A guide how to do room correction and use it in Roon


(Magnus) #101

I agree, first fix speakers and electronics, then setup speaker positions and listening positions, then treat room acoustics and only as a last step do room correction. In some cases, room correction might not be needed, and it some cases it will very much improve the sound.

I think one partial reason about the “loss of energy” and feeling of thinness is because most people are not used to a flat frequency response. But in some cases, room measurements can give very localized readings which turns out wrong, which is why I suggest RTA and averaging measurements over an area.

But when all is said and done, its your ear that should be the judge. Some people like +15 dB on bass and higher frequency (my wife’s son being one), some like an almost straight house curve (for example, the default curve in Dirac is very flat).

Personally I like a curve like this, makes it a bit more fun and livelly:

The reason you need (or want) a house curve is because some frequencies does not sound flat even if they are, for example bass sounds lower at equal energy, and higher frequency sound higher. So to get a flat sounding system, you need a house curve. The default in REW is something that will produce a flat sound.


(Magnus) #102

For whoever is interested, this is the house curve I use in REW (create a text file and select in from Settings -> House curve), first number is frequency and second is dB adjustment. Remember to set HF/LF slope to 0 in settings though if using an explicit house curve, or they will both be applied.
30 6
35 7
100 3.0
500 1
2000 0
15000 -3
20000 -6


#103

If you don’t want to have any house curve baked in with actual room correction, you can do your drc using a fairly flat curve eg the rew/dirac defaults, and then use a somewhat simple PEQ in Roon to add a house curve that is easy to fiddle with. House curves, tonality preferences should be kept separate from the room correction, which is harder to tweak on the fly. imo.


(Magnus) #104

I think that depends on the speaker and result of measurements. For example, my speakers have a “built in” house curve, so bass is already a little higher and higher frequency drops off at about 1.5 db/octave. In that case, it feels wrong to first equalize them to totally flat, and then equalize them back.

I have also heard someone mention that adding Parametric EQ to Roon lowers the sound quality, although for me I haven’t been able to hear any difference.

You can still add a Parametric EQ to fine tune the sound, no matter what house curve you used. See my point 16a.


#105

Just as I was ready to start remeasuring with your suggestions, the SPL is suddenly measuring around 65dB (vs the 35-40dB that I had before), mostly in the 20Hz - 50Hz range. I suspect that a neighbors must have their air-conditioners on or some other equipment. So I unfortunately have to wait until next weekend and hope that I get a quiet window to re-measure.

In the meantime, I am enjoying listening to my music with the convolution enabled. Despite the negative comments from before, I find that it sounds more natural and is less fatiguing, even if it is “lean” at times when compared to certain aspects of the sound without the filter.


(Magnus) #106

I sometimes gets a peak at 50Hz here, nothing I can hear with my ears but it shows very clearly when using the RTA without playing anything.

Remember that dB is a logarithmic scale though, so it might not influence the measurement as much as you might think ( each 10 step in dB is 10 times higher effect). For example, adding 70 dB and 80 dB produce less than 81 dB.


(Steve) #107

Having followed this topic with interest (but not yet found time to get the mic out and measure), I’ve been thinking about REWs impulse output.

Why does it simply output the parametric EQ curves, when it has the entire measurement at its disposal and could output a full convolution at much greater accuracy?


(Magnus) #108

It outputs 131k taps, but I am not sure exactly what those are :slight_smile:
It can also output timing/phase information, but only for a single sweep measurement which makes them unusable for room correction in my opinion. As soon as you average sweeps, or use RTA, only the frequencies and amplitude corrections are included, but in the form of an impulse response made for convolution.

But this is best explained by someone who knows the math and theories behind FIR and convolution/impulse filters. Maybe some dev at Roon can explain?

Personally I don’t understand the need for the complicated filter generation, it seems so much easier to just generate a +/- dB adjustment for each Hz and interpolate between them. Making that adapt to a flat curve would be very easy and quickly computer. But I guess the convolution engine can’t handle that.

Anyway, its the result that’t most important and it does seem to work very well.


(Steve) #109

Ah OK, perhaps I misunderstood. I thought I’d read in one of the articles that it’s impulse file was exactly the same as just the PEQ’s.

Yes, indeed the results are all that really matters beyond just understanding all this stuff. I really want to try your technique. But finding a window of a few quiet hours with the lounge to myself - that seems more out of reach than programming the maths for these filters myself!


(Magnus) #110

[quote=“hifi_swlon, post:109, topic:23800, full:true”]
Ah OK, perhaps I misunderstood. I thought I’d read in one of the articles that it’s impulse file was exactly the same as just the PEQ’s.
[/quote]Its mostly correct though, what I do in this guide is correct amplitude for frequencies, first individually for each speaker and then a final adjustment if needed (step 16a), and the corrections are based on the EQ.

I compared this to Dirac (used the trial), which also does time/phase stuff, but I can not hear any difference, so whatever extra Dirac does has a small impact on the sound (people with better ear and equipment than me might hear a difference though).


#111

Thanks Magnus for this very helpfull tutorial. For a long time I listened to a system with too much low for my taste, and I am very happy with the end result your tutorial delivered me.

I do have one question however. I have fofllowed your tutorial fully and completely up to step 15. In my correction I only adjusted peak responses, so never have I boosted a specific frequency. When I apply the convolution however (this is the only DSP that I apply) I get allot of clipping issues. I need to introduce a -3.5 dB headroom to get rid of the clipping.

I do not understand why i have these clipping issues since i haven’t boosted a single frequency. Maybe you could shine some light on this? Does the introduction of headroom deteriorate the sound quality at this point?


(Magnus) #112

That does sound strange, in REW EQ open the filters (the “EQ filters” button") and check the gains to make sure. If you compare music level with and without convolution filter applied, how does their volume compare?


#113

Hello Magnus,

De filters in REW only show negative gains. Volume wise there is no difference at all between convolution filter applied and disabled.

update
Some more information; I performed the measurements at 75 dB according to the SPL meter.
However after completing the measurements, when I press the “Set target level” in Target Settings and finally press “Match response to target”, I get a message stating that 78% of the response is above target. I don’t know if this is causing the issue.


(Daniel Beyer) #114

From what I understand, reducing doesn’t just attenuate, it also alters phase relationships which can lead to an unintended boost in surrounding frequencies. That is why you need to add some headroom.


(Magnus) #115

Its probably what Rugby said. But also, don’t be to afraid to boost, take the below example from my latest measurement (had to redo since I added some acoustic treatment):

I wonder if my right speaker is feeling ok, that 10 dB drop at 2k looks bad.

Btw, I use volume leveling and have never had any problems with clipping, so that’s an option.


(Magnus) #116

[quote=“Anthro, post:113, topic:23800, full:true”]
update
Some more information; I performed the measurements at 75 dB according to the SPL meter.
However after completing the measurements, when I press the “Set target level” in Target Settings and finally press “Match response to target”, I get a message stating that 78% of the response is above target. I don’t know if this is causing the issue.
[/quote]That’s just a warning, saying that most of your measurement is above the target curve. If you don’t wan’t to boost, its good to have most of the measurement above target. I assume you set the Max boost and individual boost to 0 in REW to do that (but see my previous post about not being to afraid to boost).

Here is how I do to set target: I try to get the higher frequencies on target as much as possible, but use same target for both speaker measurements. See the picture in previous post.


#117

Thanks for the feedback guys. I am going to continue my experimenting. It’s already becoming very addictive to be honest. :slight_smile:


(Mark Allen) #118

And I thought it was just me being addicted to Room correction. :relieved:


(Pete) #119

My own experience having experimented with Dirac earlier this year - better imaging/staging but at the cost of dynamics/musicality. There must be a penalty for stretching / compressing the signal at different frequencies. So personally I prefer without RC.


(Magnus) #120

I tried that and it worked very good, doing EQ correction in REW with a totally flat curve, and then adding the 2 below filters to Roon as a house curve. As far as I can tell, the audio quality is the same, but the house curve is more easily modified.

The curve below is a little heavy on bass and light on reduction on higher frequencies, but I kind of like that:

With this curve active, and the room correction active as well, I get the following measurement when I play Pink PN from Roon, and measure in REW with the RTA and moving microphone as described in this guide (the blue target line is te default house curve in REW):