Speaker Testing: why mono is better

[Koeitje]

Stereo just creates a masking effect for issues with the loudspeaker.

 

[amirm]

I've noticed that the speakers are not toe'ed in.

Much of the testing is done in mono and listener sits on axis of speaker.

 

[Grumple]

It seems to me that evaluating in mono is the only logical setup. I don't see how people could intuit that stereo would be better. Maybe the stereo people could provide some examples of stereo sound that occurs outside of hifi? Because my intuition of the natural world, including people and their musical instruments, is that they are all in mono. Or am I missing something?

 

[ctrl]

There is no such science.

There is evidence that it makes a difference to listen to and evaluate a loudspeaker quasi monophonic as a center speaker like Harman and to listen to a pair of loudspeakers in stereo.


For instance, Dr. Toole lists one reason in his book "Sound Reproduction". The cause are comb filter effects due to the different travel times of direct sound to the ears and direct sound interference from the two speakers.


The influence of this interference is not insignificant:

The listed comb filter effect can be easily simulated by adding two ideal loudspeakers with 6dB sound pressure difference and a delay of 0.27ms (as described in Toole's text).

simu:

So, the next source for a timbre change between monophonic versus stereophonic is the AES paper by Bernd Theiss "Localization Experiments in Three-Dimensional Sound Reproduction" (1996).
Among other things, it states ("the reference" is the center speaker):

Own experience (unscientific, anecdotal):
The biggest beginner's mistake I made as a hobby speaker designer was trying to perfectly tune a speaker monophonic (quasi as center speaker) with a mono signal (down-sampled from stereo).

If the speaker sounded perfect monophonically, it later sounded way too bright when listening in stereo and was unbearable at high sound pressure levels.

When I develop a loudspeaker today, I still start with monophonic listening (the advantages have already been listed here in the thread by Amir and others), but the loudspeaker is placed sideways and the test music is heard in stereo.
There I agree one hundred percent with @amirm's listening setup.

Yet, by listening and tuning monophonically, I am not able to produce the perfect stereo tuning. Of course, this may be due to my own inability, but even with the described monophonic test setup, a timbre change can still be detected during the transition to stereophonic.

For the last 10% sound quality I have to do the tuning in stereophonic.

A somewhat more detailed discussion of the sources, can be found here and here. However, I mix the topics "stereophonic interference" and "lateral reflections" in the thread.

 

[pozz]

of course you can make the room response so bad that it drowns everything else

I don't disagree with your line of thought, particularly what's possible to do with the input signal. Where I think we differ is in our perspective of the basic situation of listening. IMO normal room response is enough to drown phase error, and that listening to regular multiway speakers and human head movement into account means that we are never in a situation where the acoustic response at our ears preserves linear phase.

If you also consider that correlated signals are localized if they arrive within a certain interval, and perceived as diffuse or enveloping if they are uncorrelated, that suggests pure phase error is subject to averaging anyway, which would capture how open baffle speakers work, for example. For low frequencies especially, where phase error is usually largest, we are used to soundfields with a lot of variation, so it wouldn't make sense that phase would have any significant impact as long as time of flight and amplitude at the listening position have been corrected (approximately rather than full inversion).

 

[dorirod]

I don't get it. What's so hard to understand. Let me put this in easy to understand words

2 bad speakers in stereo= maybe ok sound

2 very good speakers in stereo = very good sound

We use mono listen test to determine if one singular speaker is good. If one is good enough why wouldn't a stereo setting be very good?

That's just easy logic.

How do you get to the distinction/metric between "ok sound" and "very good sound"? Based on the paper mentioned on the video, I would say the "very good" and "bad speakers" were at the same double-blind judged quality. I think it's a fair question to ask since based on what the paper is saying, to determine the best designed speaker you would need to listen in mono, but to determine the best sounding stereo speakers, listening in mono may not be necessary or sufficient.

Said another way, since listening is done mainly in stereo, the more important factors or flaws may not be those revealed only by mono listening.

BTW, I do appreciate Amirm listening in mono and providing those results, *because* other reviewers do not do that. However, I wouldn't go so far as to encourage *all* reviewers to *only* evaluate in mono. *Ideally*, they should be evaluated in both mono and stereo, but again I really appreciate the insight that Amir provides considering it's perfectly obvious that it would double efforts and costs, so not really worth pursuing in his case. I just don't agree with discarding stereo evaluation since in my opinion, the paper itself is showing valuable data, namely whether you can determine preference in stereo listening between two pairs of speakers. As Racheski mentioned, if you've discovered that a $2000/pr of speakers sounds the same in stereo as a $200/pr of speakers, maybe you go with $200/pr knowing that they're not as good in mono, but perfectly acceptable in stereo. Or maybe you move on and evaluate other criteria/distincions such as build, support, size, efficiency, etc. If you are wanting to spend $2000/pr though you would want to know that they're better designed.

Or maybe, there is a test/set of conditions that would show in stereo listening that speakers AA,E, are clearly better than BB, and then should make sure stereo listening includes those conditions if they would show up in normal usage conditions. But if there is no such test/condition, then I don't see how you can throw away that data point. It seems pretty significant to me that they couldn't tell BB from AA, E in stereo.

 

[pozz]

r maybe, there is a test/set of conditions that would show in stereo listening that speakers AA,E, are clearly better than BB, and then should make sure stereo listening includes those conditions if they would show up in normal usage conditions. But if there is no such test/condition, then I don't see how you can throw away that data point. It seems pretty significant to me that they couldn't tell BB from AA, E in stereo.

Look up Francis Rumsey. One of the main researchers for multichannel audio.

He did blind testing using pro (studio engineers, etc.) and ordinary listeners. Main result was that tonal balance and timbral accuracy (so frequency response) was the leading criterion for mono, two channel and multichannel listening. All of the other aspects that come with increasing channels, sense of spaciousness, envelopment, distance, came in way lower in importance. Pros were more sensitive to those other aspects, while some ordinary listeners others barely noticed them if they sensed timbre was off.

Speaker design considerations start changing as you increase the number of channels since you have to think about how to place and integrate them in the room. That's about it. Bottom line is that if you have good information about a single speaker you can follow through with other situational decisions about setup and so forth.

Remember that the preference testing is done in blind, controlled conditions. Setting those up is not possible for any reviewer unless they spend significant effort in creating a space, determining standard test protocols and so forth. No matter what a reviewer reports, if they listen in ordinary circumstances, it has only glancing relevance. Measuring a speaker and then listening in mono amounts to a reasonable check of the data. If you add channels, you are not only complicating setup but no longer assessing a known starting output since the effects of acoustical crosstalk are not predictable or intuitive.

You, your room and speakers form one system. Adding speakers to a room makes it more and more a system-level assessment than a speaker-level assessment, the latter being the goal when it comes to reviews, not the former.

 

[dasdoing]

[OT]
It is actually very simple to implement, You do not need three source channels for that. Just set L and R according to the simple formulas L' = L - R/2 and R' = R - L/2. The center channel is simply the average and can be obtained passively at the power amp input.

[/OT]

this kind of solutions never made sense to me. they don't work for music (are intended for movie center extraction). the center channel should be used as panning, the same way it works in 2 channel stereo. I tried something like this here (without realy having much knowhow on how realizing something like this): https://www.audiosciencereview.com/...e-aka-phantom-center.13923/page-2#post-439738
also I asked about a musical solution here, and got some answer which are a little over m head: https://www.avsforum.com/threads/upmixing-stereo-music-into-true-3-channel-stereo-not-3-0.3164657/

 

[dorirod]

Look up Francis Rumsey. One of the main researchers for multichannel audio.
You, your room and speakers form one system. Adding speakers to a room makes it more and more a system-level assessment than a speaker-level assessment, the latter being the goal when it comes to reviews, not the former.

Thanks, yes that makes sense. The issue remains though, I have a set of speakers, I know how they measure. I know which speakers measure better than them or have a better looking FR. Is there an objective way I can determine that buying a replacement set of speakers will actually result in better listening experience (in stereo)? It seems that the answer is no, or rather this stereo (or higher number of channels) smearing is doing enough to actually reduce the impact of differences in FR to make determining which speakers sound better a lot more challenging. After all, I can't gather a list of 10 speaker sets, drag Amir to my house, make him tell me which ones to pick (oh and do that every time I move or my wife decides to arrange the furniture The alternative I guess is to just buy a good measuring speaker set and stop worrying about whether I can get something better. I'll print the ASR review and tape it to the back to let people (and myself) know if they argue with me.

 

[Pdxwayne]

I wish to thank Amir for testing in mono. The video explains why testing in mono is a great way to determine how well a pair of speakers would do in stereo.

The video also validated my observations in my friend's 5.4.1 setup.....All those shortcomings observed in mono testing went away when one plays stereo music with 9 speakers and 1 sub (it was professionally installed and calibrated to get all speakers to work together in the room, optimized for the listening position). In fact, it was really amazing experience listening to my friend's setup.

 

[q3cpma]

The logic fault lies in the argument-less transformation of observed property mono => more spread in MOS into proposition mono => more discernable sound quality through more spread in MOS => more discernable sound quality as if the MOS couldn't be influenced by any other consequence of the passage from stereo to mono. For example, what would be the result with two "perfect" speakers (flat on axis, CD down to a similar frequency, low nonlinear distortion) with only horizontal directivity width varying?
Another problem I've already raised here: why do these flaws matter in any way if they're not perceived in stereo?

 

[pozz]

Is there an objective way I can determine that buying a replacement set of speakers will actually result in better listening experience (in stereo)? It seems that the answer is no, or rather this stereo (or higher number of channels) smearing is doing enough to actually reduce the impact of differences in FR to make determining which speakers sound better a lot more challenging.

I think you're asking about how to quantify how much of an improvement you'll get. I think that if you have an active design, flat on axis, constant or close to constant directivity and smooth sound power you are close to hitting the limit. The choices that remain are SPL, types of directivity and low frequency extension (which should be supplemented by separate subwoofers anyway). But, for example, I suspect that for cardioids like the D&D 8c some of the excellence of the speaker can be matched by having a listening room constructed to severely damp front wall reflections and avoid SBIR (e.g., through soffit mounting).

 

[pozz]

Another problem I've already raised here: why do these flaws matter in any way if they're not perceived in stereo?

Because of signal correlation. When you play music in stereo (timing/intensity differences), you need each speaker to individually reproduce its part of the signal correctly. In the broad sense, sure, the flaws are averaged out, but perceptually they are still matter when you consider arrival times and spectra. Especially above the transition region, where direct sound of the speaker dominates.

 

[PeteL]

It seems to me that evaluating in mono is the only logical setup. I don't see how people could intuit that stereo would be better. Maybe the stereo people could provide some examples of stereo sound that occurs outside of hifi? Because my intuition of the natural world, including people and their musical instruments, is that they are all in mono. Or am I missing something?

I don’t disagree with testing in mono, but your rationale for doing so does not apply (nor make sense) In the “natural world” sound can come from anywhere, 360 degree x 360 degree. Listening in mono means not being able to localise the source in space. Stereo, along with various miking techniques and fx allows to create a space where you can localise various sources. It’s based on the fact that you have two ears. There is nothing natural to listen in mono.

 

[Ron Party]

On another forum Duke LeJeune recently posted some interesting comments on this topic. I'm hoping he will read this as an invitation to join in this great discussion.

 

[pozz]

I don’t disagree with testing in mono, but your rationale for doing so does not apply (nor make sense) In the “natural world” sound can come from anywhere, 360 degree x 360 degree. Listening in mono means not being able to localise the source in space. Stereo, along with various miking techniques and fx allows to create a space where you can localise various sources. It’s based on the fact that you have to ears. There is nothing natural to listen in mono.

Playback through speakers is artificial from the outset. Note that, in nature, it is impossible to have two of the same sounds coming from two different locations. But two channel stereo routinely plays sounds in just that manner to produce the phantom center. And on top of that takes all of the ambient directional reverberation information and confounds it to emanate from two frontal points.

What you are describing is not regular stereo, but binaural audio, which has to be individually tailored to a specific person with microphones at their eardrums (or synthesized as such), not any general collection of intensity/timing differences. True soundfield accuracy is found by using spatial sampling, which is reproduced in playback through wavefield synthesis systems, which aren't commercially available (they exist as research projects in certain universities).

Localization refers to your innate ability to perceive sounds in a space. The fact that the source is a single speaker does not in any way impede your ability to localize them to that speaker. If you are interested you can look up work on monaural vs. binaural localization. Your ability to perceive height for example is mostly monoaural.

 

[q3cpma]

Because of signal correlation. When you play music in stereo (timing/intensity differences), you need each speaker to individually reproduce its part of the signal correctly. In the broad sense, sure, the flaws are averaged out, but perceptually they are still matter when you consider arrival times and spectra. Especially above the transition region, where direct sound of the speaker dominates.

By these differences, I mean only those that could possibly be hidden in stereo. Are there any hints to the question of their nature?

 

[Todd74]

Not sure I understand the imaging comment where you say [paraphrasing] “Look at all the separation here in Mono, and then look how it’s all grouped closely together in Stereo”.

Here’s the graph.

That’s not how I perceived the chart. I perceive it as— in Stereo, there’s a drastic difference compared to Mono, with the 2nd speaker now SURPASSING the first one when it was notably behind the leader in Mono, and with the 3rd speaker matching the leader despite there being a massive difference in Mono. To me it’s like saying the Mono tests don’t carry as much weight if we can’t pick up their differences in stereo.

That’s fairly significant IMO, especially if we’re talking about a speaker that’s hypothetically less money. I mean, isn’t that the goal? —not to find the speaker that performs best in mono but to find a pair that can sound as good in stereo as the great testing and potentially more expensive speaker does in mono? It kinda feels like you’re suggesting that it’s “cheating” to have the inferior speaker sound as good in stereo as the other speaker performs in mono.

I mean, what if speaker 1 in the graph is a $5000 Revel, speaker 2 is a $3000 Dynaudio, and speaker 3 is a $1500 brand X? Are you suggesting that it’s more important to choose my speaker based on the significant difference in performance in Mono than in Stereo where the less expensive speakers suddenly look a lot more attractive?

 

[abdo123]

Not sure I understand the imaging comment where you say [paraphrasing] “Look at all the separation here in Mono, and then look how it’s all grouped closely together in Stereo”.

Here’s the graph.

That’s not how I perceived the chart. I perceive it as— in Stereo, there’s a drastic difference compared to Mono, with the 2nd speaker now SURPASSING the first one when it was notably behind the leader in Mono, and with the 3rd speaker matching the leader despite there being a massive difference in Mono. To me it’s like saying the Mono tests don’t carry as much weight if we can’t pick up their differences in stereo.

That’s fairly significant IMO, especially if we’re talking about a speaker that’s hypothetically less money. I mean, isn’t that the goal? —not to find the speaker that performs best in mono but to find a pair that can sound as good in stereo as the great testing and potentially more expensive speaker does in mono? It kinda feels like you’re suggesting that it’s “cheating” to have the inferior speaker sound as good in stereo as the other speaker performs in mono.

I mean, what if speaker 1 in the graph is a $5000 Revel, speaker 2 is a $3000 Dynaudio, and speaker 3 is a $1500 brand X? Are you suggesting that it’s more important to choose my speaker based on the significant difference in performance in Mono than in Stereo where the less expensive speakers suddenly look a lot more attractive?

That’s not how science works, Speaker 3 here is a control.

Meaning that it’s objectively and subjectively subpar.

The stereo testing shows two things:

1) people couldn’t tell that the bad speaker is bad

2) people couldn’t distinguish the quality difference between any of the 3 speakers in a STATISTICALLY SIGNIFICANT WAY. The difference in score between the the speakers is so low it could be just by chance.

So basically any data out of that stereo test is pointless because it’s not significant. => stereo testing is bad.

 

[ashegedyn]

Not sure I understand the imaging comment where you say [paraphrasing] “Look at all the separation here in Mono, and then look how it’s all grouped closely together in Stereo”.

Here’s the graph.

That’s not how I perceived the chart. I perceive it as— in Stereo, there’s a drastic difference compared to Mono, with the 2nd speaker now SURPASSING the first one when it was notably behind the leader in Mono, and with the 3rd speaker matching the leader despite there being a massive difference in Mono. To me it’s like saying the Mono tests don’t carry as much weight if we can’t pick up their differences in stereo.

That’s fairly significant IMO, especially if we’re talking about a speaker that’s hypothetically less money. I mean, isn’t that the goal? —not to find the speaker that performs best in mono but to find a pair that can sound as good in stereo as the great testing and potentially more expensive speaker does in mono? It kinda feels like you’re suggesting that it’s “cheating” to have the inferior speaker sound as good in stereo as the other speaker performs in mono.

I mean, what if speaker 1 in the graph is a $5000 Revel, speaker 2 is a $3000 Dynaudio, and speaker 3 is a $1500 brand X? Are you suggesting that it’s more important to choose my speaker based on the significant difference in performance in Mono than in Stereo where the less expensive speakers suddenly look a lot more attractive?

Regarding preference testing isn't it difficult to make these judgements on absolute differences? Preference scaling is relative. Even if you give a pseudo measure it is still relative. It is not like a frequency response graph, I don't think we know the details of the sample size and statistics generated or the confidence intervals. Without that it would be difficult to make a rational judgement on some of these issues raised. It would be great if the papers were available without the paywall. What is clear is that mono testing shows greater differences than if we do stereo testing. It seems reasonable to me from the study as well as common sense that if you test the individual speaker you will also make a rational judgement if the speaker is to be used in stereo. I think the weight of showing that stereo testing is better would need some sort of study rather than just objections with no data. If people are objecting that stereo testing is needed, are there studies from AES to corroborate this?