Estimated In-Room Response - trendline & slope

[gsp1971]

Hi everyone.

I know the Estimated In-Room Response should be sloping downwards. But how much? Moreover, depending on how the trendline is drawn, slightly different conclusions could be reached about where the EIR is over the trendline (potential brightness) or under it (potential recess).

I have been experimenting with trendlines & slopes on 6 speakers reviewed in ASR. For each speaker, I played around with different trendlines (in Excel), and picked the best (the one with the highest R-Squared metric). The selected trendline has a slope. Is there a way to determine which is better?

I am posting graphs for the 3 best speakers (out of the 6), each having a trendline with an R-Squared > 0.90 (i.e. extremely good, with little deviation around the trend).
The two Revels, though both highly rated, have quite different slopes, with the M16 being significantly steeper.

Does this make a difference in eventual listener preference or not?

Thx in advance for whatever discussion might arise from this.

Focal Aria 906, trendline slope -0.0473

Revel M16, trendline slope -0.0605

Revel M105, trendline slope -0.0399

 

[DJBonoBobo]

I don´t remember how i found this, but look at this calculator: http://www.ohl.to/calculators/targetcurve.php

 

[Matias]

Most common is -6 dB from 100 Hz to 20 kHz which is 0.8 dB per octave.

 

[edechamps]

I played around with different trendlines (in Excel), and picked the best (the one with the highest R-Squared metric). The selected trendline has a slope. Is there a way to determine which is better?

I am posting graphs for the 3 best speakers (out of the 6), each having a trendline with an R-Squared > 0.90 (i.e. extremely good, with little deviation around the trend).

r² is not an appropriate metric for estimating the "deviation around the trend" when comparing curves with different slopes. It's a common misconception. Mathematically, r² will increase with the steepness of the slope, which is not what you want here. Basically, the speakers that have the highest PIR least-squares regression r² (also known as SM_PIR in the Olive preference score papers) are the speakers that have the least deviation around the trend, but also have the steepest slope. This derives directly from the definition of r² and is a common pitfall that has already been debated to death (this example is one of the most telling). A better metric to use could be RMS of residuals, in dB.

As for what makes the best PIR slope, I guess that's still open for debate and might be related to directivity preferences. Here is a previous debate around this topic.

 

[abdo123]

The downward slope should be determined by your Speaker’s radiation pattern and DI index smoothness.

To assess Directivity preference you need to keep every other aspect in the speaker constant.

there is only one speaker brand that used to make Speakers with DSP controlled directivity index and that’s Lexicon.

If you manage to get one of their speakers and bring it to anyone with a listening room you can conduct a study and publish the results!

 

[richard12511]

The downward slope should be determined by your Speaker’s radiation pattern and DI index smoothness.

To assess Directivity preference you need to keep every other aspect in the speaker constant.

there is only one speaker brand that used to make Speakers with DSP controlled directivity index and that’s Lexicon.

If you manage to get one of their speakers and bring it to anyone with a listening room you can conduct a study and publish the results!

B&O does this too now.

 

[hardisj]

As for what makes the best PIR slope, I guess that's still open for debate and might be related to directivity preferences. Here is a previous debate around this topic.

Absolutely, this. Very wide speakers have a much more shallow slope. A fully omnidirectional speaker would have a flat line for directivity and a flat line PIR. Some people really enjoy omnidirectional speakers. Some prefer low directivity speakers (narrow radiation). And there is everything in between.

I asked Dr. Toole about this in my chat with him but I don't recall the timestamp. I believe it was near the end, if you care to watch it:

 

[MZKM]

From Olive’s speaker preference formula patent:

The ideal target slope for the on-axis and listening window curves is identical for both test samples, which indicates that the on-axis curve should be flat, while the off-axis curves should tilt gently downwards. The degree of tilt varies among curves for Test One and the larger sample. Test One includes mostly 2-way designs whereas the larger sample includes several 3-way and 4-way designs that tend to have wider dispersion (hence smaller negative target slopes) at mid and high frequencies. This suggests that the ideal target slope may depend on the loudspeaker’s directivity.

 

[sfdoddsy]

Every now and again I bang on about this. Might as well do it again.

The various sloping target curves we are all so fond of are not objective measurements of excellence.

They are responses to subjective preferences.

The fact that 74% of people prefer a -6dB slope from 100Hz (or whatever) does not mean such a slope is accurate. It just means it is liked.

I personally dislike the Harman Curve, but completely understand why so many enjoy it.

Much like many people like the skewed response of Bose or B&W speakers.

Likewise the fact that X% of listeners like the Harman curve does not mean we should design speakers or EQ to match the Harman curve unless we care more about subjective than objective quality.

Not to mention that even Floyd Toole doesn't suggest manipulating response. His hypothesis is that speakers which measure flat anechoically and have good dispersion tend to be preferred by trained listeners. They also measure with a downward sloop in room, but that is simply a result of physics.

So I think all this EIR stuff is pointless. The cool thing about the Klippel is that you can get anechoic and dispersion measurements in room.

Which gives you a Spinorama.

That's all you need.

 

[hardisj]

Not to mention that even Floyd Toole doesn't suggest manipulating response.

If you mean exactly what you said then that is incorrect. Toole stance on tone controls is they may be necessary and are perfectly reasonable to adjust to preference. Assuming you start with a known good product, first, however.

If I am misunderstanding you, please clarify.

 

[TimVG]

From Olive’s speaker preference formula patent:

Theoretically, in my mind, ideally the reflections are identical to the direct sound, minus several dB (I believe it's mentioned in Toole's book that preference for the first lateral reflection (sidewall) is minus 4 to 6dB). So constant directivity in the horizontal plane, from as low in frequency as possible. This however wouldn't score as well in the Olive metric.

 

[gsp1971]

Thanks everyone for the feedback.
I did some further reading based on the material you submitted.
It is true indeed that R-squared is not the appropriate metric to establish goodness-of-fit of a trendline, as it increases with slope. It seems Root Mean Square Error (RMSE, lower values are better) is a better metric. Based on RMSE, and the suggestion that the Estimated In Room Response line should not slope downwards more than 6-7dB overall from Upper Bass (100Hz or so) to High Mids (5kHz or so), the ranking of the 6 six speakers I had shortlisted now makes sense, as it correlates with their price (in Greece):

1. RMSE 0.57, Revel M105, €1690
2. RMSE 0.72, KEF LS50 Meta, €1250
3. RMSE 0.76, Focal Aria 906, €790
4. RMSE 0.91, KEF LS50, €860
5. RMSE 0.91, Revel M16, €1120
6. RMSE 1.25, ELAC DBR-62, €520

It seems the Focal Aria 906 is priced lower than its performance would suggest, as opposed to the Revel M16 which is priced a little high. If these two were inversed in the order, the correlation of RMSE to price would be almost perfect. See the pricing in your country, perhaps a bargain will identify itself.

I understand that there isn't a single perfect score to rank speakers - each method has pros and cons. One needs to study the Spinorama info to make an informed decision. Having said that, I wanted a simple but still logical / mathematical approach to rank speakers for an audition shortlist, and I am happy with the above.

Thanks again to all for the contributions and the feedback.

 

[alont]

I think that PIR is not the right way to rank speakers for an audition shortlist.
Instead, I would look for (by order of importance IMO):
1. LW flatness
2. ER smoothness (you can even look for a specific slope here if you know what kind of directivity you prefer).
3. Max clean output level @ 1m (I realize this is harder to quantify in an automated manner, so it should probably be calculated manually to further filter the shortlist)
4. Low frequency extension

 

[ADU]

These values are just estimates based on a quick and dirty visual analysis of the in-room spinorama curves. So your mileage may vary.

I haven't actually computed an average for these. But based on this list, I'm going to use -0.75 to -1.0 dB per octave as the approximate range for a neutral loudspeaker's estimated in-room slope for now.

SOME APPROX. IN-ROOM SLOPES IN DECIBELS SPL PER OCTAVE:

-0.70 SELAH INTEGRITY DIY
-0.90 DUTCH DUTCH 8C
-0.80 JBL 708P - ASR
-0.85 JBL LSR708I
-0.85 REVEL F226BE
-1.00 JBL HDI-3800
-0.70 JBL M2
-0.80 REVEL F226BE
-1.00 JBL 4367
-0.80 POLK R200
-0.80 REVEL F328BE
-1.20 KEF REFERENCE 5
-0.55 REVEL F206
-0.65 REVEL F208
-0.90 SELAH PUREZZA
-0.90 INFINITY INTERMEZZO 4-1T
-0.60 INFINITY R263
-0.85 MESANOVIC RTM10
-0.90 REVEL C426BE
-0.90 REVEL ULTIMA2 STUDIO2
-0.70 ADAM T8V
-0.80 GENELEC 1032A
-0.85 GENELEC 8050B
-0.90 INFINITY PRELUDE MTS
-0.90 REVEL F35 - ASR

Data for the above was collected here...

A collection of loudspeakers measurements

 

[ADU]

The estimate I use for a sound power slope is -1.0 to -1.5 decibels per octave btw. Or an average of about -1.25 dB/octave.

Speakers with less extension in the bass generally have a slightly shallower overall in-room and sound power slope btw. And the response in the lowest frequencies can easily increases the overal "tilt" of these curves by at least a couple dB on the very best extended speakers.

I'm only interested in the in-room and sound power slope of speakers with good extension in the bass though. So that's what my average ranges above are largely based on.

I don't include the falloff in the sub-bass in my estimates either. I just try to make a rough guess at the curve's slope based on its average overall tilt above that range.

 

[tuga]

Thanks everyone for the feedback.
I did some further reading based on the material you submitted.
It is true indeed that R-squared is not the appropriate metric to establish goodness-of-fit of a trendline, as it increases with slope. It seems Root Mean Square Error (RMSE, lower values are better) is a better metric. Based on RMSE, and the suggestion that the Estimated In Room Response line should not slope downwards more than 6-7dB overall from Upper Bass (100Hz or so) to High Mids (5kHz or so), the ranking of the 6 six speakers I had shortlisted now makes sense, as it correlates with their price (in Greece):

1. RMSE 0.57, Revel M105, €1690
2. RMSE 0.72, KEF LS50 Meta, €1250
3. RMSE 0.76, Focal Aria 906, €790
4. RMSE 0.91, KEF LS50, €860
5. RMSE 0.91, Revel M16, €1120
6. RMSE 1.25, ELAC DBR-62, €520

It seems the Focal Aria 906 is priced lower than its performance would suggest, as opposed to the Revel M16 which is priced a little high. If these two were inversed in the order, the correlation of RMSE to price would be almost perfect. See the pricing in your country, perhaps a bargain will identify itself.

I understand that there isn't a single perfect score to rank speakers - each method has pros and cons. One needs to study the Spinorama info to make an informed decision. Having said that, I wanted a simple but still logical / mathematical approach to rank speakers for an audition shortlist, and I am happy with the above.

Thanks again to all for the contributions and the feedback.

Unless you know your preferred target (for which you would have to listen to a bunch of speakers), the numbers are, well, meaningless...

You could buy a pair of Kefs and can play with the tilt by adjusting toe-in.


Also note that Kef recommends listening at 10º off-axis, so the Spinorama LW tilt doesn't really reflect the optimal listening conditions.

Shortlist by numbers, yes, but then listen to a few different contenders. By going on numbers only you are deluding yourself. Not that different from star-ratings...

 

[Sokel]

That's my favorite slope (after EQ of course,both are captured) and as long is a matter of taste (if all the other things are correct) I'm staying with it.

 

[ADU]

Changed the in-room slope estimates in my previous post from the overall values (across 10 octaves) to decibels per octave, to hopefully avoid a little confusion.

I may revise some of the values above a bit, if or when I have some more time to do a more detailed analysis of the in-room curves. But so far, the average of the 25 speaker values that I listed above seems to work out to about -0.832 dB/octave, which is fairly close to the middle of my estimated -0.75 to -1.0 range.

 

[Kvalsvoll]

Very wide speakers have a much more shallow slope. A fully omnidirectional speaker would have a flat line for directivity and a flat line PIR.

The slope of the in-room response is determined by the power response of the speaker system in combination with the reflected energy form the room.

A "wide" speaker can have the same slope as a "narrow" if both have a similar frequency response of the total radiated sound power. What many consider as a "narrow" speaker is often a speaker with very frequency-dependent radiation, they are omni at low f and then becomes more and more narrow at higher frequencies, and it is this that causes the sloping in-room response.

In a properly treated room, most speakers will have a gentle slope, close to flat. And it does not sound bright.

 

[Jon AA]

Yeah, I think the effort worrying about this is misplaced. With two different speakers of very different directivities, if corrected to the same room curve they are going to sound very different. Most will prefer one over the other...sort of proving one doesn't have a "prefered room curve." They more likely have a preferred speaker type which results in a certain type of room curve. If the same user sets a different speaker up such that he likes it the most, it'll likely result in a different room curve.