Lessons learned redesigning convolution filters for HQP thanks to Room Eq Wizard

[Le Concombre Masqué]

Goal /the incumbent filters: get a faster more “agile” articulate intelligible sound while keeping big bass extension

 

Crisis started when I switched off convolution to compare my setup + original Argo of Jamal at Alambra to the needle drop sold as dsd download (yes yes and to much praise by Acousticsounds ; meaningless comparison since I doubt the needle drop is sourced from the distorting but sometimes gloriously tubey original) : though convolved DSD was a clear winner over non convolved DSD or LP, it sounded too analog and too tamed, fat and slow.

 

Findings

 

Do eQ, REW is donations welcome free : try it (but you need a microphone). Doubt you get comparable results from power usb whathaveyou cables…

 

RR1 (as in https://www.innerfidelity.com/content/warren-tenbrooks-summary-head-measurements-harman) is a mighty good target house curve

 

Mind applied smoothing : a predicted response curve matching Bob Katz’s target (flat to 1 k, 6 dB downward to 20K ) created with “Psyschoacoustic” smoothing in REW resembles RR1 when looked at in VAR or traditional (ie 1/3 octave) smoothing mode (Bob Katz uses Acourate’s Psychoacoustic)

 

Multiply measurements but pick the one that resembles average, not the average summation (ie in REW the Psychoacoustic profiling only applies to single measurements ; as do many options)

 

Avoid messing with frequencies above Minimum Phase zones galore (on the left half of frequencies) (850 Hz max here, different from the theoretical 300). 

 

Lower Shelves are great : found out that RR1 paralleled the measured response in my room and that a 5 dB 600 Hz start LS puts frequencies below 600 on the RR1 target

 

Dare switch to Manual. Automatic was not at ease, I took the time to flatten the bumps and, up to 9 dB in Minimum Phase zones, fill.  

 

Listen (Stereophile CD3 and best of Chesky 2 have been most helpful, mainly the tracks with moving point sources (acoustic clickers, cowbell…) ). I did not have at hand the mike I created the measurements with, only a iPhone + SPL app, useless below 250 and probably giving a false 3 dB bump reading at 500, but measuring would be great too. In the end listening matters more.

 

Bottom line :girlfriend threatened to quit madman but is as amazed as I am in the end : WOW ! Goal met

[Le Concombre Masqué]

On 6/9/2018 at 8:51 AM, Le Concombre Masqué said:

 

 

Mind applied smoothing : a predicted response curve matching Bob Katz’s target (flat to 1 k, 6 dB downward to 20K ) created with “Psyschoacoustic” smoothing in REW resembles RR1 when looked at in VAR or traditional (ie 1/3 octave) smoothing mode (Bob Katz uses Acourate’s Psychoacoustic)

 

 

Bob Katz's target#Harman RR1as you can see ... Actually, I created in VAR smoothing targeting RR1

[Le Concombre Masqué]

"Image", I forgot to mention "image", soundstage, disturbing soundstage when I get the illusion of a large venue (ie with Solti's Ring or Dorati's Firebird) in my too small a room for my large Cabasse : distance to walls etc are just means, what matters is response curve.

 

We know, we know, but hearing what I now hear at last, far better than what I got in the much  larger room I had before, but with no eQ, is shocking 

[Le Concombre Masqué]

Actually, I could and I did better. But it's been long and painful : the infamous last mile... or whatever analogy...

 

I used Rephase in conjonction with REW.

 

Goal = Time correction

 

If I was to make it a business, provided it's possible with the generous great softwares REW and RePhase, I would not do it for less than 150 €, being provided measurements done by my customers : if really Acourate does as well (but how to do the magic trick Vector Average + MMM combo ????) , that you don't have the necessity or taken the challenge of greatly improving your system for free, buy Acourate. It took me hours and hours to get things done as described thereafter while it's supposed to be done in minutes with commercial softwares.

 

A quick sketching :

 

a tuto by Swissbear will teach you how to get wonderful looking curves based on REW' vector average of multiple measurements

 

problem is : VA does not yield satisfying amplitude corrections ; MMM (moving micro method) do

 

went through many contradictory informations

 

crazy thing is that the simple thing I settled for hasn't been described yet 

 

go Swissbear till you get gorgeous looking soso sounding (IMO) results

 

go back to the . rephase setting results you have saved

 

blank amplitude parametric eQ banks

 

via tools apply the xml generated in REW that is the eQ filter done for the MMM measurement (s ; really reproductible) 

rough sketch it, I used a simple target based on B&K C (but -9dB not _13 since my room is not a large theatre + it's then very very close to the RR1 I advocated earlier, just neater simpler) +I used Q= magic 1 only

 

do light corrections to the phase eQ 

 

clear the measurement, that is Vector average

 

import the MMM measurement

 

tweak the amplitude eQ (works much better in pre phase eQed Rephase than in REW), forget the phase eQ ; MMM has no phase information, just leave the few diagonal bars as they are. 

 

generate impulse per Swissbear and al

 

You have then generated the best possible phase eQ, for it's based on the Vector Averaging of multiple measurements, and associated it with the best amplitude eQ for it's based on MMM measurement(s)

 

Umik measured PN is on target while Impulse and phase proofing might not look as gorgeous as in Swissbear and al exemples but are pretty decent

 

 

 

 

[sb6]

Did you notice any degradation in sound quality using such s significant amount of DEQ?

[Le Concombre Masqué]

39 minutes ago, sb6 said:

Did you notice any degradation in sound quality using such s significant amount of DEQ?

No

[sb6]

That's interesting because when I use DEQ via REW implemented via Convolver in HQPlayer I hear some high frequency attenuation / rolloff.

[Le Concombre Masqué]

Well, you have my actual final measurements attached ; the Pink noise is played by HQP convolving  and measured via a second computer : no rolloff but intended downward slope adapted from Bruel and Kjaer, very close to Harman's RR1 and even Bob Katz's taking "psychoacoustic" into account

 

The SQ is pretty amazing indeed : spatialisation, clarity, speed.... all has improved (over previous quite worked on filters) I found the process complicated but I'm highly rewarded

[sb6]

Thanks for the information. I have used REW to provide DEQ for bass and looked into Phase EQ but didn't dig into how to implement it. Phase seems to be more of the challenge to implement and do so beneficially IMO.

[Le Concombre Masqué]

I think you can get a musical file eQed by Acourate (and/or Dirac) and hear for yourself the difference. Should be a real temptation to buy.

[sb6]

Thanks. I understand and have implemented DEQ via REW for frequency, all I need is to better understand phase and I'm set. Problem is - there is very little information and no real manual to assist. Eventually I'll spend the time and figure it out like you did ?

[Bob Stern]

2 hours ago, sb6 said:

there is very little information and no real manual to assist.

 

There is copious information from the developer, but the actual usage instructions are scattered among mathematical explanations of the theory.  Chapters 33–35 explain how to generate an EQ filter:

http://www.roomeqwizard.com/help.html

 

A few members of the AVS Forum (now called AV Nirvana) created a 130 page tutorial.  The chapter "Using REW to Generate EQ Filters" begins at page 95:

https://www.avnirvana.com/resources/getting-started-with-rew-a-step-by-step-guide.19/

[sb6]

Thanks Bob, just requested an account for avnirvana to dload the manual. For REW manual, I have read it, but there's little info specific to generating an EQ filter changing phase. I have many for frequency which work, but again knowledge of how / what to adjust for phase is the challenge.

[Le Concombre Masqué]

6 hours ago, sb6 said:

Thanks. I understand and have implemented DEQ via REW for frequency, all I need is to better understand phase and I'm set. Problem is - there is very little information and no real manual to assist. Eventually I'll spend the time and figure it out like you did ?

https://www.dropbox.com/s/10xdhh83jokzbxv/REW_rePhase_tuto.pdf?dl=0

 

once done 

go back to the . rephase setting results you have saved

 

blank amplitude parametric eQ banks

 

via tools apply the xml generated in REW that is the eQ filter done for the MMM measurement (s ; really reproductible) 

rough sketch it, I used a simple target based on B&K C (but -9dB not _13 since my room is not a large theatre + it's then very very close to the RR1 I advocated earlier, just neater simpler) +I used Q= magic 1 only

 

do light corrections to the phase eQ 

 

clear the measurement, that is Vector average

 

import the MMM measurement

 

tweak the amplitude eQ (works much better in pre phase eQed Rephase than in REW), forget the phase eQ ; MMM has no phase information, just leave the few diagonal bars as they are. 

 

generate impulse per Swissbear and al

 

 

http://forum-hifi.fr/thread-2003.html I have not read it but wish I did, looks thorough...

 

+ 

[Le Concombre Masqué]

The measurements of actual concert halls presented here https://pdfs.semanticscholar.org/e4c9/b3222f26472f6ba12e7fea900c2edb10c14c.pdf show a much more drastic drop of the frequencies above 5K than suggested by target curves such as B&K C, Bob Katz's or HK RR1 

 

The bass roll-off is consistent with the used monitor FR while the drastic treble roll off appears to be the hall response.

 

I created new convolution files to mimic SI R2, preferred by assessors cf https://users.aalto.fi/~ktlokki/Publs/JASA_lokki2012.pdf

 

Although I can't say I suffered from brightness with my previous filters, the first impression is very good and paradoxal : more precision in instruments placement (while I had in mind that more treble favours soundstage) 

[Bob Stern]

4 hours ago, Le Concombre Masqué said:

Although I can't say I suffered from brightness with my previous filters, the first impression is very good and paradoxal : more precision in instruments placement (while I had in mind that more treble favours soundstage) 

 

Thanks, Cucumber, for those interesting articles plus your own findings.

 

Your impressions of your new filter are consistent with my impression that frequencies above 6 KHz produce airiness and delicacy, not brightness, whereas brightness is produced by frequencies in the range of 3 to 5 KHz.

[Le Concombre Masqué]

53 minutes ago, Bob Stern said:

 

Thanks, Cucumber, for those interesting articles plus your own findings.

 

Your impressions of your new filter are consistent with my impression that frequencies above 6 KHz produce airiness and delicacy, not brightness, whereas brightness is produced by frequencies in the range of 3 to 5 KHz.

Agreed. And/but live gigs don't show off as much airiness and "soundstage" as some high end systems do

[Le Concombre Masqué]

end of year 2019 UPDATE:

 

1) it s more cumbersome but more rewarding to eQ the Vector Average than the MMM response. It s not simple for the curve in RePhase is ugly being the VA with MMM FR adjustments (one needs to convolve VA with designed for MMM filters, measure MMC, correct in REW, export in Rephase a few iterations to get results below (impulse and Step look just as good as in my signature but are better because being those of the actual corrected response)

 

2)It s quite difficult to build differentiated B&K and Harman targeting corrections for 3 dB is close to the margin error (see below actual measurements of my system/room) ; yet they sound very different but not in a trivial way (that would be more bass etc)

 

3) There s rational for the optimum target as found by Sean Olive (cf Harman) being not the right one, often.

 

Translation (read attached Sonarworks white paper) conscious engineers who have chosen to output their work following B&K have maximised the chances of a right translation, see confirmation by Sonarworks (attached picture of average of many systems in many rooms). Then when listening to a consciously made record chances are it fits BK better for it was a safe bet for the engineer. 

 

4)I would advocate that Olive s target would be adopted by both engineers and audiophiles rather than the solution promoted by Sonarworks for the simple reason that a downward slope is natural and that unnatural makers/users standardisation via x curve has already proved wrong in the movie industry 

 

 

whitepaper.pdf

[Le Concombre Masqué]

3 weeks later :

 

 

Parametric eQ to scalar NON FDW average of at least 10 sweep measures 

 

Should match MMM measure over 300 Hz, check

 

eQ phase to FDW Vector Average

 

look for tweeter peak in Impulse ; say it’s -37 ùs, set t offset in REW control before export, in Rephase, set phase offset 37

 

don t eq phase  below 100 lightly till 300 or so

 

do iterations from rew to rephase at 65536/96, no more, for faster results

 

keep phase eQ, building RR1 on B&K, then Synthesis

 

use pano shuffler presets in rephase

 

enjoy 3 targets 

 

 

[Le Concombre Masqué]

ALTERNATE perspective. 

[luisma]

Thank you @Le Concombre Masquéfor taking the time to put your experience in this thread, doubt the girlfriend is still around after all this, I will be implementing something similar and your journey is very helpful, I'm surprised you don't get more comments on this.

So currently you're using REW and RePhase exclusively correct? You mentioned pink noise by HQP, I was using REW to generate pink noise from a laptop fed directly to the DAC (no HQP on this phase)

 

[Le Concombre Masqué]

On 4/24/2020 at 6:15 PM, luisma said:

Thank you @Le Concombre Masquéfor taking the time to put your experience in this thread, doubt the girlfriend is still around after all this, I will be implementing something similar and your journey is very helpful, I'm surprised you don't get more comments on this.

So currently you're using REW and RePhase exclusively correct? You mentioned pink noise by HQP, I was using REW to generate pink noise from a laptop fed directly to the DAC (no HQP on this phase)

 

Thank you @luisma, ask @The Computer Audiophile to reload my latest findings he deleted.

In short : REW + RePhase, yes.

Vector Average from REW should be eQed to an inverse S 10 dB Standard Harman Curve, not the usual suspects RR1, B&K or what have you, that apply to steady state measurements to which a JBL Synthesis target might be applied 

[Moedra]

This is a most interesting thread. Over the last year or so I have been delving through the scattered information on this topic just as I'm sure many of you have. The information here is indeed compelling. I have been working on this very subject myself with REW + rePhase and I believe I have worked out an elegant and effective solution to the matter. I'm interested in finding a few people to take my filters for a test drive. All that's required is a copy of the most recent version of REW and a measurement mic - and of course a convolution processor to run the filters. They are 64-bit wav files with 65536 taps. I make them for independent channels (I run parallel convolvers to process them) but the two files can be joined into a stereo 32-bit file for applications like Roon.

I've attached a screenshot of the systemwide configuration I use for PC-based playback. The host is an application called Pedalboard, and I use "Satson Channel" to control the trim, two instances of Melda's MConvolutionEZ to process the left and right filters, and VUMT at the end of the chain to control the output level and verify that there is no signal clipping.

 

 

If anyone would like to compare notes, I'm up for that as well.