle of page 7)... we are looking for an ideally perfect music experience so the measurements should try to conform to that.

Ciao, Flavio

[Mitchco]

My contribution is here - https://drive.google.com/file/d/0BxdmSMpV-t3GRTFFa18tdWJKME0/edit?usp=sharing

I couldn't manage 6 positions as I couldn't work out a way to get the mic into the 6th position without being in the way of the speakers. Floor plan is attached, naming convention for the measurements is

<mic position>-<speaker>-<DRC status>

so 1-L-On means mic on the far left of the pic, L speaker with DRC (provided by acourate hosted by jriver) on, acourate is a single measurement system and I use position 2 for measurements.

Speakers are MK MP150 Mk2
Attempts at close mic measurement can be found in this post - http://yabb.jriver.com/interact/index.php?topic=88942.msg614015#msg614015 - it's a bit tricky though as there are 3 tweeters that have different crossovers for dispersion control reasons.

My comments;

- room is completely untreated except for normal furnishing, it's a lounge home cinema system I use for music
- the amount of correction for my sub is quite large, the results without correction are therefore quite meaningless so I would ignore <100Hz
- it seems I wasn't super careful with mic orientation so you might want to ignore >15kHz too
- there is certainly evidence of suboptimal correction in the other locations <1kHz
- one might expect the system to sound rather bright without correction as there is consistently a dip in the low kHz range and then it rises up again towards 10kHz before rolling off, in contrast the "on" view consistently rolls off as per the target curve
- subjectively... it sounds vastly superior with correction on, both for stereo and for surround

I've attached a graph (averages.jpg) showing the average of the 5 positions at 1/6 smoothing for the L and R with DRC on and off.
There are also some more graphs (1/6 smoothed) showing the individual collection of L and R measurements with DRC on and off.

FWIW I know the rationale of taking multiple measurements but I did take a conscious decision to optimise for a single seat (mine). Position 3 is where you go for a lie down during a film and is just outside the axis of the L speaker, attempting to account for that seems a waste of time. Position 1 is tucked into a corner and has always been rubbish so if you care about sound you don't sit there. Positions 4 and 5 would involve leaning forward so file under not going to happen. Despite all this, my wife reports that it sounds great from both those positions & much improved over no correction so what do graphs know  

Hey Matt, I downloaded your REW .mdat file and having a look at your L and R averages with DRC on:



This is certainly within the +-5 dB window I mentioned in an earlier post and is +-3 dB from 100 Hz on up.  Given your room layout constraints - must sound pretty darn good and looks to me that the DRC is indeed doing its job across the listening area with one measurement used for generating the filters.

Congrats!

Cheers, Mitch

[Hest]

I have achieved superb sound in my living room with active speakers and Audiolense XO.

Everything changes when each driver can be corrected individually instead of correcting the speakers as one.

Multiple measurements are not needed. At least not with my setup. I have tried but never got any improvement.

[)p(]

Hi,

I have been playing with the dirac trial. For practical reasons I had to move the speakers in a much worse position. So I was interested in what drc could do here. Red and green are the un corrected speakers. Blue and purple corrected. As you can see from the attached screenshot it certainly does its work in the bass region.  Also imaging is a bit smaller but with better focus. And the results for most part are within Mitschko's +- 5db range. Although I think the results are positive in this setup I do prefer the results I got with Denis Sbragion open source drc software. Loopback with jrmc does not work on the htpc so I could nor measure Denis drc output with rew. I have not tried acourate or audiolense yet.

[)p(]

After some feedback from dirac I redid my measerments with the mic directed fowards instead of to the ceiling as was advised in the dirac manual. This is because my mic was calibrated on-axis.

The result sounds very good to my ears. In the measurements you can see that the new filter flattens the high end a little better then the previous one.

I think it now sounds at least as good as with Denis Sbragion's DRC soft filter. Because DRC is free and the differences are small it's an easy choice to stick with the latter. Although I will really miss the virtual sound device and thus the ability to AirPlay spotify to the htpc with correction applied.


If you want to try Denis Sbragion's DRC, it's very easy with the GUI frontend made by Alan Jordan:
DRC Designer:http://www.alanjordan.org/DRCDesigner/HelpFrameset.html


I attached measerements of the left and right speaker with dirac on and off.

[Flak]

This subject is treated by many, including myself, in a somewhat simplistic way... on the other side a full discussion would involve a lot of math so I have asked Jakob to try translating those concepts into english wording and in pills:

* If you measure in a single point you cannot know what is common, and what is different due to room reflections, in different points in the listening area.
Without that information it's impossible to tell what is robustly invertible.
Using information from multiple measurements the inverse can be made as detailed as allowed by the spatial variability.

* By applying heavy smoothing to a single point it is often possible to capture the behavior in neighboring points as well, as the smoothing will remove most of the details.
(Minphase) filters based on such a model are likely to be spatially robust in an area around the measured position.
The resolution of the ear have very little to do with the performance of this filter, what matters is if the filter is valid for the behavior to be corrected in points around the measured point.

* For this to be true the filter will have to be smooth in most cases, but with some exceptions.
Assume you have a speaker with a jagged response placed in a room that does not affect the response significantly.
In this case the response will be similar in a large area, and it is possible to design a filter based on a single measurement that is spatially robust and has a high resolution.
But using a single measurement it is not possible to know what resolution can be used as there is no information on what is common and what is different.
In cases where the direct wave is not dominant averaging over space corresponds reasonably well with averaging over frequency.

* If you want to correct the impulse response (not just the amplitude response) and make sure the resulting filter do not introduce pre-ringings you again need information from more than one point.
It is imperative to find out what is varying and what is static, and that information cannot be synthesized from a single measurement.
Again, various smoothing techniques (in the time domain this time) can be used to make do with only a single measurements, and sometimes the smoothed model will be a good approximation of the behavior close to the measured point; sometimes it won't be.
How well a measurement can be made to represent a large area varies from room to room and speaker to speaker.

This is it, ciao
Flavio

[AudioVero]

This subject is treated by many, including myself, in a somewhat simplistic way...

Yes, but I fear that "somewhat" is just an optimistic description.

* If you measure in a single point you cannot know what is common, and what is different due to room reflections, in different points in the listening area.
Without that information it's impossible to tell what is robustly invertible.
Using information from multiple measurements the inverse can be made as detailed as allowed by the spatial variability.

Different averaging algorithms will give different results of "common". Your result may be robust under mathematical aspects (e.g. invertibility). This is the same with other "common" results (calculated by power response, Sammon maps, envelopes, arithmetic mean, geometric mean, moving average, weighted average, root mean square etc.). Are the "common" results now identical or not? If not, why are they called "common"?

But does this give a real answer to the example below?

Let's assume some arbitrary values. The average is 0 dB at 40 Hz (we just pick one frequency for discussion). Curve 1 shows -20 dB and curve 2 shows 3 dB.
If the target is 0 dB this means that the correction is also 0 dB as the average is already there. So listening point 1 still has -20 dB after correction, point 2 is still 3 dB.

But how is the average calculated? There are many different algorithms to calculate an average. So another algorithm may lead to an averaged curve with -6 dB at 40 Hz. To reach the 0 dB target the correction thus will boost by 6 dB. So at the listening position 1 we get -14 dB, which is an improvement and at position 2 we get 9 dB , indeed a deterioration.

At the end we get one correction filter. IMO it should be optimal for the desired listening position. Optimal for the listener, not for the mathematician (who should clearly know, that the result will be suboptimal at other positions).

Flavio, are you really convinced by listening that single point measurements lead to wrong results? Tell me why. Or are you just repeating Jakob's arguments under marketing aspects? Please tell me what's wrong with Acourate correction filters.

Cheers, Uli

[Flak]

Jakob:

I hope this discussion have been informative to the readers of this forum, and it seems it should probably be ended about now. Here are a few closing comments from my side:

1. We are not discussing any particular products here, but the merits of single vs. multi point based compensation.
2. We don't know how Acourate produces the filters it does, and we have no knowledge about any flaws the Acourate filters may have.
3. As we have stated numerous times, measuring in a single point can work just fine, but in general more robust results can be achieved by using more information. If the extra measurements contain no additional information, a good compensation algorithm will disregard them.
4. The result should be optimal for the listener, but short of hand tuning every single room together with the intended listener there need to be an algorithm to produce a filter that results in a listening experience that is better than the uncompensated system. This algorithm must have a mathematical representation as it is performed by a computer.
5. Optimality by definition implies the existence of a criterion that should hold. This criterion need to be evaluated by said algorithm. There are probably more such criterions than there are compensation techniques. And again, this is expressed using mathematics.

Flavio:

I have listened at length to another good mixed-phase solution based on one point measurement (and apparently the same principles) which I do not mention as well as I have never mentioned Acourate.
In my opinion multipoint measurements led to better results especially because I do not listen from a single fixed position... but I do not think that my judgement (as well as Mitch's one) are especially meaningful, I believe that everybody should judge with his own ears (so it is important that a two weeks free trial of Dirac Live is available)

This is all, we leave you the last word
thanks for your time, Flavio

[dallasjustice]

Multi point works well for DIRAC.

Single point works very well for Acourate in my room. 

I can't say which measurement method is better.  However, I can say that I have a better sounding sytem using Acourate. I now own both softwares.  I have spent substantial time working with DIRAC. I use very good test gear.

I like the result with Acourate better. Acourate is not a fully automated process like DIRAC.  It requires some learning to use it. I am still learning so I am no expert.  I look forward to learning more about crossovers.   

[retro]

Dallas,

So you've abandoned Dirac..interesting..
Was it the crossover functionality that made you switch..?

Can u describe in what way acourate sounds better..?

[dallasjustice]

There's nothing wrong with DIRAC. I confess to having a wandering ear. I previously tried acourate out last year. I didn't have appropriate test/playback gear at that time. I was very impressed my second time with it. Now, my filter is even better with Uli's help.
Listening to Acourate after DIRAC reminds me of my experience listening to DIRAC for the first time after no DSP. Everything is more clear. Better resolution bottom to top. Just me experience. It's free to try.

The primary reason for my curiosity is crossover though. I have to get a Hilo first. After that, I'll work on that next. For now, I am very happy with Acourate.

Michael.

[retro]

Interesting!

Hope you start a thread at "the other" forum about Acourate and describe your findings.

Now I have licenses for both Dirac and Audiolense..maybe I should get license for Acourate too..

[dallasjustice]

I used to have license for Audiolense about 6 years ago.  I was too impatient to really stick with it.  So Bernt let me sell it for $100. Since then I've grown a small part of brain to allow for some DSP knowledge.   

I'll post more once I finish the crossovers and speaker correction filters.

[satbox]

Hi,

First of all, sorry for my English... I'm new in room correction but very interested to test some software (at least, Acourate and Audiolense).

I use a Macbook Pro with JRiver MC and an Ayre QB-9 asynchronous USB DAC to play music. So, my first idea was to install Acourate on my Macbook Pro using Parallels and Asio4All and to do the measurement with a USB mic (like the UMIK-1 for example) and my QB-9 USB DAC. But after reading this topic, in order to perform good time domain corrections, I think it's not a good idea    Do you confirm?

I understood that I have to use the same device for input/output and avoid to use Asio4All. Is it correct?

So, is it possible to use my Macbook with Parallels without Asio4All? If yes, what equipment do you advise me to buy (mic, etc) in order to do accurate measurements and, preferably, use my USB DAC to output the sound to my amplifier?

Thank you for your help!!!