Why evaluating the sound of a single speaker is essential

[Matt_Holland]

Why would increased loudness sensitivity in our hearing at some frequencies make us more sensitive to a loudspeaker crossover at the same frequencies?

Hifi reviewers are always saying that 2-3KHz is a bad crossover frequency because that’s where our hearing is most sensitive. However I think what these reviewers were experiencing was bad crossover implementations and the resulting non-linearity in direct and reflected sounds.

 

[Duke]

Why would increased loudness sensitivity in our hearing at some frequencies make us more sensitive to a loudspeaker crossover at the same frequencies?

Because anomalies tend to occur in crossover regions, off-axis if not on-axis. If this happens in a region where the ear literally hears better, the anomaly is more likely to be detected.

 

[bmc0]

Now the bumpage seems to be centered on about 1.5 kHz, which is arguably good new for speaker designers: Seems to me it's usually easier to cross over at 1.5 kHz than down around 1 kHz.

I don't see why equal loudness contours for pure tones in anechoic conditions should have much relevance to crossovers. Is there any evidence that sensitivity to spectral modifications of a wideband stimulus is significantly reduced around 1.5kHz (or, conversely, significantly increased around 3kHz)?
Lower/higher sensitivity to pure tones does not imply worse/better in all metrics. Interaural phase sensitivity, for example, deteriorates rapidly above ~1kHz.

 

[Duke]

I don't see why equal loudness contours for pure tones in anechoic conditions should have much relevance to crossovers.

Crossover or not, ime the 3-4 kHz region is the most critical part of the spectrum, and I believe this correlates with the ear's heightened sensitivity to SPL in that region. Both too much and too little energy in that region, even if it's only a decibel or so, is ime disproportionately audible.

Is there any evidence that sensitivity to spectral modifications of a wideband stimulus is significantly reduced around 1.5kHz (or, conversely, significantly increased around 3kHz)?

I take the equal loudness curves themselves to be evidence.

Lower/higher sensitivity to pure tones does not imply worse/better in all metrics. Interaural phase sensitivity, for example, deteriorates rapidly above ~1kHz.

I do believe the equal-loudness contours are telling us something useful about perceived loudness. And to the extent that perceived loudness correlates with audibility, imo equal-loudness curves are telling us something useful about audibility.

 

[amirm]

I don't see why equal loudness contours for pure tones in anechoic conditions should have much relevance to crossovers. Is there any evidence that sensitivity to spectral modifications of a wideband stimulus is significantly reduced around 1.5kHz (or, conversely, significantly increased around 3kHz)?

How audible are flaws at 30 Hz or 22 kHz for a bookshelf speaker? And how would you know?

 

[amirm]

What this tells me is that we really must listen in mono when evaluating a single speaker.

That would obviously highly depend on what music you picked to listen to. For what I use, it makes no difference as modern music tends to not have had pans and such.

Also, when you try to sum, you can get both overflow and cancellations. It is quite tricky to create such a mono mix after the fact.

Net, net, listen to your test music and make sure it is full range and continuous in each channel and go with that.

 

[MarkS]

That would obviously highly depend on what music you picked to listen to.

Yes. And in my experience, many classical music recordings have out-of-phase information that is essential to approximate realism. This is not studio panning, this is what is coming directly into the mikes, especially with minimally-miked recordings like those done by Telarc.

As a test signal, I like to use out-of-phase pink noise. I think this is a far more revealng test of speaker directivity + room interaction (and especially how that changes with frequency) than any mono test. (And recall that Harman research has shown that, in mono, pink noise is the most revealing of all test signals.)

Interestingly (to me, anyway), I have found that narrow directivity speakers tend to have subjectively wider projection (beyond the speaker boundaries) of pink noise than do wide directivity speakers. My hypothesis is that this is because the out-of-phase signal is being less contaminated by reflections.

On the other hand, for orchestral music, wide directivity speakers usually sound more realistic to me. Here my guess is that the room reflections are simulating the hall, in a way that is not true to the original recording, but is still subjectively more realistic to me.

 

[bmc0]

How audible are flaws at 30 Hz or 22 kHz for a bookshelf speaker?

Care to explain your point? You can't hear flaws at frequencies that you can't hear, therefore the shape of individual equal loudness contours predicts sensitivity to spectral coloration of non-flat, wideband signals like music in a reflective room? I'm not following the logic here...

 

[amirm]

Care to explain your point? You can't hear flaws at frequencies that you can't hear, therefore the shape of individual equal loudness contours predicts sensitivity to spectral coloration of non-flat, wideband signals like music in a reflective room?

How do you know you can't hear? Answer: you used FM curves!

By your logic, noise shaping is of no value just the same. Right?

 

[dualazmak]

....
Hifi reviewers are always saying that 2-3KHz is a bad crossover frequency because that’s where our hearing is most sensitive. ....
....

I essentially and generally agree, even though I have never done/experienced intensive blind ABX comparison between passive/active crossover in 2-3 kHz vs. without such crossover.

This is one of the reasons I still stick to excellent-and-wide-dispersion midrange driver covering ca. 500 Hz to 6 kHz Fq zone. In my case, it is "still-the-world-heritage" (I believe so) 8.8 cm vapor deposited Beryllium dome midrange driver YAMAHA JA-0801 in heavy (39 kg) rigid sealed YAMAHA NS-1000 (not NS-1000M) cabinet where the JA-0801 is directly and dedicatedly driven by ACCUPHASE E-460 integrated amplifier.

Since the active DSP "gentle/mild" high-pass XO filter is -12dB/Oct Linkwitz-Rilley at 500 Hz, and also "gentle/mild" low-pass XO filter is -12dB/Oct Linkwitz-Rilley at 6 kHz, the JA-0801 midrange driver excellently sings together with woofer even in ca. 300 Hz to 500 Hz Fq zone and also excellently sings together with tweeter even in ca. 6 kHz to 10 kHz Fq zone; JA-0801, therefore, is still one of the best midrange drivers ever produced.

For the details of my present audio setup and Fq-responses thereof, please visit my posts #931 and #1,009 on my project thread.

 

[Torin Krell]

I do believe that single speaker listening tests are the best way to evaluate speakers especially with respect to all the research I have read, and informed comments here. However speaker matching between the left and right pair should also be considered and measured. Manufacturing tolerances vary significantly among speaker companies and a mismatched pair can significantly impact stereo presentation, imaging, soundstage and envelopment.

 

[Mike123]

By the same reasoning, the 3-4 kHz region (and maybe somewhat to either side) looks like an especially challenging place for a crossover, as this is where the ear is most sensitive and likely to be the least forgiving.

In addition to the ear's sensitivity, there is also a spectrum of musical information. It is quite possible that 1/2/3 harmonics of all musical instruments and voice are below 3 kHz. Therefore, cutting bands at 800-1000 Hz may be more audible than at 3 kHz.

 

[Hayabusa]

I don't remember where it comes from; I saved it years ago, and just now was unable to find it online again.

I did not realize how different the more recent curves are!



Now the bumpage seems to be centered on about 1.5 kHz, which is arguably good new for speaker designers: Seems to me it's usually easier to cross over at 1.5 kHz than down around 1 kHz.

most sensitive area is still 3-4KHz from what I see?

 

[Heinrich]

Now the bumpage seems to be centered on about 1.5 kHz, which is arguably good new for speaker designers: Seems to me it's usually easier to cross over at 1.5 kHz than down around 1 kHz.

It isn't the bump for its own. It could be, as a speculation, the non-constant relation between the physical intensity and subjective loudness, e/g double intensity => (not) double as loud. Little deviances get amplified by such, namely in the bass. Since the subjective sensitivity increases with higher levels, a too much technically would impact the impression a bit more. But, that's a small effect, me thinks.

The bumpy shape as such wouldn't affect anything. And as he said, it isn't thechnical data, it is subjective data and as such will be different from person to person.

And again with the following posts, with all due respect, I see much speculation that isn't marked as that. As if everything is picked up and pressed into a theory. (Can be done with virtually everything for virtually all theories: confirmation bias. The adds use that.) That for, science is about falsification - the contrary process.

 

[Heinrich]

Let's formulate a hypothesis, shall we?

Context is evaluating speakers in regrad to subjective preference by a wider audience. The test panel (real people) is not all trained in evaluating, but some are. Since we have no predefined 'neutral', we first ask for discrimination of speakers one against the other, secondly we ask for a preference considering the differences. Then after we design a rank on an ordinal scale, and consequentially lift that on a rational scale by counting the individual verdicts, doing some statistics.

Given this, we hypothize that:
- listening to just one speaker, in particular not two in stereo or many in surround, the discrimination is more stable, consitent over the panel (what the necessary mono signal is, is left to be determined)
- the so found preference translates to the real use of a speaker pair, or a surround setup respectively

In answer to the hypothesises I would argue that
- discrimination is there, the data shows it, preference is there also
- translation is missing, just because the dicrimination is diminished in stereo etc and so is preference

Ironically, if you don't mind, the call for more discriminative testing contradicts itself. It doesn't translate, which very fact is taken as the motivation for mono testing.

NB: I handed the above over to an LLM (AI). It got it right, as far as I'm concerned, and putting all the sticky praise besides, it concluded that there's a "philosophical" framing:
- methological purity
versus
- ecological validity

In short: "We can measure well in mono, but what we measure doesn’t matter as much in stereo."

To argue after the fact, that the better mono speakers are found to be the more sound (*g*) in terms of engineering isn't valid, given the above context, right?

Won't speculate on why the discrimination is more robust when doing mono, because, again, it won't help too much.

Have fun, no offense intended!
I'm decidedly not 'anti', see my post

• #596

 

[Floyd Toole]

Yes. And in my experience, many classical music recordings have out-of-phase information that is essential to approximate realism. This is not studio panning, this is what is coming directly into the mikes, especially with minimally-miked recordings like those done by Telarc.

As a test signal, I like to use out-of-phase pink noise. I think this is a far more revealng test of speaker directivity + room interaction (and especially how that changes with frequency) than any mono test. (And recall that Harman research has shown that, in mono, pink noise is the most revealing of all test signals.)

Interestingly (to me, anyway), I have found that narrow directivity speakers tend to have subjectively wider projection (beyond the speaker boundaries) of pink noise than do wide directivity speakers. My hypothesis is that this is because the out-of-phase signal is being less contaminated by reflections.

On the other hand, for orchestral music, wide directivity speakers usually sound more realistic to me. Here my guess is that the room reflections are simulating the hall, in a way that is not true to the original recording, but is still subjectively more realistic to me.

Just a comment. What you call "out-fo-phase" information in classical recordings is actually long-delayed reflections (longer than the delays in small listening rooms). Phase is not the issue. Telarc's minimally miked (spaced omnis for a while as I recall) collected masses of reflected sounds generating a pleasing, for some pieces of music at least, spacious illusion. Microphone pickup is in principle no different from "studio panning", except that the mics collect useful early reflections. The classic "Blumlein pair" is pure amplitude panning plus reflections.
Pink noise is the most revealing signal for detecting the presence of resonances in multiple-loudspeaker comparison tests. That's all.
Cheers,

 

[Floyd Toole]

Because anomalies tend to occur in crossover regions, off-axis if not on-axis. If this happens in a region where the ear literally hears better, the anomaly is more likely to be detected.

Back in post 599 the curve in (b) by Stevens was done using bands of sound in a normally reflective room and arguably is the most relevant to this discussion.

 

[gnarly]

Why would increased loudness sensitivity in our hearing at some frequencies make us more sensitive to a loudspeaker crossover at the same frequencies?

Hifi reviewers are always saying that 2-3KHz is a bad crossover frequency because that’s where our hearing is most sensitive. However I think what these reviewers were experiencing was bad crossover implementations and the resulting non-linearity in direct and reflected sounds.

The acoustic design of the DIY speakers I'm into, synergy/MEHs, allows crossover frequencies to be changed without altering lobing potential much, if any.
So I've played with varying frequencies and crossover orders quite a bit. I've become pretty convinced that well done crossovers can be placed anywhere in the spectrum without audible compromise.

So I tend to agree with your assessment that reviewers saying 2-3kHz is a bad range, were probably due to bad implementations.

Because anomalies tend to occur in crossover regions, off-axis if not on-axis. If this happens in a region where the ear literally hears better, the anomaly is more likely to be detected.

That also makes sense to me.

I think the foundation of excellent crossover design is the acoustic design, the physical spacings between the various sized drivers being used.
Seems to me, the acoustic design ultimately determines the degree of lobing that a speaker can't escape, no matter how well the electrical crossovers are implemented.
If the crossover lobing occurs in the ear's sensitive range, it seems to me the odds of it being problematic have to increase.

 

[Heinrich]

If the crossover lobing occurs in the ear's sensitive range, it seems to me the odds of it being problematic have to increase.

The "sensitve range" is detected somwhere else every other day. I would say, all the range is important, if one is after perfection. Not the least, the mechanism by what a flaw like lobing would act on increased sensibilities is not explained. And if there were, it wouldn't be validated by experimentation, correct?

For the actual least, the lobing isn't addressed for decades other than by Genelec and KEF successfully, and some other firm w/ non-coaxials. In case, how come that a significant flaw is ignored for so long? More so, some are categorically talking against the presence of a problem, not only ignorance but straight out denial with virtue.

Now you resurrect it as a specuative confirmation of a theory. It is not the first aspect to be absorbed by the latter. But, as explained earlier, science is not about confirmation. All the speculations on what might contribute to the relevance of the theory are useless, until there's no way ro prove the theory wrong (see "Sir Karl Popper", "positivism" (sic!), "science", "falsification".) The issuer of a theory is asked to hand out a falsification method. "If so and then, my theory is disproven."

I perfectly see, that this is about engineering to begin with, applied science. It follows established rules and, for any time being, much is taken for granted.

But the statement on 'mono' as a better method lies in the realm of science. Won't repeat myself, see post #615.

Just in case: why bother with mixing down stereo to mono? Why not record mono and let a sound engineer optimize the recording for mono playback? A more sound approach, wasn't it? Never mind.

 

[bmc0]

By your logic, noise shaping is of no value just the same. Right?

That in no way follows from what I wrote.

In post #603, I simply asked for corroborating evidence that a small blip in individual ISO 226:2003 contours is actually relevant to a different case with different stimuli in a different acoustic environment than that under which the equal loudness data were obtained. If you disagree with that, I don't know what to tell you. Assuming data obtained under entirely different conditions and intended to answer an entirely different question to be relevant and directly applicable is certainly not my definition of good science.

I further suggested that the particular inference in question—crossover audibility, specifically in the midrange—may be a misapplication of the equal loudness data, but I made no concrete claim that it actually is. Again, I asked for supporting evidence.

I'd posit that how the contours change with level may perhaps be more relevant than the shape at any single level. The contours bunch together at low frequencies, meaning that a small change in SPL is perceived as a relatively large change in loudness. From Olive's loudspeaker preference model, it is apparent that bass is very important to sound quality of music despite the fact that tones in this frequency range must be much higher SPL than higher frequencies to be perceived as equally loud. Is this at all relevant to crossover audibility in the midrange? I don't know, but it may be worth mentioning that the ISO 226:2003 model indicates reduced sensitivity to SPL changes from 1.25kHz to 6.3kHz compared to 1kHz and below.