Understanding the State of the Art of Digital Room Correction

[al2002]

A deep dive presentation on the fundamentals of "proper" Digital Room Correction (DRC). Includes hands-on DSP FIR Filter Designer demos using Acourate and Audiolense.

SNIP

Thanks for posting your very informative video.

I’ve had Acourate for many years and agree with your assesment of the advantages of pc based DRC as opposed to hardware based DRC. However, a PC based system is not always family friendly so hardware based DRC, e.g. miniDSP running Dirac, still has its place.

A couple of questions re. the presentation:

1. When calculating the width of the 600 Hz window I see you took 4 Hz as the lower frequency. The reason is not clear. Can you explain why 4 Hz was picked, and not, say 1 Hz which has the calculation advantage that log1 = 0, or alternatively 15 Hz, which would be a more realistic lower cut off frequency for a system with subs? Of course, the change in window width would not really matter, but it would be good to understand the reason behind your choice of 4Hz as the lower frequency.

2. In you list of recommended software packages you did not mention Rephase which is freeware. I’ve not used it, but looking at the feature list and reading the long thread on diyaudio it seems to tick all the boxes. Can you comment?

Great work. Keep it up.

P.S:- Is an updated edition of you book in the works?

 

[JRS]

A bit off topic, but one of the advantages of a PC based system would be scalability to whatever number of channels with ones encoding flavor of choice (not sure which of the major brands have a personal use license). Question:

Being an audio cheapskate, I really would rather not buy a stand alone processor if I could do it all on a computer, that is if it's at all feasible to obtain Trinnov type MCh performance, say the equivalent of their studio 8 channel product which retails for 8k? Seems like a future proof solution with the ability to use different flavors of encoding--or am I deluded in thinking this is even possible?

 

[Chromatischism]

I prefer to equalize the sub, at least partially to a workable shape, before bass management.

This doesn't make intuitive sense to me. How does the processor know what is going to the sub channel?

 

[ernestcarl]

This doesn't make intuitive sense to me. How does the processor know what is going to the sub channel?

Do you mean an AVR processor? I do not use one, sorry.

My method uses a PC to create predictions from prior gathered multi-position measurements:

Measurements are made of all channels without EQ. I convolve the weighed average. And create summed predictions. It usually works really well -- not perfectly, of course -- some adjustments must be made afterwards with actual real spatio-temporal measurements after changing/setting new bass-management parameters.

 

[fluid]

While this is true in any case, I suspect that reason for your satisifaction is also the fact that your central point sweep correlates very well with the average of multipoint sweeps/MMM. Can you please put them together on a graph with Psy smoothing?

You are correct they are pretty similar, and the closer they are to each other the better single position measurements work for me. This is a reduced set of measurements (9) in the Vector average.

that makes it useless

can you share the mdat, please?

I don't think it is "useless". This works much like beamforming where the more measurements you average the more the room response is averaged out, the trick is finding the right balance. In the most reflective room with the speakers listened to slightly off axis it gave the best subjective sound. It does require a different approach and modification to the target to account for the spectral tilt. In another room the vector average was quite awful. So the "best result depends on everything" has been true for me.

This is the difference in ETC between a single vs Average

I have uploaded the mdat to my google drive for you as it is 50Mb in size, please don't make me regret sharing it.
https://drive.google.com/file/d/1izyXY_VFWDjumdIku_1cYTvda8gTolVW/view?usp=sharing
These speakers are a full range line array and nothing like a standard bookshelf. There is no EQ in these measurements they are raw.
After EQ it looks something like this

 

[fluid]

Meditate on : "The stated or implied sales pitch is: give me any loudspeaker in any room and my process will make it "perfect". A moment of thought tells you that this cannot be true."

There is no need to meditate as this is absolutely true and I don't see anywhere where Mitch is claiming to be able to polish a turd to perfection.
If the speaker has horrible directivity and the user brute force corrects to a target then the proof of Floyd's statement will be heard. There is a large amount of middle ground between those two positions though.

When I was young and naive and had a computeraudiophile account, I thought @mitchco was a good guy sharing insights on a forum where people help each other. Fishing for customers is not bad per se but I've lost hours misguided by his half truths and twisted presentations of serious research, downward slopes only good for boom boom etc etc .

In all my interactions with Mitch I have found him to be genuine, pleasant and helpful, I have no idea what caused you to have formed such a different opinion, I sincerely doubt he had any intention of misleading or twisting facts. If you got too much boom then you could well have put too much of a slope on the response. The difference between great and boomy is not very much for me. 0.2 to 0.5dB increase to the slope or having the slope continue down to where the speaker rolls off. I have the in room response flatten out between 100 and 50 Hz and for me this helps avoid things sound overly bassy. This is something that not everyone seems to agree on.

If you have seriously engineered speakers, unless you're a guy (any lady here?) confusing the map and the territory, forget about eQ above the transition frequency, full range targets and other nasties and then the debate about FDW etc will become mostly moot and the messing with time domain conter productive

@jtwrace has posted here and in a thread he started. He is using JBL M2 speakers and multiple subwoofers. He has heard them as they are and after "correction". He can and has spoken for himself so what you say above is not universally true. An amount of trial and error is needed to find a correction that sounds good. I have literally spent years trying out different options and the difference between good and great was not really so much in the end. But I have also totally destroyed the sound with something not that different either.

Having nice looking graphs is great for showing off on internet forums but the filters used to make them need to sound good too. I think you can have both but they don't always come together.

 

[j_j]

Please elaborate. For example an analysis of low frequency room modes requires high resolution measurements. Smoothing means loss of resolution.

This is basic physics. Perhaps some other issues need to be tackled first?

You are aware of the relationship between wavelength and frequency, yes? You are aware that your ears are 6" or so apart, too, yes?

 

[mitchco]

Thank you @fluid for sharing your measurement data and observations. Folks be like @fluid As Zilch (RIP) says, "More Data, Less Wank."

 

[j_j]

Regarding your remark about wavelength vs. head position - same ratio between wavelength vs. head position is there when a live person is speaking in our rooms from the other side of the coffee table but no such artifacts occur when we move our head as the ones caused by bad correction.

Actually, no, not the same thing. The SINGLE SOURCE from the person speaking provides that "leading edge" that your auditory system locks into. Two speakers reproducing the same thing do not have the same effect (furthermore this shows very elegantly why 3 front channels are required, but that's another thread).

As to 'bad correction', that is simply a result of NOT SMOOTHING. Again, consider the relationship between wavelength and frequency. If you don't smooth, your "correction" is only valid for the middle of the head, and neither ear is there, furthermore, there's no HRTF's involved, so all of volume velocity, coherence length, and time delay come into play if you don't smooth.

Which is why you MUST smooth, single point, multi point, WHATEVER.

Your point about "can not be detected by the ear" is also valid, for very narrow, very sharp variations, of course, but conveniently the two issues work together, rather than oppose.

 

[AudioJester]

Please read the collection of Dr Toole's quotes collected by @thewas : https://www.audiosciencereview.com/...ut-room-curve-targets-room-eq-and-more.10950/ . Meditate on : "The stated or implied sales pitch is: give me any loudspeaker in any room and my process will make it "perfect". A moment of thought tells you that this cannot be true." When I was young and naive and had a computeraudiophile account, I thought @mitchco was a good guy sharing insights on a forum where people help each other. Fishing for customers is not bad per se but I've lost hours misguided by his half truths and twisted presentations of serious research, downward slopes only good for boom boom etc etc . If you have seriously engineered speakers, unless you're a guy (any lady here?) confusing the map and the territory, forget about eQ above the transition frequency, full range targets and other nasties and then the debate about FDW etc will become mostly moot and the messing with time domain conter productive since the right modal frequency eQ will be essentially coupled with the right time domain mod : https://www.hometheatershack.com/threads/waterfalls.7135/ .

If you disagree with something in the video its fine to point it out and argue its merits, with a factual basis. Iam sure Micho would welcome robust constructive debate.
However, attacking the man and his intentions says a lot more about the poster and their own insecurities.

For real, Mitcho years ago brought computer based dsp to the common man - I was one of them. Got me and many others off the component bandwagon and into great home based audio.

 

[j_j]

Rather than continue this discussion about smoothing regarding people who apparently refuse to comprehend the relationship between smoothing and "validity to position" I'm going to introduce a concept, and let you all thing about it.

The term is "coherence length". The coherence length at a given frequency is the point at which two pressure measurements (i.e. omni measurements) cross-correlation drops to 1/e.

As you can see, this is highly related to wavelength. Or so I hope you can see that.

Now, please relate that to the effective range of a single point correction on a sphere, and then to how that will allow smoothing to widen the accuracy of a correction around that sphere.

This is pretty obvious, I think. Obviously not all people do.

 

[j_j]

As an aside, the people questioning intent, and using straw man tactics to attempt to frame Mitch here are starting to convince me of malice.

I see no place that anyone, Mitch included, has claimed that one can do a "perfect" job, yet suggesting as much, and then questioning a person's intent based on that insinuation, to me, shows punitive intent.

 

[jtwrace]

@jtwrace has posted here and in a thread he started. He is using JBL M2 speakers and multiple subwoofers. He has heard them as they are and after "correction". He can and has spoken for himself so what you say above is not universally true.

Yes, and it should be noted that I've owned them for ~ 5 yrs without AL so I'm quite familiar with their sound. If AL didn't please me, I'd happily ditch it and quite frankly, that's what I really thought would happen. In fact, a couple of times when Mitch sent me a filter I was like, OMG, this is awful. Guess what? I was wrong.

 

[jhaider]

Thank you Folks be like [USER=30766]@fluid As Zilch (RIP) says, "More Data, Less Wank."

Agreed. It would be useful to see what the software does to the direct field response in the mids and treble of a well behaved speaker, as shown ideally by Klippel NFS or anechoic chamber measurements, but at least by gated quasi-anechoic measurements taken with care at a sufficient distance for the drive units to sum properly. A single design-axis measurement would suffice, though one could argue a listening window average is better still.

Orherwise, there hasn’t been much (useful) data this thread. Plots taken in someone’s random room at some single point aren’t useful or interesting data.

 

[jtwrace]

@jhaider since you're a reviewer, why not review it yourself?

 

[al2002]

A bit off topic, but one of the advantages of a PC based system would be scalability to whatever number of channels with ones encoding flavor of choice (not sure which of the major brands have a personal use license). Question:

Being an audio cheapskate, I really would rather not buy a stand alone processor if I could do it all on a computer, that is if it's at all feasible to obtain Trinnov type MCh performance, say the equivalent of their studio 8 channel product which retails for 8k? Seems like a future proof solution with the ability to use different flavors of encoding--or am I deluded in thinking this is even possible?

Not sure how many channels you need, but it is pretty straightforward to run multichannel convolution on a PC. I’ve run 4 channels of filters on a dedicated convolution PC built around an i7 3770T. The CPU was loafing. Could have easily run a 7.2 configuration on that computer.

Bear in mind that immersive audio codes are not available to run on a PC. I believe Trinnov writes their own software to decode the new formats.

 

[j_j]

A Plots taken in someone’s random room at some single point aren’t useful or interesting data.

I must disagree. There could be better data, but single point data, examined with an understanding of the acoustic and psychoacoustic considerations, can carry quite a bit of information.

 

[j_j]

Not sure how many channels you need, but it is pretty straightforward to run multichannel convolution on a PC. I’ve run 4 channels of filters on a dedicated convolution PC built around an i7 3770T. The CPU was loafing. Could have easily run a 7.2 configuration on that computer.

Bear in mind that immersive audio codes are not available to run on a PC. I believe Trinnov writes their own software to decode the new formats.

Hm, (warning, something I have something to do with), check out mixcubed.com. Try the jazz clip for starters. Free to listen to the demos.

This was prompted by the comment about immersive apps, for which you just made a very interesting point. Hmmm.

 

[jhaider]

@jhaider since you're a reviewer, why not review it yourself?

Why? I have zero personal interest in routing my audio through a general purpose computer, and even less interest in infecting my audio system with windows. Someone would have to pay me better than my day job rate for that hell.

I don’t think it should be too much to ask for people praising something to the hilt to show meaningful measurements of the thing.

 

[jtwrace]

Why? I have zero personal interest in routing my audio through a general purpose computer, and even less interest in infecting my audio system with windows. Someone would have to pay me better than my day job rate for that hell.

I don’t think it should be too much to ask for people praising something to the hilt to show meaningful measurements of the thing.

Clearly you have a beef towards Mitch, Audiolense or both based on your comments. The fact that you're not willing to use Windows alone says it all. Whatever.