Kali Audio IN-8 Studio Monitor Review

[Jon AA]

So how comes that steady-state curve, which above Schroeder practically equals PIR, cannot be used as accurate descriptor of SQ but LW, ER and SP can?

To add to what edechamps said, it's because your two ears and a brain are more sophisticated than a microphone.

The steady state curve heard by the microphone adds all late arriving reflections on top of the direct sound and early reflections and gives you the sum. Your ears don't hear it that way--they will filter out the late arriving reflections and your brain puts them in the bucket of "room sound," or "the space you're in" while it puts the direct sound and early reflections into the bucket of "the sound coming from your speakers." The later is what matters to your ears/brain in judging the sound quality of the speakers (with particular emphasis on direct sound) which is why you don't want to sacrifice them in order to improve the steady state curve.

 

[Jon AA]

As one experiment, I eq'd my very narrow directivity JBL 4722 cinema speakers the same as a pair of wider directivity KEF LS50's "in-room" to a target response. Both speakers used the same subs to take out the low frequency differences: https://audiophilestyle.com/ca/revi...ker-comparison-with-binaural-recordings-r768/

That's interesting, thanks. However, that may not be the best example to show the differences as the dispersion of the two speakers isn't really as much different as some may think at first glance. The 4722 actually has pretty wide dispersion for a controlled directivity speaker, especially vertically (and from most informal tests, sounds better with two channel music than most more narrow dispersion CD speakers). The LS50 does use a waveguide and its dispersion is much more controlled than many flat baffle speakers that use no waveguide.

I think the differences would be more striking comparing a typical flat baffle speaker that's more nearly omni-directional to a higher frequency with a CD speaker that has a narrower 60 degree, etc, horn. Two speakers like that, if perfectly flat on-axis, would have pretty dramatically different shaped PIR curves.

I'm contending the tonal balance of the two when listened in-room would sound the most similar if both speakers are EQed flat on axis anechoically, resulting in much different steady state room curves (and different spacial presentations). If both are corrected to the same in-room steady state response, you'd be dramatically altering the direct (on axis anechoic) sound of at least one of them and should result in very different tonal balance.

 

[jhaider]

JBL has its 705p plagued by a bad port,

If you listen with your eyes on a graph, yes it's a terrible port.

If you listen to music, there's nothing untoward to hear.

The knock on 705P may be reliability.

I think all of this is unnecessary complicated. The predicted in-room response tells us if the directivity is good or bad.

No, the polar maps tell you if directivity is controlled or not.

Regarding predicted in room response, a lot of people seem to be making what maybe we should call the Mitchco fallacy:

The statements from research are

X [here, flat and smooth on axis + smooth off axis] -> Y [here, higher preference]
X -> Z [here, pretty PIR chart]

The Mitchco fallacy is Z, therefore Y.

However, the two statements do not tell you anything about Y if Z is true.
If Z is false, then you know X is also false (ugly PIR -> nonflat or unsmooth on axis or unsmooth off axis), i.e. the contrapositive. However, not X does not tell you anything about Y either.

 

[Absolute]

You may find it surprising but predicted in-room response actually tells us what response to expect in our room.

Take a look at this snapshot from official CEA-2034 document:

So yes, from above of app 400Hz predicted in-room response correlates with the actual in-room steady state measurement.

Below 400Hz this cannot be predicted as room modes, which are dependent on room dimensions, will affect the actual response.

While the curve predicts what happens above a certain frequency, in most cases you're looking at something like the Harman "preferred" curve.
This is the curve;

This is in-room response of both Devialet Phantom Silver and Kii Three in my room;

There you go. Neutral speakers will more likely look like the Harman curve than the flat -10 dB in actual rooms. If you look around at people's in-room responses, they'll look like that with good speakers most of the time. Hence the Harman-curve, which isn't actually a in-room target to equalise after, it's the curve you're likely to get with good speakers in normal rooms.

 

[mitchco]

To add to what edechamps said, it's because your two ears and a brain are more sophisticated than a microphone.

The steady state curve heard by the microphone adds all late arriving reflections on top of the direct sound and early reflections and gives you the sum. Your ears don't hear it that way--they will filter out the late arriving reflections and your brain puts them in the bucket of "room sound," or "the space you're in" while it puts the direct sound and early reflections into the bucket of "the sound coming from your speakers." The later is what matters to your ears/brain in judging the sound quality of the speakers (with particular emphasis on direct sound) which is why you don't want to sacrifice them in order to improve the steady state curve.

Absolutely agreed 100% No question. But this is not how "room correction" works and I think this is where all the confusion is in general.
Not aimed at you Jon.

If I stand a measurement mic up using REW's default 500ms window in my room, I am going to get exactly what you say above, so is Harman when they stood a mic up in their "typical living" room with the F208's and so did whoever measured versus the PIR of the loudspeaker I referenced out of the CTA 2034-A spec. Standing waves, room reflections, early late, direct sound and all. It is the combination of everything. But again, that is not how room corrections works.

Good room correction software first uses a transient analysis derived from the steady state measurement and then separates the later reflections from the early and direct sound. Let's talk transient analysis first:

Our ears perceive that music is mostly transient in nature. The DSP software analyzes the full frequency range of the transient behaviour and this analysis leads to a new frequency response that follows the peaks, but not the dips as can be seen in the frequency response chart above.

If one studies the chart closely, we can see the envelope response is following the peaks but not the dips. (Reference JJ's article on psychoacoustics room correction in an earlier post). Note the marker at just over 100 Hz, one can see a very narrow dip, but the new analysis does not follow that narrow dip. This new frequency response can be considered as an upper envelop of the original magnitudes. The spectral envelope is used as the basis for further calculations. Also note the new analysis sits high in the comb filter region and avoids over correction of dips. Key point to remember as we move to part 2 of the analysis.

Just to be clear, the DSP software is still in analysis mode and not correction design. Now that we have the transient response, we know we want to modify the low frequency standing waves to a target response, both in the amplitude and time domain (there is an ideal loudspeaker in an ideal room, but that is for a full article). Above the rooms transition zone, we "may" want to gently modify the direct sound only and not the later room reflections. This is accomplished by using a frequency dependent window (FDW):

Note the length of time the window is open at different frequencies. The user can control how much window to open at low and high frequencies on a curve as a straight line in a logarithmic frequency plot . This is so we can correct the bass in the room, but not the (later) early reflections and later reflections are not included in the correction design. If we are making a small semi to anechoic adjustment at high frequencies if the speaker is a little dull or bright for example, it is a broad band, low Q amplitude correction, again only to the direct sound. And from the transient response analysis above, it is also not correcting every little dip.

How? there is FDW math that allows you calculate this window size for your size room. I am just going to leave the link here as this is (already) way too much for a thread post: https://www.audiovero.de/acourateforum/viewtopic.php?f=11&t=14 The short of it is I can calculate a window width that is open for 500ms at 20 Hz down to 1ms at 1 kHz and .1ms at 20 KHz.

Playing with the FDW for both amplitude and time and at both low and high frequencies allows one a level of control over the room reflections in different frequency zones and the direct sound from the loudspeaker. At low frequencies I get a nice smooth frequency and phase response with no room disturbances and above the rooms transition frequency, it is either a partial correction, which means full pass through or used as a "tone control" if required.

With the FIR filter correction in place, I drop the mic using REW's 500ms default window to let it all in and I get the "preferred frequency response" curve as shown in Sean Olives slides (no smoothing), but both the frequency and phase response is smooth as butter below Schroeder and then you can see the regular room reflections entering the picture starting around 350 Hz as we are only adjusting the direct sound of the loudspeaker:

Sounds great to my ears Folks may want to read up on how room correction actually works. Try some experiments and listen with your own ears.

Happy Friday!

 

[q3cpma]

If you listen with your eyes on a graph, yes it's a terrible port.

If you listen to music, there's nothing untoward to hear.

The knock on 705P may be reliability.

While it's unlikely to be audible with most musical content, it isn't proved to be for all; and I'm especially wary when I've seen some people (SOS review and an ASR member doing sweeps) saying they could hear it. But even in that case, I just don't want to spend this much money on a problem that could have been fixed easily by putting the port in the back (assuming it's cancellation due to midrange leakage); I mean, 2000€ isn't a small amount of money.

 

[thewas]

I think the differences would be more striking comparing a typical flat baffle speaker that's more nearly omni-directional to a higher frequency with a CD speaker that has a narrower 60 degree, etc, horn. Two speakers like that, if perfectly flat on-axis, would have pretty dramatically different shaped PIR curves.

I'm contending the tonal balance of the two when listened in-room would sound the most similar if both speakers are EQed flat on axis anechoically, resulting in much different steady state room curves (and different spacial presentations). If both are corrected to the same in-room steady state response, you'd be dramatically altering the direct (on axis anechoic) sound of at least one of them and should result in very different tonal balance.

In my limited experience till now those 2 loudspeaker sound tonally also quite different when both EQed flat on axis anechoically, so although the direct sound importance is unquestionable, its seems to be not the only one.

 

[Jon AA]

Absolutely agreed 100% No question. But this is not how "room correction" works and I think this is where all the confusion is in general.

I really think we need to better define "room correction" though, as there are many types/brands and they don't work the same way. Even within Audyssey, for example, the older/lower level MultEQ was much more aggressive at higher frequencies correcting many little bumps and dips that were caused by room reflections (and should be left alone), while the top level XT32 doesn't--instead making smooth, broad tonal shaping corrections at higher frequencies. But I'm pretty sure even it isn't doing what you're describing, at least not very well. When I take 1m gated measurements, I can see where XT32 changed the direct sound where there's a directivity change in the speaker, making it less flat quasi-anechoically.

What you're using above is no doubt much better. If it's successfully eliminating all room reflections from its measurements, then it really is "speaker correction" and would be exempt from many of the generalized criticisms of guys like Dr. Toole. When he speaks of the in-room response (and what the PIR predicts with its formula) is not based on FDW measurements.

But if it can really do that from 5m away from the speaker in a room, why would we need anechoic chambers and Klippels? Theoretically after correction, all speakers would measure flat anechoically. Which puts me right back to where I stated what room correction should do. I have confirmed with some software (XT32) in order to do that you need to take 1m or 2m gated measurements and "fix" what it did manually in the target curve. If there's software that can do the same thing from the MLP, that would be much more convenient. I need to do more reading on the operation of specific brands you may be talking about.

 

[Jon AA]

In my limited experience till now those 2 loudspeaker sound tonally also quite different when both EQed flat on axis anechoically, so although the direct sound importance is unquestionable, its seems to be not the only one.

How were they EQ'd flat anechically? Anyway, I don't think anybody is saying the direct sound is all you're going to hear regarding tonal balance, but it probably is the most important.

The real question is, would either of those speakers have sounded better when EQ'd such that they are not flat on axis anechoically?

 

[thewas]

How were they EQ'd flat anechically?

By myself (gated measurements).

Anyway, I don't think anybody is saying the direct sound is all you're going to hear regarding tonal balance, but it probably is the most important.

I agree that its the most important but not the only one, also the examples I gave before like high shelf filters on good monitors to deal with different room acoustics and that the Harman metric considers the PIR and not just the LW or DS all tend to support this.

The real question is, would either of those speakers have sounded better when EQ'd such that they are not flat on axis anechoically?

I personally tend to prefer EQs on my (decent) loudspeakers that smoothen my locally spatially averaged measured listeners position response above transition frequency rather than just the direct sound or listening window, but that's possibly just me or part of audio's circle of confusion and as I said I think more experiments need to be done in that direction. Also the differences of the EQs are quite small and tend to rather linearise also the direct sound just not fully.
On few loudspeakers I had with not nice directivity neither correcting just direct sound or sound power/room response often gave good sounding results and often the compromise of the manufacturer between both was a better sounding choice, like Toole says in these cases correction doesn't bring much and recommends just getting better loudspeakers.

 

[thewas]

while the top level XT32 doesn't--instead making smooth, broad tonal shaping corrections at higher frequencies.

The XT32 was imho also unusable in most cases unless you bought the Pro-Kit or when much later the Audyssey app was released so you could limit the correction or modify the target curve.

 

[Jon AA]

By myself (gated measurements).

Ok, that's what I'm doing as well.

On few loudspeakers I had with not nice directivity neither correcting just direct sound or sound power/room response often gave good sounding results and often the compromise of the manufacturer between both was a better sounding choice, like Toole says in these cases correction doesn't bring much and recommends just getting better loudspeakers.

That's certainly the best option. If you're stuck with the speakers, you can't fix the directivity problem, but you may be able to fix other things with it (assuming it has other things wrong with it...which it probably does) so there may be some improvement to be had. Chances are high when a manufacturer pays such little attention to directivity, they didn't pay much to the overall frequency response either.

The XT32 was imho also unusable in most cases unless you bought the Pro-Kit or when much later the Audyssey app was released so you could limit the correction or modify the target curve.

Agreed. The App has changed everything though. Along with the free Ratbuddy program, it's a whole different animal (in the right hands). Initial testing with this method has given me some shockingly good results running 30+ control points modifying the target curve.

I know, I know, there are much better systems available than XT32 but the cost difference can be large. So many people already have XT32--basically any D/M AVR over $800 has it built in already and can do all 12 channels. Aftermarket solutions that also require outboard amplification are in a whole different ballpark price-wise. Processors with Dirac are generally much more expensive as well. Thus, my current focus on squeezing all the capability out of XT32 in a way others could replicate.

 

[jhaider]

I just don't want to spend this much money on a problem [sic] that could have been fixed [sic] easily by putting the port in the back (assuming it's cancellation due to midrange leakage);.

It's not. It's a resonance due to the very long low tuned port in a small cabinet.

The only way to "fix" it would be to replace the port with a passive radiator, or multiples.

 

[bobbooo]

Technically, it doesn't even use the LW to calculate the score. It uses the on-axis data instead, which is contradictory. As I understand though, it is only in the last 10 or so years that the listening window has become the preferred metric for 'direct' sound with Harman. The regression model dates from 2005 and with that in mind, is possibly outdated to the latest findings.

LW is actually included in the preference rating formula, as it makes up 12% of the PIR, the latter in turn contributing a total of 38% to the model, via its narrow-band deviation (20.5%) and smoothness (17.5%). Also note, as @edechamps touched on, the variables and weightings of the model were determined by principal component analysis of 23 possible variables, an objective statistical procedure that maximises the predictive power of the model while minimising the number of variables and their collinearity (correlation with each other) - they were not determined by any 'preferred' metric of Harman or Dr. Sean Olive.

 

[Jon AA]

LW is actually included in the preference rating formula, as it makes up 12% of the PIR, the latter in turn contributing a total of 38% to the model, via the narrow-band deviation (20.5%) and smoothness (17.5%) of the PIR curve.

I think those numbers don't accurately reflect the situation though on their face, or at least can be deceiving. Even with perfect directivity control, if you have a really lousy on-axis response, you're going to have a lousy listening window and you'll have really lousy early reflections and sound power...so that lousy on-axis response is compounded into the score as, in the end, it's a factor in all the other curves. This is why I took exception to the implication that the on-axis response wasn't that important because it only made up a small portion of the score (and why I think EQing it can improve speakers quite a bit).

The formula does allow for separation of speakers that may be fine on axis but have other problems (directivity, lobing, etc) off axis. But despite the small percentage, it's hard to conceive of a speaker that could score highly with a really bad on-axis response.

 

[bobbooo]

I think those numbers don't accurately reflect the situation though on their face, or at least can be deceiving. Even with perfect directivity control, if you have a really lousy on-axis response, you're going to have a lousy listening window and you'll have really lousy early reflections and sound power...so that lousy on-axis response is compounded into the score as, in the end, it's a factor in all the other curves. This is why I took exception to the implication that the on-axis response wasn't that important because it only made up a small portion of the score (and why I think EQing it can improve speakers quite a bit).

The formula does allow for separation of speakers that may be fine on axis but have other problems (directivity, lobing, etc) off axis. But despite the small percentage, it's hard to conceive of a speaker that could score highly with a really bad on-axis response.

On-axis response doesn't make up a small proportion of the score. Narrow-band deviation of the on-axis response contributes 31.5% to the model, which is comparable to the PIR contribution of 38%. The on-axis response also makes up 1/9th of the listening window, which pushes its contribution up a bit more. So on-axis response is definitely being adequately accounted for in the final preference ratings.

 

[QMuse]

To add to what edechamps said, it's because your two ears and a brain are more sophisticated than a microphone.

The steady state curve heard by the microphone adds all late arriving reflections on top of the direct sound and early reflections and gives you the sum. Your ears don't hear it that way--they will filter out the late arriving reflections and your brain puts them in the bucket of "room sound," or "the space you're in" while it puts the direct sound and early reflections into the bucket of "the sound coming from your speakers." The later is what matters to your ears/brain in judging the sound quality of the speakers (with particular emphasis on direct sound) which is why you don't want to sacrifice them in order to improve the steady state curve.

I'm familiar with all you've said, but that didn't answer my question.

 

[QMuse]

That's not correct. Not according to @Floyd Toole, anyway. A more correct statement would be:

"As soon as you put speaker in a real room what a single microphone at the listening position will measure will be according to the predicted in-room response (PIR)"

There, Toole would agree with you.

This might seem like a subtle distinction. It absolutely is not, and is the main reason why so many people here vehemently disagree with you on this.

The reason why these two statements are very different is because the predicted in-room response (or an actual in-room response for that matter) does not tell you what part of the response is the direct sound and what is reflected sound. But our auditory system can - Toole spends multiple chapters on this very topic. Therefore the in-room response does not contain enough information to fully characterize what you ear.

This is why Toole is often quoted as saying things like "Two ears and a brain are massively more analytical and adaptable than an omnidirectional microphone and an analyzer" (source - his book also phrases it similarly).



Not a good speaker, no. Not necessarily horrible, though.



Definitely not. Yes, despite the good PIR. Because, again, the PIR doesn't tell you the whole story.



It's interesting that you would bring up the topic of the Olive score, because that actually works against you here. Let me remind you of the coefficients. 38% is directly calculated from PIR (NBD_PIR, SM_PIR). But there is also 31.5% calculated directly from On-Axis (NBD_ON).

Now ask yourself this: if PIR really is the be-all end-all when it comes to perceived performance of a loudspeaker in a room, as you seem to think, then why does the score formula mention NBD_ON at all? Keep in mind Olive's model was developed using a statistical method (PCA) whose whole point is to remove variables that are bad predictors, and only keep the smallest set of variables that make it possible to predict preference with a reasonable degree of accuracy. If On-Axis is just a distraction and PIR is really what matters, then how come the statistical model assigns such a large weight to On-Axis, almost the same weight as PIR?



To be honest, I'm not entirely sure. Its limited usefulness is probably why it's not part of the standard spinorama graph as specified by CTA-2034. I suspect it might be a good indicator of overall tonal trend (i.e. to judge an overall bass/treble tilt), I don't know.

I have reservations on scoring formula which uses on-axis instead of LW. My head is not tightened to chair while listening as mic is to mic's stand.

I also have reservations on scoring formula which scores Harbeth slightly better than Revel C52. Don't you?

Let's have a look. Upper curves is C5s, score 7.6, Harbeth is lower curves, score 7.7:

So, they score the same. Would you say the same by looking at their curves?

Btw, from what I can see Harman guys clearly optimised C52 for LW and PIR, not for on-axis.

 

[TimVG]

LW is actually included in the preference rating formula, as it makes up 12% of the PIR, the latter in turn contributing a total of 38% to the model, via its narrow-band deviation (20.5%) and smoothness (17.5%). Also note, as @edechamps touched on, the variables and weightings of the model were determined by principal component analysis of 23 possible variables, an objective statistical procedure that maximises the predictive power of the model while minimising the number of variables and their collinearity (correlation with each other) - they were not determined by any 'preferred' metric of Harman or Dr. Sean Olive.

Agreen in general and thanks for correcting, as it is technically included. What I'm implying is that since the development of the model, it seems that whereas in former days the primary target was flat on-axis, with smooth off-axis behaviour (will use the JBL LSR6332 in this example, which is apart from a visual update identical to the LSR32)

It seems these days they have shifted, Revel included, to put the primary focus on the listening window, as shown here in the M2. I am suggesting that this is done for a reason. I am unsure how this shift would translate into the overall score with the regression model as shown, but I'm just saying there is likely a reason for the change in priority.

 

[TimVG]

I have reservations on scoring formula which uses on-axis instead of LW. My head is not tightened to chair while listening as mic is to mic's stand.

I also have reservations on scoring formula which scores Harbeth slightly better than Revel C52. Don't you?

Let's have a look. Upper curves is C5s, score 7.6, Harbeth is lower curves, score 7.7:

View attachment 53163

So, they score the same. Would you say the same by looking at their curves?

Btw, from what I can see Harman guys clearly optimised C52 for LW and PIR, not for on-axis.

It seems the 1/2 octave interference dip centered around 900Hz on the Revel accounts for more then the general broadband 1 octave dip centered between 3 and 6kHz on the Harbeth. There is also no deduction for the geneneral 6dB downward slope from 100Hz onward.