Share your in-room measurements?

[Chromatischism]

Yeah, it's unnecessary to measure that loud. Typically I measure at 85 dB, which should be enough. Try it and see. If you have an AVR that's calibrated to reference, you should get there at the -20 MV setting.

You're right in pointing out what I'm seeing, so thanks for explaining that you're going for the Harman curve. It certainly does cause your upper bass to look depressed. Down in the subwoofer region, my own testing over the years tells me I wouldn't like the blue line as much, and would prefer your measurement, where the bottom end rises above the mid bass. It gives everything a nice weight that anchors the sound. Therefore my target would bring 60-100 Hz downward a bit and merge those two areas better. There is probably some personal taste here. I find when that region is elevated, the upper bass gets a little cloudy, including vocals. So for me it's a clarity improvement.

Is the 350-1000 Hz rise seen from your speaker in anechoic measurements?

 

[richard12511]

Yeah, it's unnecessary to measure that loud. Typically I measure at 85 dB, which should be enough. Try it and see. If you have an AVR that's calibrated to reference, you should get there at the -20 MV setting.

You're right in pointing out what I'm seeing, so thanks for explaining that you're going for the Harman curve. It certainly does cause your upper bass to look depressed. Down in the subwoofer region, my own testing over the years tells me I wouldn't like the blue line as much, and would prefer your measurement, where the bottom end rises above the mid bass. It gives everything a nice weight that anchors the sound. Therefore my target would bring 60-100 Hz downward a bit and merge those two areas better. There is probably some personal taste here. I find when that region is elevated, the upper bass gets a little cloudy, including vocals. So for me it's a clarity improvement.

Is the 350-1000 Hz rise seen from your speaker in anechoic measurements?

No rise in the anechoic measurements that I've seen, so it's gotta be a room/placement issue. I'll try your EQ suggestion. The box I'm using for Dirac lets me have 4 different presets that I can switch between at any time.

 

[nick-v]

As measured from your desk chair?

Yes, that's a single measurement from the mlp at ear level. I had the Umik1 on a boom stand at ear level with the chair where it normally sits.

 

[thewas]

Ok, I re ran measurements this morning and I formatted the graph to your specifications, here. Var smoothing, 15-20,000, 5 degree increments and 50dB scale, 1920 width.

View attachment 87342

The problem still shows(peaking from 330-1,500Hz) , though I guess it's a little less obvious than it was at 1/12 smoothing. I suppose I could keep my EQ only under 300Hz strategy and still fix the problem by bumping up everything below 300Hz by 2-3dB? I'm not overly concerned with headroom.

Target curves depend on the room reverberation and loudspeaker directivity characteristics, so I don't see a peaking problem between 330-1500 Hz but rather the problem of using a predefined target which was made using different loudspeakers and rooms.
In most rooms I have measured till now good conventional loudspeaker with smooth increasing directivity don't show the "ski slope" Harman curve you use but rather a linearly decreasing curve with about 0.8-1.0 dB per octave and that is what was also @mitchco confirms in his reviews and actually that is what listeners preferred at Harman, see slide 24 http://petoindominique.fr/pdf/The Subjective and Objective Evaluation of Room Correction Products.pdf

You curve actually is quite close to such a 1.0 dB per octave target

except a bit too few energy in the 100-300 Hz region which is a very commonly observed and usually results from various SBIRs. You could try to increase that region about 2-3 dB and compare it directly if you prefer its warmer ground tones or the more analytic ones you have now but I wouldn't really "correct" anything else.

 

[richard12511]

Target curves depend on the room reverberation and loudspeaker directivity characteristics, so I don't see a peaking problem between 330-1500 Hz but rather the problem of using a predefined target which was made using different loudspeakers and rooms.
In most rooms I have measured till now good conventional loudspeaker with smooth increasing directivity don't show the "ski slope" Harman curve you use but rather a linearly decreasing curve with about 0.8-1.0 dB per octave and that is what was also @mitchco confirms in his reviews and actually that is what listeners preferred at Harman, see slide 24 http://petoindominique.fr/pdf/The Subjective and Objective Evaluation of Room Correction Products.pdf

You curve actually is quite close to such a 1.0 dB per octave target

View attachment 87488

except a bit too few energy in the 100-300 Hz region which is a very commonly observed and usually results from various SBIRs. You could try to increase that region about 2-3 dB and compare it directly if you prefer its warmer ground tones or the more analytic ones you have now but I wouldn't really "correct" anything else.

Interesting. I'll try a more linear slope and see if I like that better. I know I like the "ski slope"(like that name) curve more than the default Dirac curve(linear .5dB per octave slope), but I've never directly compared the ski slope curve against a linear 1.0dB per octave curve.

 

[Chromatischism]

Depending on the room, sub, and speakers I usually like ~5-10 dB from 20 Hz to 100 Hz, then another ~5-10 dB from about 100 Hz to 20 kHz. The important thing is that for me I tend to think of those two ranges separately, and I guess Harman does too.

 

[ernestcarl]

I find it quite confusing how REW suggests different "default" image widths. It would be nice if there were a drop down list for only the most common standard widths.

 

[ernestcarl]

The reason I chose 1920 is for future proofing. Ever looked back at old images in dismay at how tiny they are on your screen because you used a lower resolution back then? So I'm attempting to avoid that. I'll get the laptop out and take a screen shot of my settings in a minute.

My gripe with 1920 is the uber small font size (barely readable if viewing on a small device):

2dB divisions is also too much.

I think a 1280 pixel width setting might be a good average even on a bigger higher res screen.

Taller than what I would want, but very readable on a small device -- and also not too pixelated on a much bigger screen.

 

[Chromatischism]

Yeah, I wish the text scaled up with the image size. Maybe I'll put that in as a feature request.

I still get 5 dB increments though. Part of the problem might be that we're using different limits (I use 45-105).

I think 1280 will be too small when you try to look back on your measurements in the future with a 4k or 5k screen.

 

[ernestcarl]

Yeah, I wish the text scaled up with the image size. Maybe I'll put that in as a feature request.

I still get 5 dB increments though. Part of the problem might be that we're using different limits (I use 45-105).

I think 1280 will be too small when you try to look back on your measurements in the future with a 4k or 5k screen.

The setting to change the graph text size was staring at us in the face all along!

1920pw 25dB/decade

I presume using HTML code one can resize the descriptive title as well.

1920pw 50dB/decade

above is compressed but saves screen space quite a bit

Evidently, simply changing the captured graph font size brought back the 5dB per division vertical scaling for some reason (for the first graph!).

*Personally, if one were purely aiming for standardizing REW graphs and capture settings specifically within this thread, my vote is for the first one (25dB/decade) simply because it keeps the 5dB per division lines intact and it doesn't seem to matter what the resolution width is set e.g. 800, 1280, 1920. The second one would be perfectly okay too if it just did the same. I dunno if the differences we're seeing is due to one's OS or program/version settings...

 

[Chromatischism]

Nice, that must be a new feature. A benefit of using higher resolution (like 1920 for example) is we get to see more detail. I also increase the trace thickness at higher resolutions to 3 or 4 (depends if it's an averaged measurement, which is already thicker).

 

[ernestcarl]

Nice, that must be a new feature. A benefit of using higher resolution (like 1920 for example) is we get to see more detail. I also increase the trace thickness at higher resolutions to 3 or 4 (depends if it's an averaged measurement, which is already thicker).

I was mistaken about the 50dB/decade graph being not as good as the 25dB/decade due to the lines... as mentioned, experiment by changing the capture graph font size and you can get the 5dB division lines right back where they should be. I find this to be a rather strange quirk with the program -- as it would have been better if there were a bulleted option to choose from e.g. 2dB/5dB/10dB main division lines for the vertical scaling within the capture graph settings.

But yeah, if more resolution is what's desired, then it's best to just go with 1920pw and 25dB/decade (& adjust fonts accordingly) by default for a permanently fixed aspect ratio that makes it super easy to compare between different graphs.

*Although, more often than not (esp. if not using MMM or spatial averaging), measurements may not be too linear (when placement and room acoustics is not at its optimum) -- so I think an 80-100dB window, 50dB/decade, and applying 1/6 or 1/12 or psyc or var smoothing makes it much easier to see the overall 'big picture' -- this also makes more sense if one wants to see more 'detail' in a distortion graph.

That deep, persistent and quite audible cancellation between 400-500Hz is from desk bounce -- it goes away when the desk is removed with only speakers on stands left in place.

I had to move furniture around quite significantly (pulling my desk further away from all walls and sacrificing what little room space I have) to achieve as linear a response as possible right from the start -- even before EQ. It turns out the sacrifice was worth-it, and the only somewhat consequential EQ boost I needed(?) to apply was around the 400-500Hz null area. A sub proved crucial in this case as you lose some bass as you move your monitors far away from all walls (i.e. removing 'boundary gain').



Let me just re-emphasize for everyone else might be just starting to learn to EQ:

OPTIMIZE PLACEMENT and ROOM ACOUSTICS FIRST BEFORE EQUALIZATION!

Will save you a lot of time and minimize your need to perform innumerable post 'tweaking' adjustments later on.

For context (and maybe as an update to my own desk setup),
Here's L+R before summing with sub (miniDSP output channels)

Not much EQ surgery necessary!

All that furniture moving made me too lazy to EQ each channel individually above 200Hz. Anyhow, I checked each channel invidually to verify -- and, while admittedly less than perfect as it's not done individually, it still looks fine overall despite then summed filter generation. May partly be due to me keeping boundaries and placement as optimally symmetrical and clear from obstructions as possible.

Post summing with sub (miniDSP input channels)

*Still a few manual adjustments were made to get things where I want...

*oh yeah, (esp. if you only have a single sub) optimize your phase alignment to get the best possible summing between all drivers! In this particular case, a high pass filter for the sub and low pass for speakers was not necessary -- and made phase response more linear and alignment between speakers easier.

Nearfield and a substantial distance away from walls.

The only significant reflection/cancellation I'd say now is the one between 400-500Hz, inevitably caused by the presence of a desk.

 

[hege]

But yeah, if more resolution is what's desired, then it's best to just go with 1920pw

Uh, FYI forum resizes all images to 1280px max, so your suggestion is quite pointless and just reduces quality. IMO even that is complete overkill for graphs, something like 800-960px actually fits in thread post area as is. Honestly this whole thread is horrible to read. The graphs should be manually resized to like half the post area width, so people can click them bigger if needed, but this forum doesn't seem to support the feature.

 

[RayDunzl]

forum resizes all images to 1280px max

But if you click them they'll open full size, up to the byte-size storage limit you can save here.

 

[ernestcarl]

Nice, that must be a new feature. A benefit of using higher resolution (like 1920 for example) is we get to see more detail. I also increase the trace thickness at higher resolutions to 3 or 4 (depends if it's an averaged measurement, which is already thicker).

I haven't experimented much with changing trace line thickness settings, but definitely may come in handy to clarify things even further.

 

[hege]

But if you click them they'll open full size, up to the byte-size limit you can save here.

My mistake. Accidentally tried image which was originally 1280px. Duh. Well my point still stands, most people probably watch this forum on max 24" screens and phones. Those 1920px images above are terrible screen hogs (+ resized mess). @amirm should install some plugin which allows users setting the image display size inside post.

 

[ernestcarl]

Uh, FYI forum resizes all images to 1280px max, so your suggestion is quite pointless and just reduces quality. IMO even that is complete overkill for graphs, something like 800-960px actually fits in thread post area as is. Honestly this whole thread is horrible to read. The graphs should be manually resized to like half the post area width, so people can click them bigger if needed, but this forum doesn't seem to support the feature.

Interesting. I had no idea that it did that. Well, my original suggestion was 1280. While I do appreciate your suggestion, I'm not sure why you are saying the thread is a horrible read. This was specifically for graphs generated using REW's 'capture' option. I check the thread from time to time just out of interest of people's measurements, and don't see the need to treat its entirety as a continuous, singular piece.

 

[hege]

I'm not sure why you are saying the thread is a horrible read. This was specifically for graphs generated using REW's 'capture' option.

Maybe a small exaggeration. Most images are decend here. Mostly a large height eating the whole screen bothers me. There's a point where there simply isn't more information in an image, no matter how large you make it. Posting 800-960px original images makes them non-resized and sharp in posts.

 

[ernestcarl]

Yeah, actually, I'd be happy if people just substantially increased the fonts and limit the vertical window to a max of 100dB (although Ideally half that or 50dB) -- high resolution is not absolutely necessary since we're not exactly producing archival-level spinorama data here.

 

[Chromatischism]

Guys, if there are issues with images taking up too much of the screen, we could adopt a rule by committee to display them like this instead:

(Use the Insert: Thumbnail button)



One issue this creates though is scrolling up and referring to someone's graph while writing a response, it would be too small to make out anything. Maybe we could work around it.