A Broad Discussion of Speakers with Major Audio Luminaries

[rdenney]

...
Can confirm the existence and scale of this problem, usually it is easily visible when increasing the SPL in steps of +5 or +10dB and see the true Helmholtz resonance frequency shifting (which I prefer to identify in the nearfield measurement of the active woofer, i.e. the frequency at which its diaphragm comes to a standstill). Same compact sealed design are also prone to this shift, probably due to overly high compression. An indirect variant to identify it is just looking at the dynamic FR:

View attachment 467991

This Revel model is a floorstander afaik. We see compression occurring in the 80-150Hz region as well as dynamic expansion around 50-60Hz. The latter is indicative of an increase in port fs with rising SPL. Imagine a kick drum sound with approx 50Hz fundamental and 100Hz first harmonic. The spectrum of its transient sound will shift in relative 1.5dB.

Does not sound like a lot, but for synthesized kick drums sounds, a significant drift in resonance frequency hence lower cutoff frequency and group delay depending on the actual SPL, can become am audible problem as it comes with other implications. Cannot deliver solid evidence that this is the single reason, as compression and port noise occurs simultaneously, but most of speakers it tried showing this behavior, also ´don't really kick´ in the lowest bass region. EDM sounds are either boomy or hollow, if that makes sense.
...

One reason I bought the Revel F12's was it's ability to play loudly without compression. The Canadian NRCC anechoic test by Soundstage reported this finding:

They did not usually go this high, but with the F12 they decided to because it did so well at lower levels. Note that this is a speaker playing an output at 2 meters at 100 dB SPL in an anechoic chamber, so at 1 meter that would be 106 dB SPL. This was beyond the capability of most speakers they tested, and they only provided this for a few models. The chart shows about 1dB of compression at 50 Hz, about the same amount in the band from 90 to maybe 150 Hz, and 3 dB of compression in a band centered on 1 KHz and again in the top octave. It doesn't show the how it varies with level (at 95 dB the linearity was within half a dB), but what it shows is about 4 dB louder than the F35 you showed. I have this feeling that the change in timbre resulting from a dB of compression in the bass region is going to be hard to detect given how loud this is, and kick-drum peaks are probably too spectrally diffuse and short to draw much conclusion about timbre from a difference this small. I haven't conducted that experiment, and doing so at this level would upset domestic harmony. But I did play--very loudly--the drum solo on the Chesky demonstration CD that was recorded with as much dynamic range as they could muster, and the loudest peaks were flashing the clipping lights on my NC502MP amp (read: at a power output of maybe 350 watts into each speaker for those peaks, so the peaks were probably at 112 or 115 dB at one meter). I didn't detect any change in timbre with changes in level even that loud, though I'm not sure those changes wouldn't be masked by the ringing in my ears

Their testing for the F12 was conducted in 2006 and did include listening window measurements but did not include a full spinorama, unfortunately. But for that, I depended on Revel's established reputation for wide, smooth directivity.

The F12 had two 8" woofers plus a 5" mid and a dome tweeter in a Revel waveguide. This is more radiating area than the later (and smaller) F35.

Rick "wishing this chart was included in all speakers tests" Denney

 

[gnarly]

And can someone clarify for me what "timing" means in relation to group delay and phase? If both of those are below thresholds of audibility, does this make timing acceptable?

Ok, I'll give it a go.

For me, and my practice...
Time alignment is when each driver sections' impulse response arrives at the same time to the measurement mic. So time alignment involves fixed constant time delays between the sections, and nothing else.
For IIR crossovers, the delays align the driver sections' impulse starts, their initial rise from zero. For linear phase crossovers, the delays align the sections' impulse peaks.
Each driver section needs its crossover filters and any other response smoothing filters in place, before measuring the impulse arrivals, and setting the fixed delays.

Phase alignment is when adjacent driver sections' have their summation region achieve an acoustically complementary crossover, in which case their phases traces will overlay.
So for me, phase alignment involves first choosing an acoustically commentary crossover type, frequency, and order that will work between sections. Then on each of the sections, first apply in-band, and out-of-band minimum phase frequency magnitude smoothing (which we know smooths phase too), and after that add whatever high-pass or low-pass is needed to match the desired target crossover.

When both of those alignments are accomplished for all driver sections, phase traces of individual sections will lay on top of each other throughout critical summation regions.
And the group delay curve will primarily reflect the all-pass rotation inherent in IIR crossovers. The GD curve will transition rather smoothly between the flat region levels away from crossover.

If time alignment is not correct, group delay or rather so-called excess group delay, will show a time difference between the flat regions of the GD curve. Each flat region being that of an individual driver section. This excess group delay timing difference is the same as the difference found via measuring impulse time arrivals. (Much prefer to use impulse timing)

If only phase alignment is off, the GD transition region near the crossover frequency will show a more kinked look compared to a smooth transition when phase is aligned.

When both time and phase are off, flat regions of GD become curved, transition kinks become sharper.

I have to say, I think group delay is a curse on audio....the most misunderstood technical term going.
GD's first problem is confusion over whether it includes fixed delay or not. Personally, I think all fixed delays should be removed from audio measurements before assessing GD or phase. GD and time don't go together imo, as GD is frequency dependent and time is not.
Next, GD would have us believe the difference in reported times between flat regions represents real time. The only situations where that is true is when there are real timing differences between impulse arrivals. The reported time difference GD shows for the sides of IIR crossovers or all-pass filters, are not real time and are not good substitutes for delays. (I get why the passive folks have used GD in the past as a fixed time substitute given the difficulty in making LCR delays. It's time to let that kludge go)

Lastly, I think many audibility tests of GD have missed the boat.
Because I think GD audibility really needs to be focused on the frequency region of changing group delay. Which creates so-termed envelope distortion for transient signals that have frequency content in the region's bandwidth.
I find short tone bursts within the frequency region of changing GD...... no GD, vs varying orders of GD.....to be audibly different. (and certainly measurably different)

No point in me griping about group delay with linear-phase, since there isn't any! Now if there were just a way around the dang FIR filter fixed delay times.

Anyway, thx for reading if you made it though. I'm self educated on this, which always comes with the risk of gaping holes in knowledge and understanding.
Please correct or advise if appropriate, and I'll say thank you sir !

 

[Holmz]

My reading is rather 0.3ms, not seconds ;-)

Luckily I had also converted that to distance Sp we worked i tout together.

A runtime difference of 0.3ms might look ugly on the step response graph, same to the inverted midrange, but I see no evidence that these are audible if it is just phase shift or group delay between midrange and tweeter (identical for all channels). On the other hand, a potential delay of the lowest bass frequencies caused by a steep subwoofer lowpass or port resonator, might cause audible group delay but is difficult to identify on such a graph spanning over just 10ms.

Step response is IMHO rather useless for direct predictions on audible aspects.

That might be true that is useless for predictions on audibility, but I am not sure.

That 0.3 ms oud indeed be a phase shift for say the harmonics of some note that starts off in the MR and then gets into the tweeter.
Amps do not create those phase shifts with harmonics.

That is not much information from the step responses.
And I assume @pma was more asking along the line of which speaker has “kicking bass" and which has not or other results in respect to SQ and preference. If phase/step response is so important that should be easy.
…

Well I assume that he knows what it should have shown me.

Sorry I didn't answer you sooner. Are you aware of all the studies conducted by these disciplines, including physics, mathematics, engineering, and neurophysiology, and have you created speakers that respect or utilize this knowledge? Have you ever listened to speakers designed and built for time and phase coherence in controlled systems? Just answer yes/no to these two questions.
Thanks

I am not going to answer for Suono - but let’s consider this from this example…

1. We take a speaker with a poor step response and use DIRAC live on it, which reportedly sounds different than not using it.
   1. And e largely recover a better time and phase performance from the system
   2. But maybe the cross over regions are still a bit janky.
2. We take a speaker with a s good step response and use an EQ to correct the FR.
3. We use an Active XO.

In case #1 the passive XO is usually LR2 or LR4, or sometimes other higher orders.
And the system has more stored energy in the XO.

In case #2 the passive XO is usually 1st order.
So any FR EQ should not perturb the time-domain behaviour.

Maybe Case #3 is the most optimal as the speaker has no stored energy in a passive XO?

^that is all sort of a question^ or maybe a though experiment for the “Audio Luminaries”.

 

[pma]

I would guess that the first is a 3 way with 1st order or no xover between low and mid drivers ( mid hi pass from a separate enclosure ) all drivers are in phase and I don't know about the tweeter slope.
The second looks to have limited hi freq. output so probably a single driver.

Both are the same 2-way speaker (with horn loaded compression tweeter driver). The only difference is that the bass/midrange driver in step2 is connected with opposite phase. None of the versions is optimal, but step2 version has more flat on-axis frequency response.

 

[krabapple]

I was never talking about automatic corrections. I was talking about corrections above Schroeders frequency.

? If Joe Audiophile runs Audyssey -- an automatic correction -- it's going to correct below and above Schroeder, unless Joe downloads the app and specifically limits the correction bandwidth.

I asked the same question earlier in the thread, but got no response.

I am getting the impression that some members here have committed themselves to some in-principle ideas, maybe even put a lot of expense or effort into developing their hifi around such ideas, and are then resistant to evidence that the ideas are not perceptibly beneficial. For example:-

This argument is exactly the same as the argument that super-high sampling rates and bit depths eg 32/768 are an extremely good idea because..."getting the input and output signals to exactly match" is "technical excellence" and hence 24/48 will never do. And any arguments that 24/48 is completely transparent to the ear will be dismissed as "yada yada and more yada".

(etc)

I agree (with the whole post which I clipped for space)!

It's interesting watching the two 'sides' go at it on this. I know sweet f-a about building a good sounding speaker, but from here it looks like one side (Dr. Toole et al) says, in the end, what matters is what difference we can hear and thus choose (or not) in real rooms, and these are backed with experimental data. He's been careful to note the ambiguities and corner cases in the extant literature/measurements. The other side (Holmz, gnarly et al) seems convinced that the goal of 'exact input/output matching' or 'perfect time/phase alignment' is both intuitively desireable and audibly consequential, based on their own (typically non controlled) listening and work (but not always just that -- they also cite literature and measurements) .

Perhaps it comes down to being a personality trait, an inner imperative. What used to be called 'perfectionism', the pursuit of which is a source of pleasure. That means it won't easily be countered with words. Not always a terrible thing, but as you say not always helpful.

 

[Arindal]

This seems like reasonable performance from a pair of 5 1/4" woofers in Revel's entry level floorstander, the F35. If one wants serious high level kick drum sound this little speaker will be stressed, but if high passed and supplemented by a subwoofer, things would be more satisfactory.

It might look like a reasonable compromise in terms of lower cutoff frequency and max. SP. Nevertheless it is supporting my claim that even medium-sized floorstanders are showing hints of the lower bass not being ideal in terms of dynamics and port frequency stability with electronic sounds (don't know how this particular models sounds, it was just an example of typical results which are commonly associated with such impression).

Otherwise, according to Amir's and Erin's data, this is a respectable speaker at the price. As always there are choices, and knowledge and data help make them.

The question is if it is a reasonable compromise for what common pop music lovers expect. It is a 3-way, 4 driver design, bigger than anything a common music listener would ever let into his or her living room. Explaining people why such a chunky speaker does not deliver a precise timing and ´kick´ with EDM (if that would be the case, we cannot be sure based on the measurements), or advising them to switch to something even bigger or a multi-subwoofer array, is IMHO not the best way to bring people to hi-fi in general or convince them that a science-based approach to hi-fi is leading to a maximum of personal listening enjoyment.

People are used to portable bluetooth speakers 1/10th the size of that floorstander and 1/5th of its prize to deliver satisfyingly precise, low-reaching, kicking EDM beat that makes you dance. At least the class-leading models are capable of that, if you don't take max SPL as a goal of its own into account. If a review based on measurements cannot tell which floorstander can deliver a similar amount of ´fun bass´ and which is just producing some bloated boomy bass blob, isn't it understandable they all completely ignore measurement-based reviews and fall for subjectivistic influencers dancing on TikTok to the sounds of some bluetooth cans?

 

[pma]

It is a simple physics, we need 10” woofer at least to get cone area and reasonable excursion. Small diameter cone + large excursion is not the way to get high SPL undistorted bass.

 

[Purité Audio]

Or multiple smaller drivers.
Keith

 

[Blueprint]

Can someone point me to these models that are 1/10th the size of the F35 but deliver clean, low bass? I’m in the market for about ten of them.

 

[Arindal]

Because I think GD audibility really needs to be focused on the frequency region of changing group delay.

Which frequency bands and thresholds of GD delta do you have in mind? How do you expect such a listening test to be designed?

I am aware of the definition of phase shift and group delay and I have worked with loudspeakers which are said to be class-leading regarding time coherence. I took part in several experiments, some of them controlled and blind, to find out what is the difference between linear-phase and minimum-phase x-over design, and the results were pretty sobering for everything above 100Hz. Some trained listeners attributed a slightly more stable imaging to the linear-phase variant, but that was basically it. Most of participants did not distinguish any difference between two scenarios looking wildly different on the step response graph.

The only exception was the lower bass region with artificial kick drum sounds, like EDM. Linear phase mode down to 20Hz does surprising things in this case, but it is barely (or not at all) visible in the step response. And such concepts are vastly unpopular due to high system latency.

We take a speaker with a poor step response and use DIRAC live on it, which reportedly sounds different than not using it.

Is it possible to correct solely the time domain with this system, independent from amplitude response and with not indirect influence on the latter? I am not aware of such mode, and amplitude correction would always dominate the perceived differences.

Can someone point me to these models that are 1/10th the size of the F35 but deliver clean, low bass? I’m in the market for about ten of them.

Try such:

Not really loud, and bass limiter will kick in if you listen at higher levels, but for average Joe this delivers pretty clean bass impulses.

 

[amirm]

For me, and my practice...
Time alignment is when each driver sections' impulse response arrives at the same time to the measurement mic. So time alignment involves fixed constant time delays between the sections, and nothing else.

Why is that a good thing other than appealing to your lay intuition? An impulse is an idealized test signal with infinite bandwidth. It is not representative of any music.

Beliefs like this is what leads some speaker designers to forgo higher order crossover filters and with it, produce horrible responses with woofer break up and such. Another example is stacked drivers which invite diffraction galore:

For IIR crossovers, the delays align the driver sections' impulse starts, their initial rise from zero. For linear phase crossovers, the delays align the sections' impulse peaks.

The job of the crossover better be first and foremost to create a flat on-axis response and good off-axis. Don't confuse its job with that of a watch.

I find short tone bursts within the frequency region of changing GD...... no GD, vs varying orders of GD.....to be audibly different. (and certainly measurably different)

You find? How do we know that is the reality absence of controlled testing? How do you know that is the only thing going on? And how are "short tone burst" relevant to what you use the speaker for?

As I said, the problem here is that the few of you seemed to have put the cart before the horse. You have convinced yourself, not based on research and controlled testing, that somehow things being "time aligned" means good sound reproduction. Damn the many flaws created in the process of achieve such goals.

Learn to build a simulation tool, perform controlled testing, and then bring that here. Until then, you are piling supposition on top of supposition.

 

[ROOSKIE]

View attachment 467991

This Revel model is a floorstander afaik. We see compression occurring in the 80-150Hz region as well as dynamic expansion around 50-60Hz. The latter is indicative of an increase in port fs with rising SPL. Imagine a kick drum sound with approx 50Hz fundamental and 100Hz first harmonic. The spectrum of its transient sound will shift in relative 1.5dB.

Does not sound like a lot, but for synthesized kick drums sounds, a significant drift in resonance frequency hence lower cutoff frequency and group delay depending on the actual SPL, can become am audible problem as it comes with other implications. Cannot deliver solid evidence that this is the single reason, as compression and port noise occurs simultaneously, but most of speakers it tried showing this behavior, also ´don't really kick´ in the lowest bass region. EDM sounds are either boomy or hollow, if that makes sense.

Ceratinly, those are not speakers designed for 100db average playback levels.
But this makes no sense to me to examine this 'compression' without factoring that at 102 vs 86 the human perception of bass-midrange-treble balance is completely different.
Any boomy or hollow 1 to 1.5db variation from 50-150hrz is swallowed up by that fact that we do not perceive increases in bass in a fashion that is linear with increases in the midrange frequencies.
If someone is listening at 85db average levels from 250-2500hrz and a kick drum hits big peaks at 97db (which sound to /are perceived by the listener as about 3-4 times as loud as the 85db averages), are they really thinking geez, there is an extra 0.5db at 50hrz and that 100hrz harmonic just sounds a little thin? As you said, is something else at play?

What if they listen at 70db averages and the then the peaks are 82db. How does this affect the perceived sound and tonal balance vs the louder example above? Probably a lot more than the slight compression issues.

50-150 hrz is in the range where many room modes, SBIR issues and other things are deeply in play. Not to mention what size is the room they are in and are the close to the front wall or pulled out. I don't see the minor changes in port tuning as SPL changes and driver temperatures change to be nearly as big of an issue in terms of noticeable effect as any of the things I just mentioned.

Curious folks usually just don't have the ability to test this stuff outside of our biases and without a lot of testing limitations due to complexity. I'm sure we all would.

 

[gnarly]

Which frequency bands and thresholds of GD delta do you have in mind? How do you expect such a listening test to be designed?

The only GD band of interest to me anymore, is the system high-pass band. The 25-30 Hz zone for what I build. Since I use linear-phase crossovers, there is no GD to consider anywhere else.
For ported and sealed subs, I've compared given IIR high-pass orders, usually BW3, to same high-pass linear-phase, to degrees of phase linearization of the IIR high pass.
Outdoor listening, with a processor that can silently and instantly switch between any scenario.
Short term results are inconclusive, because one track sound better this way, next track better that way. I've learned to just stick with a trial and let it play for a few weeks both outdoors and in, and see if an impression forms vs a lot of different music. Anecdotal for sure.

As far as linear-phase vs minimum phase.....I think well done minimum phase, which means smoothly gradually sloping phase across the spectrum, and linear-phase are too close to call. I think kinks in the phase curve, just like high Q PEQ's get easy to hear.

The overwhelming advantage of linear-phase imo, is how easy it is to achieve the beautiful spins ASR judges everything by. I value the spins too, and have a 4 mic vertical mast off of deck, in combination with turntables to make spins. About 25ms before first reflections.
When I set up 4-5 ways with IIR, it is an incredible bitch to get the same quality spins as I can with lin-phase FIR. It is so easy to align impulse peaks for timing with IIR. It is so easy to get phase traces to overlay when phase traces are flat at zero degrees. It is so easy to choose whatever crossover frequency gives best polars between sections, with effecting any other sections. Bottom line, it makes for a better speaker ime, with NO downside other than latency.
And surely ranks higher, has more potential than chaser electronics' SINAD for heavens sake.

 

[gnarly]

Learn to build a simulation tool, perform controlled testing, and then bring that here. Until then, you are piling supposition on top of supposition.

You presume so much about me, and obviously without fully digesting what I post, it hurts.
I suggest you consider more out of the box testing, measuring, and listening.
Klippel NFS spins showing only magnitude response, and small room audio science, are not the be all end all to audio.

 

[sceptical1]

This is turning into another incoherent mess.

 

[kyuu]

Klippel NFS spins showing only magnitude response, and small room audio science, are not the be all end all to audio.

Klippel is capable of characterizing more than magnitude response. Trouble is, the evidence we have is that the magnitude is by far the most important metric. All this kerfuffle about precisely lining up impulse responses, on the other hand, has weak or no evidence.

 

[Arindal]

Ceratinly, those are not speakers designed for 100db average playback levels.

Surely not, and I was not meaning to say they were a reference. It is just an example of what is an average speaker for average popular music lovers with pretty clean and tight bass response. My guess would be people who had such, would not want to take a step back in terms of bass quality, and it is not as easy to beat as it looks from jus technical spec point.

But this makes no sense to me to examine this 'compression' without factoring that at 102 vs 86 the human perception of bass-midrange-treble balance is completely different.

We might have a misunderstanding here. I did not say the 1.5dB compression/expansion is the sole cause of what people might not like in the bass of a typical passive bookshelf or decently-sized floorstander. It was rather a hint that resonance frequency of the port is shifting to a certain degree with SPL, which is a potential explaination why bass precision and ´timing´ might not be satisfactory. So rather several phenomena hinting to the same cause than one causing the other.

The compression itself looks rather harmless with this floorstander, which I mentioned in reply to the comment that rather very compact bookshelf designs are affected. They are even more so, naturally. Have stumbled across several very compact active vented speakers with pretty astonishing specs in terms of max SPL and lower cutoff frequency, which did not really sound convincing to me in the lower bass region.

50-150 hrz is in the range where many room modes, SBIR issues and other things are deeply in play. Not to mention what size is the room they are in and are the close to the front wall or pulled out. I don't see the minor changes in port tuning as SPL changes and driver temperatures change to be nearly as big of an issue in terms of noticeable effect as any of the things I just mentioned.

That is exactly the problem. If port tuning freq at modest SPL is not hitting a mode, it might sound fine, but with certain sounds like kickdrums at higher levels it excites a mode for a moment.

 

[gnarly]

Why is that a good thing other than appealing to your lay intuition? An impulse is an idealized test signal with infinite bandwidth. It is not representative of any music.

What does an impulse response have to do with being representative of music? It's simply a measure of system response.

Beliefs like this is what leads some speaker designers to forgo higher order crossover filters and with it, produce horrible responses with woofer break up and such.

Beliefs like what please?

Higher order crossovers? I've been publicly recommending 96dB/oct complementary linear phase crossovers for years.

 

[Holmz]

…
It's interesting watching the two 'sides' go at it on this. I know sweet f-a about building a good sounding speaker, but from here it looks like one side (Dr. Toole et al) says, in the end, what matters is what difference we can hear and thus choose (or not) in real rooms, and these are backed with experimental data. He's been careful to note the ambiguities and corner cases in the extant literature/measurements. The other side (Holmz, gnarly et al) seems convinced that the goal of 'exact input/output matching' or 'perfect time/phase alignment' is both intuitively desireable and audibly consequential, based on their own (typically non controlled) listening and work (but not always just that -- they also cite literature and measurements) .

Perhaps it comes down to being a personality trait, an inner imperative. What used to be called 'perfectionism', the pursuit of which is a source of pleasure. That means it won't easily be countered with words. Not always a terrible thing, but as you say not always helpful.

I am not sure that we can call meta analysis, or hundreds of ABX runs to see what people statistically like as being “hard science”.
However it certainly gleans a lot of insight into things.

People generally have to pick their compromise.
And then they throw some digital correction on it.
Or in the case of using throw woofers, then go to a more expensive motor which is more linear over a longer throw.
But those were designed from the ground up to be linear, and not the result of ABX testing.

Whether they pick their compromise with ears, or with data, or with looks, or cost… they get something that is generally not 100 perfect.
Hence my questions to “Luminaries” is what is at least in theory the most perfect?

Or does that fidelity not even matter because we generally choose to just ignore the time domain and FR is 99% of it.

Why is that a good thing other than appealing to your lay intuition? An impulse is an idealized test signal with infinite bandwidth. It is not representative of any music.

I think that it being a measure of “system response” was already mentioned.

Beliefs like this is what leads some speaker designers to forgo higher order crossover filters and with it, produce horrible responses with woofer break up and such.
…

You are not alone in your revulsion to low order XOs, and many others share that boat with you.

Hence the question for the “luminaires” had parts 1, 2, and 3.
With which is best in theory:

1. The L/R based passive XOs, time domain corrected with say DIRAC, or some other phase correcting FIR, and ideally one where the taps can be seen/controlled.)
2. 1st order passive XOs (FR corrected with DSP/EQ)
3. An active XO, which could include FIR designs. Sort of needs to be FIR to correct phase.

And I suppose #4 being - we do not care about system response, etc… maybe mostly FR, pattern, and then compression in a distant 3rd, etc.
Which then makes me wonder why Klippel even includes step function response, and impulse response, as an output plot(s).

 

[Chr1]

Been following this thread tho feeling pretty stupid as most of it's way over my head unfortunately. Learning tho

Be interested as to why my two favourite speakers can be technically so different.

Active three way Neumann KH310s, and passive 12" dual concentric Tannoy V12s.
Used with twin subwoofers.
Love them both, but they are so different.