Creating convolution files for Roon in the new Room EQ Wizard (REW) with Inversion

Yeah, but should the lefthand windowing even matter in this case? Because you are running a minimum phase filter, all lefthand coefficients/taps are at zero. They certainly are not affecting the frequency domain. If you were running a linear phase filter, the lefthand windowing would matter, since it would contain the the righthand coefficients/taps in reverse order.


1 Like

That’s it, AJ :+1:

And I zoomed in on your screenshots. Righthand windowing is the same 1.730 ms in both, hence why your frequency domain resolution is the same for both filter durations.


1 Like

Yes, and shortening the left window shortens the latency of the filter to audibly remedy echo effects when grouped with other zones…

Your DIY speakers are a special case where you don’t need any crossover phase correction hence the flat impulse response before time 0. It’s especially the low frequency phase shifts that generate the ringing and are very dependent on the window size.

1 Like

Hi all, hoping to get a bit of guidance since this is my first time doing any measurement + DSP processing.

I followed the steps provided in the guide. Something of note is that REW complained my sample volume was low, but I set my amp at my typical listening level.

I ended up having to apply -8dB of headroom management with this technique. Has anyone else had to apply so much headroom? Many tracks clipped, from instrumental music to pop/electronic music.

This is my RMS against the Harman target:

Here’s what I ended up with after following the steps. Wondering how it looks to those who may have done this before? Note that I have my bass rolled off a few dB on my amp at 200Hz because I have a subwoofer that I disconnected for this. The sub has to be controlled manually at all times anyway because my amp only has full volume tape-level out.

I usually need -5, -6 dB headroom to completely get rid of clipping for my entire collection according to Roon but I didn’t use any for years. -8 dB headroom wouldn’t be a problem. My personal belief is that you wouldn’t run into any serios problems with convolution even with headroom off. For the bass rolloff issue, you will need to measure and correct with the sub on as the convolution engine will correct for these frequencies with or without the sub. Just play REW sweeps from your usual system.

As promised:

This one is about subwoofer alignment but more importantly shows how to time align multiple point measurements using the cross correlation feature in REW.

Measuring at multiple points and averaging with this method rather than a single mic position measurement is giving slightly better results. I also started using VAR smoothing before inversion. However I am still struggling to invert excess phase without getting pre-echo effects. When/if I can do that, it will be worth a new Roon convolution with REW video :wink:


Thank you. Cannot wait to finish work hours to watch

1 Like

Thanks @Serkan_Gur, I took the time to remeasure the system. I took a two part approach and will share for those that come across this and have a similar setup.


  • two channel integrated amp with Tape (full volume line level) Output sends over RCA to subwoofer
  • Subwoofer is SVS SB1000
  • Volume adjustment on amp doesn’t change subwoofer level. Only SVS app (or buttons on sub) change sub level. That is, it’s totally manual.
  • I use amplifier EQ to roll off bass at 200Hz and below so speakers work less hard to carry these frequencies. This attenuation does affect the sub volume but it is VERY minimal.
  • I use the LPF set at it’s max of 200 Hz in the SVS app.

Measurement modifications:

  • Using Windows for measurements (instead of Linux) got rid of an error about timing
  • I chose to set my levels at 75 dB SPL instead of my listening level. My listening level was too quiet and REW complained.
  • I set my both speakers and subwoofer using the two different pink noise generators to 75 dB so they are matched
  • From here, did typical sweep operations and followed guide
  • Harman curve worked fine for me

Cleanup and adjusting to taste:

  • Set my speakers to quiet, normal, and loud listening levels
  • Adjusted subwoofer level to taste at each of these

This lead to a much better response than trying to measure without using the subwoofer.

1 Like

You basically managed to create different bass curves for different volumes which of course is ideal.

Why are you crossing over your speakers at 200Hz? Because they’re very small or because of the peaks in their response? I wouldn’t normally go over 80Hz.

That’s a fair question, since it’s not really what’s recommended.

For a typical listening I listen at fairly low levels (65-70 db(A)), so the subwoofer crossing over at 200Hz provides a very subtle and nice foundation to the loudspeakers, much like a ‘loudness’ feature. My amp is rated for 35W/ch, so I chose to trade bass response in the loudspeaker for more usable volume range overall.

At higher volumes, I was crossing over at 130Hz or 80Hz, depending on how loud I turned everything up. Turning the sub volume up with a 200Hz crossover didn’t sound very good, so it was primarily used that way for quiet listening.

1 Like

I see. I guess with only one sub and relatively at a central position, you wouldn’t mind localization of bass frequencies as well.

I wonder if you and John Mulcahy can come up with some kind of procedure or “filter” to deal with these phase issues like RePhase seems to be able to do… I have been looking at SoX to see if it can deal with the phase issues as a post processing pipe from the FIR.wav output from REW…

BTY I can not understand why anyone would crossover full range speakers with a sub(s) above 80Hz in a stereo system… Its not like you can save those excess Hz(s) in a mayonnaise jar to use latter on…

Excellent video, thanks.

As a point of interest, have you re-measured the response of the resulting dual subwoofer configuration? How closely does it match the predicted results?

Thank you. Of course I did. It’s almost an exact copy, maximum 0.5 db difference at no more than 2-3 points.

1 Like

Amazing, thanks!