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

(Magnus) #141

Actually, 90 deg is pointing towards ceeling, and 0deg is pointing towards speakers. See below quote from UMIK-1 page:

We provide two calibration files to be used depending on your application.

  • For stereo system (e.g. 2ch dirac live, single speaker measurement), use the 0deg file and point the UMIK-1 at the speakers
  • For multichannel system (E.g. 5.1/7.1) or a surround application where multiple speakers are spreadout around the room, use the 90deg file and point the UMIK-1 at the ceiling.

(Steve) #142

Thanks will try both then I guess and see what difference there is.

Yes, previously I was using the 90 degree for upright with my umik-1. Why they don’t just call them horizontal and vertical to make it a bit clearer I’m not sure.

As an aside, I read that the only measured calibration is the zero degree. The 90 degree (vertical) is calculated from that based on some historical data.

(Magnus) #143

I did a comparison between pointing the microphone towards ceiling and using 90 degree calibration file, and pointing the microphone towards speakers and using 0 degree calibration file. As you can see in the picture below, it matters very little which one you use, at least in my room (don’t laugh at my ridiculous bass behavior in my room :slight_smile: )


That’s some bass response! :slight_smile:

I’m glad to see your results comparing 0˚ and 90˚. I still have not been able to remeasure with the 90˚ since there seems to always be too much background noise in the lower frequencies and my SPL is measuring 65dB with all of my noise making appliances & equipment off.

I’ve reached the point where I’m about to try to walk around my building with my laptop and UMIK plugged in and pointing in various directions to try to find the source of the noise. :wink: But since it is all in the 20Hz - 50Hz range, I’m not sure I would be able to track it down. It’s incredibly frustrating as I feel that I am so close to a much better room correction – for my room since the 0˚ meant that my spirals should have always been pointing fairly accurately and consistently at the respective speaker and as I am not sitting far from the speaker the margin of error is quite high. The 90˚ calibration and orientation should decrease that sensitivity.

(Magnus) #145

My mic measures around 70 dB at 50 hz in my room (nothing I can hear though). I guess its some electronic ground loop or something that causes it. But I have measures at 80 dB anyway, and it worked good (use ear protection at 80 db or higher though, pink noise is not very nice to listen to).

(Steve) #146

Today I finally found a window to do it, and will definitely agree with that! Even at 70 db in my case I still wanted headphones for a bit of cover.

Sadly no time to write up properly right now, but after a little confusion along the way, and a couple of cockups, I made it through.

Did it all in a real rush and not ideal conditions, but first impressions are favourable!

(Steve) #147

OK, a bit more detail…

It was REALLY windy today outside, and living in a city doesn’t help things like measurements - I must have picked literally the worst time - people outside slamming car doors, horns, people talking, leaves in the trees blowing like mad… Actually I think it was like that last year on the day I did my Dirac measurement!

Anyway, I had a short window with the house to myself and a gap in work, so had to just get it done.

I’ve read this guide a few times and played with REW in the interim, but I wasn’t 100% sure I knew what to expect.

I wanted to do it all through Roon since thats one of the benefits of the Pink Noise techniques - that there’s no phase/timing stuff - and it seemed sensible to use the whole chain from the start as I’d like to measure the result too.

I saved out the Pink PN as Wav files, (and a log sweep for later) and added to Roon. I was a bit confused as to whether the volume levels were baked into the files so I just ignored and moved on.

I made a zone from my laptop, using Devialet AIR (wirelessly on this occasion) to the Amp. That worked surprisingly well. So, long USB cable on the Umik-1, into laptop, and on to the Amp via Roon and Devialet virtual sound driver.

So far so good.

The first faff was realising that I needed to mute one speaker at a time. I found the answer in Roons procedural EQ > ‘Mute Channels’ filter. (If I did it again I’d save presets so I didn’t have to keep going in through the menus to toggle the channels on and off).

Setting the levels made me fear for my speakers a bit. I ended up at about 70dB, and even then I stuck my Senheissers on for a bit of protection (and sorry neighbours). I’m not sure if PN is harder on the speakers at not - especially since the averaging method means its putting that noise out for extended periods - but in any case I erred on the side of caution and didn’t go as loud as I should. There’s a lot of b/g noise where I live so 75 would have been better for sure, but even at 70 the amp was up pretty high for my listening levels.

On the the measurements. That was kind of OK. I forgot a few of the settings in the RTA window, and didn’t quite have the technique of what to press in what order. Took a few goes to get it right - the Pink noise is pretty distracting so once that was running I tried to get the rest done quickly. The it took a few goes at a technique for moving the mike- speed and path. Obviously its not possible to get out of the way like it is with log sweeps with the mic on a stand, but I used the pole part of the stand to get a bit of separation.

Measurements done, I went through the tutorial again and played with EQ - following the same settings as much as possible.

After I’d done the first channel I was pretty confused about what to do next. I couldn’t for the life of me work out how to save the result, or how to view it back in the main window. Re-read the guide a few times to see if I was missing a step. No, just couldnt get it. It later dawned on me that maybe there was no way to save and REW just kept it ‘live’. It worked for me that way, but if there;s a better way Id like to know.

I did the other channel, even letting some bass frequencies boost (I was in a rush), exported them (a bit tedious - shame REW doesn’t just have an ‘export for all sample rates’ feature), and pulled the zip into Roon.


It worked. Except it sounded distorted. I see the clipping light very firmly ON!

OK so this bit I really don’t understand, because my target meant nothing was boosted above the average measured signal, so I wouldn’t have expected clipping. Maybe this is where my understanding falls down vs Dirac. It actually took a -10dB headroom to get the clipping under control, and it seemed to me this should have been an option in the convolution itself? Other wise if I disable convolution, I’ve got a bit volume drop. That said, when I disabled convolution the volume didn’t change, so I’m not sure why there’s so much clipping if the filter isn’t making the sound louder? One to research further another time…

Anyway, first impressions after the shock of distortion was pretty good. As you’ll see from my measurements below I’ve got a horrible room, and uncorrected sound isn’t very nice - lots of bass suckout and boom, and excess treble - the result a thin and slightly harsh sound. Through PMC’s that are pretty unforgiving, its been killing me not having Dirac running - finding it was a revelation.

With the Convolution, things are much like the generic PEQ I was using previously in terms of overall sound feel, but with much more definition and better stereo image. The PEQ had taken ages to get to where Id ended up - eyeballing from old Dirac screenshots - and was never quite right. Looking at the difference between left and right speakers its obvious thats a lot of the issue.

Tonight I’ve had a bit of time to play with REW and try and make some better filters and house curves. Shame you can’t just click and draw a curve, but I see Magnus posted above how to get a curve in using numbers, so I may try this. Hopefully get a chance to listen to it properly tomorrow.

Here are the ‘pointing to ceiling’ measurements (I may have messed up saving the 0degree ones). Sadly I ran out of time to measure the result which is annoying as thats the bit Im most interested in and not sure when I’ll be able to get all the kit out again, but never mind. Little steps…

Some observations / questions / thoughts from todays session:

-Log Sweeps are definitely nicer to listen to than Pink Noise!
-There’s an uncertainty in my mind about the technique for moving the mic - in terms of speed/distance/duration.
-Are the speakers being stressed? How long is too long when playing pink Noise at relatively high volume?
-Its a real shame REW writes out a convolution file only of the PEQ’s and not the complete picture of the PEQs and the measured values, which used together could give an absolutely perfect result and contain a lot more detail and precision.
-Ultimately, it would be easier if it were possible to import PEQs from REW, into L and R channels in Roon. That way you would also get a visual indicator in Roon, and could manipulate the EQ’s further.
-I would never have been able to work this out alone - complete respect to Magnus for taking the time to document this approach and all the steps needed!
-I’m not 100% sure I had the correct calibration file. I added one in the settings, but when I connected the Umik, REW prompted for a calibration file. When I changed from horizontal to vertical, Im not sure the caleb file was picked up. Seems its not that important - not relative to my room anyway!
-Dirac still wins hands down for ease of use, and UI, but then again its relatively expensive, and cant be used with Roon so isn’t an option. That said, I think this whole setup restricts Room Correction in Roon to the more adventurous and technically minded people (and its already limited to that to some degree already), which is a shame.

(Magnus) #148

That decline at higher frequencies makes me wonder if you didn’t use the wrong calibration file. The difference from measurements with microphone towards speakers or microphone towards ceiling is only small if you use the correct calibration file for each measurement (90 deg for microphone towards ceiling and 0 deg for microphone towards speakers).

You speaks about pink sound, you should use “Pink PN” sound and not “Pink” sound, and I don’t think they are that bad for speakers. You also need to use higher volume, your dips looks like they hit the sound floor.

If you want to apply house curve manually, then you can set LF/HF slope to 0 in REW, so you correct towards a 100% flat target and add the house curve in Roon. The result will be the same (or very similar) but it makes it easier to fine tune the sound. This is how I do it, below is the house curve I use in Roon

(Magnus) #149

About setting target: I usually find a place on the measurement that looks flat and set the target to that area (same for both speaker), on your measurement that would be 300 Hz to 1kHz at around 65 dB.

(Steve) #150

Haha, 65 dB I picked by eye too! :wink:

Yes, I was using Pink PN.

The treble looks fairly consistent with some old Dirac measurements, I think the PMCs have a bit of excess treble (and a big ‘BBC dip’ as I’ve heard it called, at 2k).

Yeah I need to do again louder. I’m using this as a trial run but at least I have some data to play with.

I’d actually set my target curve a little more ‘Dirac-like’ but I think I’ll try your PEQ method which suits the way my brain works better, and you’re right it makes it much easier to experiment with EQ. I guess the always the thought dies it degrade SQ pulling the signal dead flat on the convolution, only to adjust it again with PEQ. Shouldn’t make a difference I know, but they are two different pieces of code as far as I’m aware.

Any ideas on the clipping?

I’m enjoying it so far but know it can be better. I’ve upgraded amp since using Dirac (over a year ago now too) so a comparison isn’t possible, but I have a feeling I’m not at that ‘wow’ moment I got from Dirac yet. But it’s much improved on my ‘eyeballed’ EQ.

Your room’s not that bad BTW. I’m guessing the large majority of the population - even audiophiles - would see similar if they measured their hifi a domestic, untreated room. That’s why I’m so amazed people (including me, although less since I discovered all this stuff) deliberate hifi components sounding bright or boomy or whatever, and buying exotic cables to try and adjust things. Who knows what the rooms like.

Looking back at my old Dirac measurements, I think the lowest frequency peaks are new. Either a speaker position change introduced this, or it was the wind!

Out of interest, how many averages do you generally use?

(Magnus) #151

I often use 15 averages/spiral, so 75 in total for each speaker. Movement speed is something like 5 cm (2 inch) per second. Don’t know about clipping, I even have +7 dB on speaker setup after my room correction without ever getting any clipping. But I also use volume leveling which helps.

Lots of people spending lots of $ on minuscule improvements, while sitting in an untreated room without room correction. And many of them are scared to make changes, thinking that it would take them away from the sound the artists intended (the sound that was achieved in specially built room with super-acoustic treatment, i.e. a studio which also often uses room correction even with all that acoustic treatment).


I gave up waiting for the ambient noise in my room to get down to 40dB, so I raised my test signal to 85dB and tested using the 90˚ calibration and process.

Before I did this, I also re-checked the speaker distances and levels that were set in my receiver and manually adjusted them (manual measurements rather than relying on the auto measurements) and used the UMIK-1 to more accurately adjust levels (vs my previous method of using the Checkmate SPL meter). Even without he convolution filters, this make a noticeable improvement in balance and stereo imaging. It’s an obvious step and a silly oversight on my part – I thought I had previously set these correctly and did not bother checking them again.

The measurements:

I then added the convolution filter in Roon and remeasured in REW in Stereo just to see what the result looks like:

The sound:

  • it was a much less prominent difference than the previous round of filters I had created, but almost all of the negative effects were now avoided (esp the thinning of the sound signature)
  • the bloom I was getting with sound sounds like certain vocals, pianos, strings is much more in control with the filters in place and the instruments are easier to place
  • bass is ‘better’ – more controlled, i can hear some of what was missing before and it has added significantly to the foundation and drive of the music
  • overall, it is a cleaner and less fatiguing listening experience
  • i tried to apply PEQ settings in Roon to adjust further and to raise the lower bass butI I was not happy with the results so I ended up deactivating them
  • i felt that some of the atmosphere was overly stripped out (which prompted the below)

Tweak (20 - 1k):
I then adjusted my filters to only 20 - 1050/1500 for the channels based on the target curve mapping (as per @Magnus suggestion):

While you can see that there is a significant spike in the higher frequencies, with some music this still sounds more natural (in others it is overblown).

I’ll be swapping back and forth between there two filters for a while until I determine which I like more for most of the time.

Of course, once my room treatment comes in and my subwoofer is repaired, this process begins all over again. :slight_smile:

So far, either of these two filters have made a very enjoyable improvement. I look forward to the music.

Update: I have also added a PEQ High Shelf filter at 4k (-2dB) when using the 20 - 1k filters to help tame the excessive high frequencies. I’m liking this a lot now.

(Magnus) #153

Nice work, you should hear a definite improvement. You might want to be a little careful about the bass dips you have at 78 Hz left and 52 Hz right. You can try regenerating the filters with lower maximum peak and overall adjustments. The corrected measurement wont look as nice, but it might improve the sound.

You can also try generating flat results, and apply the house curve in Roon to make it easier to fine tune (see my post from a day earlier with the Roon PEQ screenshot).

(Steve) #154

Refined a new set of filters. Still need to do new measurements but for now I kind of like what I’m hearing.

Still confused about the clipping, as the max boost here is about 5db, yet I need 10 db headroom to stop the clipping. Then I finish the bass end of the house curve in PEQ, and that needs another -5db gain, so all in I have a 15db drop in gain, where I’m only boosting about 10db in total. Can’t work that one out.

Anyway, I have plenty of power in the amp so its nice to finally crank it up a bit. This really gets the bass pumping like I haven’t heard since I moved to this house!

I’m now going to try a set not boosting any bass, but really I need new measurements, so wont spend too much time on it.

I was going to do the whole house curve in PEQ, but I’m kind of happy with the treble, so Im going to experiment just with the bass - which even small adjustments in freq and gain can have surprisingly big effects (good and bad) so once thats refined I might try and put it all back in a house curve in REW.

(Magnus) #155

I heard someone else also had problems with clipping, maybe you can get clipping internally during the calculation inside the convolution engine, or something similar. Looking at your adjustments, you should not have any clipping problems in the output to the DAC.

(Marco) #156

I have a similar experience as Steve, I am not bumping up any frequency more than Steve is doing and still I need a -7 headroom to play without clipping. Some records works with less but for example 40 Licks from TIdal need a lot of headroom. (Probably the recording is not great at the start but I use it s clipping benchmark)

(Andrew Cox) #157

Hi Steve,

As you’ve found digital clipping can occur where frequencies are only being cut. You avoid this by creating headroom until the clipping light stays off, which is what you’ve done.

It feels counterintuitive that cutting frequencies can cause clipping. Let’s ask @brian if he wouldn’t mind explaining how that can happen.

Edit: In very broad terms I understand that the transfer functions used to perform digital operations on signals have phase effects that can result in localised reinforcement peaks, even when the waveform as a whole is being attenuated.

Edit 2: This article has a good non-mathematical description of the requirements when fitting a Dirac processed signal back into 16 bits. The same issues occur when Roon is fitting the results of its 64 bit DSP back into the bandwidth of the output.

(Steve) #158

Thanks Andy. I was aware of this phenomenon, but assumed it was quite minimal. But I have an admittedly limited understanding…

I made a new set of filters last night with zero boosting, but haven’t had a chance to check what minimal headroom I can get away with.

(Magnus) #159

[quote=“hifi_swlon, post:158, topic:23800, full:true”]
I made a new set of filters last night with zero boosting, but haven’t had a chance to check what minimal headroom I can get away with.
[/quote]I have read some suggestions that you should never do boosting anywhere in the measurements, which can be easily achieved in REW by setting the Individual/Overall max boost to zero. The reasoning behind this is that most dips are destructive room nodes and they should not be boosted. However, I think that’s to restrictive, but care should definitely be taken.

You can easily disable generated filters in REW in the “EQ Filters” window, so when I see REW boosting a sharp dip I manually disable it, but I allow safe boosting, see this picture for a couple of examples:

(Steve) #160

I set the max and individual boost to 0bd on the latest. Will share some screenshots later.

To find out what’s truly destructive, what I plan to do on the next (proper) measurements is boost all dips, then re-measure the corrected output. Any dips that remain (or are worse) can be assumed to be destructive and will be omitted from the final filter. That way anything that can be corrected will be. Seem like reasonable logic?

One thing that’s dawned on me about this approach vs sweeps, is that reflections are just being calculated as part of the overall frequency amplitude, so it may mean some frequencies are reduced more overall than they should be - something like Dirac would use time dependent filtering and remove later sound.

How much of an issue this is, and phase for that matter, I couldn’t say without an up to date comparison. Certainly the result using this technique are far better than the bit-perfect sound, or my manual PEQs.

[I did find the REW filters often overlapped so it wasn’t always easy to disable ones controlling individual peaks/dips in my case. I guess it depends how complex (erratic) your measurements are.]