What is your favorite house curve

[CREMA]

Extremely low frequency. MORE, MORE!

 

[txbdan]

I have a pair of 8030Cs and Dirac Live in my office/studio setup and have had them set to a Olive-Toole (as described in this thread for better or worse?) for months. The other day I thought to myself that It was lacking a bit of "sparkle" so I clicked over to my B&K preset and found it opened things up. Then I clicked over to my 3.5db tilt about 1kHz. And then to the stock Dirac curve with is a 2.5db tilt about 1Khz. I've been digging the 3.5db tilt best. Compared to the O-T, it has a little more kick and at the same time more clarity and treble. Music is more fun. I reserve the right to chance my mind in six more months.

 

[GalZohar]

If we measure direct+reflected sound but want to correct only the direct sound, then isn't that an effort doomed to fail no matter what target curve is used? Doesn't this just mean that the advice to limit the EQ to low frequencies where we hear the summed direct+reflection sounds which is what we can actually measure and try to compensate for? And even if we do hear the high frequency reflections as some kind of distortion, it seems like EQ can't really help in that department anyway.

I want to finish reading the rest of the thread, but for now it seems like the conclusion is to just "give up" on EQ of the higher frequencies, at least for EQ systems that can't separate the direct sound from reflected sound...

Am I missing anything?

 

[txbdan]

I apply correction full band. Sounds better to me that way. I am in a near field situation though so reflections are less dominant.

 

[Sal1950]

I want to finish reading the rest of the thread, but for now it seems like the conclusion is to just "give up" on EQ of the higher frequencies, at least for EQ systems that can't separate the direct sound from reflected sound...

I apply correction full band. Sounds better to me that way. I am in a near field situation though so reflections are less dominant.

You'll find a lot of opinions on the subject, much of which gets based on person preference and that's fine too.
MHO on the subject is that the best practice is to use DRC to smooth out the bass range along with subwoofer placement, using multipul subs in alternate locations if possible. Room treatments like bass traps, etc are great but it's extremely difficult to obtain desired results since room modes tend to be large and very hard to tame.
Then use some room treatments to tame the worst of the reflections in the mid-top end, that's where the treatments can be most effective.
All things being equal, room treatments would be the best approach everywhere but as I already mentioned it's very difficult to do in the bass range. We'd like to minimize the rooms effect on the sound whenever possible, listening mainly to what's on the recording as presented by the speakers. Some like to use room reflections as a tool to artificially increase soundstage size but IMO that is simply a distortion of the source. Similar changes have been offered by software built into some hardware from Bob Carvers Sonic Hologram preamps, to more modern use of DRC. Again these are simply distortions of the source and not the best way to approach this effect. If you really want to expand your soundstage, go to a multich system using 5.1 or better source recordings, or possibly multich upsampling of stereo source to multich playback but discreet multich recordings are vastly preferable..

Unfortunately real life has a habit of getting in the way of the best options, causing us all to make some compromises.
Don't let audiophellia spoil your enjoyment of the music, grinding your brain over how it could be made better.
Just let the music flow to your soul.

 

[dasdoing]

I am at a point where I think that there is just something inherently wrong by looking at smoothed curves. our ears don't average like that. in the speach recognition world they rather look at envelope curves. In my testing flatish envelopes on the unsmoothed graph result in smoothed graphs reminding of house curves. Im am pretty sure that the true answer is in that direction. with that beeing said, a true neutral response is not always wanted, but when you arive at flatish envelope you can than add a bass bosst f.e.

 

[ernestcarl]

I am at a point where I think that there is just something inherently wrong by looking at smoothed curves. our ears don't average like that. in the speach recognition world they rather look at envelope curves. In my testing flatish envelopes on the unsmoothed graph result in smoothed graphs reminding of house curves. Im am pretty sure that the true answer is in that direction. with that beeing said, a true neutral response is not always wanted, but when you arive at flatish envelope you can than add a bass bosst f.e.

How do you preview and analyse the envelope ideally in something like REW?

There was a thread (or two) that discussed how to read spectral decay and ETC types of graphs... even there smoothing could certainly benefit "readability" of the graphs. I'd rather look at the raw data file so I can change the smoothing as I want... but you end up with too large of a file size to attach here -- static images are more convenient in that regard.

 

[dasdoing]

How do you preview and analyse the envelope ideally in something like REW?

I asked in REW forum for a feature like this. DEV said he wont do it cause it's Acourate "proprietary". Acourate has the function but it will only bring the peaks down: https://www.audiovero.de/acourate-wiki/doku.php?id=en:wiki:funktionen:td-functions:psychoacoustics

however: it is hard to see in REW, but you can do it.
My process is quite complicated atm, and I am still experimenting, but here is a graph for a 1 position correction (the fall after 17k is an error in convolution of the meassurement; it's not real):

as you can see there are no significant peaks above that curser line.
but it still has a lot of what I call "holes" (if those make a diference is another question, just showning how I read it):

these ranges could still be EQed up to the line for example. garanted that in this case those are 1dB-ish and it probably doesn't make sense. more so because at the end it has to be the flatest it can be not for one, but for several meassurements in the LP (that is the part I still have to experiment)

btw: the same meassurement var smoothed:

it seams wrong that it goes a little up again after 9k,
but I think it makes sense, since my room seams to be more anechoic up there

 

[ernestcarl]

My process is quite complicated atm, and I am still experimenting, but here is a graph for a 1 position correction (the fall after 17k is an error in convolution of the meassurement; it's not real):

Oh, I thought it was the envelope time curve as viewed in the ETC tab you were talking about. You actually mean the upper envelope in the magnitude trace. So this type of filtering is not really something that REW is able to show easily with the same kind of flexibility. I see.

Looking at your graphs, I still find this difficult to interpret where the actual audible benefits of such micro corrections are (FYI: I'm not totally dead-set against it). It's hard enough to ascertain with absolute certainty whether something sounds truly better afterwards using listening tests alone -- program material variances and one's general personal mood at the time can also influence one's opinion of the amount of adjustment needed even when only some very simple shelving type correction filters used in the bass and treble.

I guess if you are uncertain whether an average of multiple measurements is better, then there's nothing wrong with taking more than one approach -- comparing between the two (or more) to find out whichever works best for you in that regard.

I think I prefer the quick and dirty MMM/pink noise approach in addition to some very "brute force" (i.e. averaging of many, many spatially spread-out measurements across the listening position/s) myself. The latter is annoyingly tedious, though.

However, I do think that even a single measurement can already reveal a lot.

Below is from a single L+R sine swept measurement at the main listening position (~2m+ distance from the Sceptre S8 mains with a single bass-managed sub) of my couch.

80 dB vertical scale

Sceptre S8 monitors are angled so they are on-axis to the MLP for this test.

Decay tab's view options are all set to their default apart from the amount of smoothing and rise time.

*could be wrong, but the mic calibration file seems to reflect improperly in this tab view.

Let's focus in the HF for now...

I am currently only using four linear phase parametric EQs (one is a shelving filter) as my generalized tonal balance correction to these speakers' HF response.

My quasi anechoic off-axis measurements -- as well as Resolution Magazine's -- show a deep notch growing at 3.5 kHz as one progressively moves well beyond 30 degrees. Initially, the on-axis response shows this area as quite relatively flat but the decay says otherwise. Additionally, above 10 kHz the spectrum energy traces in time decay very fast which I believe maybe partly why I find the Sceptre's less fatiguing than the Neumann KH120s.

The on- and off-axis curves of the S8 are more uneven and complex in comparison to the KH120 and so EQ in the HF is not nearly as intuitively straightforward. I have to to see the individual on- and off-axis response as well a more generalized average LW -- so for these monitors, I do think one needs to perform many, many measurements.

I could equalize this speaker to within +- 1dB, for sure, but at what optimum angle since the response gain will be enhanced or cut low depending on which trace angle one looks at... and the overall timbre (which I already am quite satisfied with) would surely be significantly altered from its default.

Hmmmn... As a matter of my own subjective "psychoacoustic" smoothing, since I'm already satisfied with the quality of the sound, there is no need to equalize the system's response any more than what's minimally necessary.

Personally, I would like to see more measurements looking at the spectral decay in other people's listening rooms in detail -- I think that this can explain how people might prefer certain "house curves" more than others -- but, unfortunately, I've noticed focus is very much frequency response magnitude measurements mostly -- even if directivity performance is emphasized, too, at times.

 

[Snarfie]

Not a Curve but a flat neutral White Line. Compared to the Grey found frequency Curve it creates a level playing field and sounds amazing.

https://imgur.com/T0I1hrT

 

[ernestcarl]

Not a Curve but a flat neutral White Line. Compared to the Grey found frequency Curve it creates a level playing field and sounds amazing.

https://imgur.com/T0I1hrT

Why is the measured FR SPL offset above 200Hz so huge?

 

[ernestcarl]

I guess if you are uncertain whether an average of multiple measurements is better, then there's nothing wrong with taking more than one approach -- comparing between the two (or more) to find out whichever works best for you in that regard.

More of the same setup measured but using multiple sweeps instead:

100 dB vertical scale

I suspect spectral decay dips and uneveness here and there make the "timbre" of this speaker unique compared to more "flat" or neutral sounding and/or measuring monitors like the Neumann KH120.

Scanned from Resolution Magazine review:

Simple linear phase parametric EQs using FIR convolution in blue trace

*I believe my microphone's 90 degree calibration file is not so accurate in the HF above 5kHz -- but it doesn't matter since I adjust the overall amount of shelving based on listening -- no, I don't think I'm really attempting very hard to make the "house curve" line ruler flat in the HF here -- though, it may incidentally look like it.

Measurements averaged across the entire width of the couch actually produce a flatter, sloping "power response" between 100 Hz to 1kHz -- so I do not to apply too many corrections between that area to make the response "extra linear" more than is necessary.

Additionally, I keep a separate set of quickly adjustable IIR filters inside JRiver to reduce the number of FIR taps required (2048 ~21ms delay at 48 kHz) to equalize the bass between the sub and satellite MCH outputs used.

 

[Snarfie]

Why is the measured FR SPL offset above 200Hz so huge?

Because of realy bad room acoustics. If i switch to bypass mode it will hurt your ears that bad.

 

[ernestcarl]

Because of realy bad room acoustics. If i switch to bypass mode it will hurt your ears that bad.

I see... the plateau offset in level does seem rather oddly extreme.

 

[Flak]

If we measure direct+reflected sound but want to correct only the direct sound, then isn't that an effort doomed to fail no matter what target curve is used? Doesn't this just mean that the advice to limit the EQ to low frequencies where we hear the summed direct+reflection sounds which is what we can actually measure and try to compensate for? And even if we do hear the high frequency reflections as some kind of distortion, it seems like EQ can't really help in that department anyway.

I want to finish reading the rest of the thread, but for now it seems like the conclusion is to just "give up" on EQ of the higher frequencies, at least for EQ systems that can't separate the direct sound from reflected sound...

Am I missing anything?

Dirac Live will leave the late reflections untouched as appropriate and the algorithm can accomplish that by taking advantage of the multiple measurements... those late reflections are late because they are bounced multiple times by the walls so that the response is position dependant

 

[ernestcarl]

Simple linear phase parametric EQs using FIR convolution in blue trace

Looking at this again, I realize that I had accidentally disabled my 300 Hz -3dB PEQ in JRiver. Peak should have been reduced as it also is present outside the range of my MLP.

 

[dasdoing]

Looking at your graphs, I still find this difficult to interpret where the actual audible benefits of such micro corrections are

as I said, this is an allready corected graph, and I only pointed out the "imperfections" to show how I read the unsmoothed graph

It's hard enough to ascertain with absolute certainty whether something sounds truly better afterwards using listening tests alone -- program material variances and one's general personal mood at the time can also influence one's opinion of the amount of adjustment needed even when only some very simple shelving type correction filters used in the bass and treble.

that's a real problem. iirc this is something that was discussed in another topic. you can't AB improvement in sound. you have to listen to it for hours.
Also I think that program material variances are best adressed by a neutral playback, cause if you take the mean of all variances they should be back to neutral. the fact that a flat envelope FR viewed in a smoothed graph comes close to resulting curves of trail and error experiments (Harmann) kind of evidences this.
all these years I have been wondering: neutral has to be flat. if flat doesn't sound neutral what is wrong is the method of the ploted graph. The first solution to this problem I found is, as you know, the flat direct sound. it doesn't realy sound satisfying though, though it is probably the best for mixing; you kind of hear the room seperated. Now I think that the envelope method give you a married (direct+room) neutrality.

if you are uncertain whether an average of multiple measurements is better

average of multiple measurements is not an option for envelope correction, since it will bring the peaks down. I am thinking of correction multiples in an overlay fashin instead. once again REW wont give me an option to make a single overlay meassurement out of multiples though.

the envelope method is probably undoable over a whole couch. and I never understood why you would do this. out of the sweet spot the sound is crap anyways. all you do is sacrificing the sound in the sweet spot for a small improvement outside of it and is still crap.
but sure multiples are still necessary if correcting for sweet spot only, but they will be much closer to each other

(I will answer the rest of you post another hour)

 

[ernestcarl]

the envelope method is probably undoable over a whole couch. and I never understood why you would do this. out of the sweet spot the sound is crap anyways. all you do is sacrificing the sound in the sweet spot for a small improvement outside of it and is still crap.
but sure multiples are still necessary if correcting for sweet spot only, but they will be much closer to each other

The measurements I posted before actually tried to equalize based on the spatial average results restricted around the MLP. Effort was made to limit PEQs used just for the sake of principle. Now, if I put more weight to the seating beyond the MLP of the couch (up to the very edges quite close to the room side-walls), the MLP would look even more "crappy", as you say.

In balance, though, I'm finding the overall compromise here rather insignificant:

single point measurement

A little more than half of the weight was placed in the center MLP -- either left and right sides of the couch were allocated less than a quarter.

But then this really depends a lot in the acoustics of the room and positioning as some spaces out there will vary much more widely than others.

Neutral to me in a room is generally a flattish response with a downwards slope. The shape of the curve is largely dictated by the speakers and room acoustics, among a few other things. Instead of forcing a completely "flat" curve in-room, I'd rather adjust the curve a little bit closer to fit the natural overall trend of the measured response -- after finding the best placement spot in the room, of course.

For example, in the exact same room, but different location: I can't really elevate the sub bass much around my desk the same way in my couch listening space without incurring a disproportionate increase in the amount of ringing and distortion (will run out of headroom at loud volumes). One would not understand this if not actively looking at other metrics like spectral decay and distortion info. And a "cleaner" sounding sub (ultra low distortion), for example, can be played much more loudly without feeling any disturbance like from a really cheap sub box that just adds "extraneous noises" with increasing volume. Speaking of which, very loud playback also necessitates a sharper downward slope in the HF.

you can't AB improvement in sound. you have to listen to it for hours.

Really? Is what Amir doing, toggling between simple PEQs during his subjective listening tests an invalid exercise, then? I think a bit more qualification needs to be added to complete that statement.

In another thread with a youtube video review of the Adam T8V reinforced with concrete, we could hear improvement in sound by A/Bing the same bass-heavy clip in a matter of seconds! Even you remarked that you could hear a difference.

But, anyway... I do get your point.

 

[Weeb Labs]

8dB slope for me.

 

[Puddingbuks]

8dB slope for me.
View attachment 159961

Is that 1db per octave?