AVS Forum banner

How to Choose a Loudspeaker -- What the Science Shows

550K views 6K replies 267 participants last post by  Rex Anderson 
#1 ·
Choosing a loudspeaker may be the biggest challenge for music and home theater lovers. There are countless brands from which to choose, and even more claims and counter-claims. Since the room has such a profound impact on the sound of a loudspeaker at lower frequencies, and it is impossible to listen in a blind test at an audio store, if they can find one, there is little that an audiophile can do to make a rational decision. Fortunately, science has come to the rescue with a set of measurements that have been proven to demonstrate an extremely close correlation with sound quality, as based on carefully controlled double-blind listening tests. This group of measurements have been adopted as the industry standard for measuring loudspeakers, as ANSI/CEA-2034-A. https://standards.cta.tech/apps/group_public/project/details.php?project_id=165

Contradicting the oft-repeated claim that choosing a loudspeaker is a very personal choice, research has proven that regardless of age, culture, or listening experience, all people with nominally normal hearing generally agree on which speakers sound better than others. Indeed, there is a universal definition of what sounds good. http://www.aes.org/e-lib/browse.cfm?elib=12794 and https://secure.aes.org/forum/pubs/conventions/?elib=12847

In this thread, we will publish the results of these measurements. In addition, we will discuss their correlation to double-blind listening tests, http://seanolive.blogspot.com/2008/12/part-3-relationship-between-loudspeaker.html as well as publishing the results of formal listening tests, when available. We will add measurement results as they become available. The intention of this thread is for it to be reality-based, and to inform and discuss loudspeaker measurements and listening tests. The papers that really started it all are now available for free from the Audio Engineering Society here: http://www.aes.org/e-lib/browse.cfm?elib=5276 and here: http://www.aes.org/e-lib/browse.cfm?elib=5270
 
See less See more
#3 ·
An explanation of ANSI/CEA-2034-A ("Spinorama") Data

Years of experiments and studies conducted by Harman Research Scientists https://www.routledge.com/Sound-Rep...f-Loudspeakers-and/Toole/p/book/9781138921368 and https://secure.aes.org/forum/pubs/journal/?ID=524 has led to a series of 70 measurements which, after post-processing, provide an excellent indication of a loudspeaker's sound quality. This group of measurements, dubbed "Spinoramas," have proven to be so meaningful in characterizing the sound of loudspeakers in rooms that they have been codified as the industry standard method with which to measure loudspeakers as ANSI/CEA-2034-A. https://www.techstreet.com/mss/products/preview/1868536

These 70 measurements, measured as a sphere surrounding the loudspeaker, are post-processed to provide six curves that, taken together, correlate closely with carefully controlled double-blind listening tests. http://www.aes.org/e-lib/browse.cfm?elib=3833



On-axis
The on-axis frequency response is and has been the most commonly used loudspeaker measurement. However, when used by itself it is a questionable indicator of the speaker's sound quality. If it is poor, the speaker will not sound good, but the on-axis response can be good, while the speaker does not sound good. The measurements described below are necessary to more fully characterize the sound of a loudspeaker. It is no wonder that speaker measurements have had such a poor reputation regarding their relationship to sound quality, given that they have generally been of the simple on-axis variety.

Listening Window
This updated measure of a speaker's direct sound output is composed of an average of nine frequency response measurements on-axis, at ± 10° vertical, and at ± 30° horizontal off-axis angles. Since this measurement is a spatial average, it attenuates small fluctuations that are merely the result of acoustical interference that is far less sonically significant than it appears to be based upon a single on-axis measurement. These relatively acoustically-benign local interference phenomenon change with tiny adjustments in the microphone position, and their visual distraction can mask actual performance problems--resonances in particular. Resonances are "bumps" in the response that can be significant sonic problems. Since resonances tend to radiate over a wide area, they will remain visible in the spatially-averaged Listening Window response, while insignificant fluctuations that would change with tiny microphone position adjustments are suppressed. Loudspeakers with smooth and flat listening window responses tend to excel in Harman's Double-blind Listening Tests.

First, or Early Reflections
Most of the sound we hear in rooms is reflected. The second-loudest sound, after the direct sound, is the first reflected sound from the loudspeakers. In fact, Harman research has discovered that the first reflection from side walls, both from the wall adjacent as well as the opposite side wall are critically important. The acoustic output of a loudspeaker far off-axis horizontally is very significant, and should match the response of the Listening Window as much as possible. This goal is technically challenging, but is essential for optimum timbre, as well as to provide a sense of seamless coherency. Revel's waveguides, along with optimum engineering choices such as crossover points and slopes, relatively small midranges and tweeters that can be safely used to lower frequencies than typical designs contribute to far off-axis responses that are close to the direct sound, as seen in Listening Window measurements. http://www.aes.org/e-lib/browse.cfm?elib=6079

Sound Power
Sound Power is composed of the weighted average of all 70 measurements. Each measurement is weighted to properly represent an area of a sphere. Sound Power is a measure of the total sound radiated by a loudspeaker. One of its uses is to detect resonances, since response aberrations seen in both the Sound Power and other curves are likely true resonances. Harman research has determined the threshold of audibility of resonances. All Revel speakers are designed to keep resonances below the threshold of human audibility using the most sensitive stimulus. http://www.aes.org/e-lib/browse.cfm?elib=5163

Directivity Indices
The Directivity indices describe how the speaker's radiation changes as a function of frequency. The Sound Power Directivity Index is defined as the difference between the Listening Window curve and the Sound Power, while the First Reflection Directivity Index utilizes the Listening Window curve and the First Reflections curve. Both curves should be smooth and change gradually. A DI of 0 dB indicates an omnidirectional speaker, while a larger DI indicates greater directivity.
 
#2,451 ·
Years of experiments and studies conducted by Harman Research Scientists https://www.routledge.com/Sound-Rep...f-Loudspeakers-and/Toole/p/book/9781138921368 and https://secure.aes.org/forum/pubs/journal/?ID=524 has led to a series of 70 measurements which, after post-processing, provide an excellent indication of a loudspeaker's sound quality. This group of measurements, dubbed "Spinoramas," have proven to be so meaningful in characterizing the sound of loudspeakers in rooms that they have been codified as the industry standard method with which to measure loudspeakers as ANSI/CEA-2034-A. https://www.techstreet.com/mss/products/preview/1868536

These 70 measurements, measured as a sphere surrounding the loudspeaker, are post-processed to provide six curves that, taken together, correlate closely with carefully controlled double-blind listening tests. http://www.aes.org/e-lib/browse.cfm?elib=3833



On-axis
The on-axis frequency response is and has been the most commonly used loudspeaker measurement. However, when used by itself it is a questionable indicator of the speaker's sound quality. If it is poor, the speaker will not sound good, but the on-axis response can be good, while the speaker does not sound good. The measurements described below are necessary to more fully characterize the sound of a loudspeaker. It is no wonder that speaker measurements have had such a poor reputation regarding their relationship to sound quality, given that they have generally been of the simple on-axis variety.

Listening Window
This updated measure of a speaker's direct sound output is composed of an average of nine frequency response measurements on-axis, at ± 10° vertical, and at ± 30° horizontal off-axis angles. Since this measurement is a spatial average, it attenuates small fluctuations that are merely the result of acoustical interference that is far less sonically significant than it appears to be based upon a single on-axis measurement. These relatively acoustically-benign local interference phenomenon change with tiny adjustments in the microphone position, and their visual distraction can mask actual performance problems--resonances in particular. Resonances are "bumps" in the response that can be significant sonic problems. Since resonances tend to radiate over a wide area, they will remain visible in the spatially-averaged Listening Window response, while insignificant fluctuations that would change with tiny microphone position adjustments are suppressed. Loudspeakers with smooth and flat listening window responses tend to excel in Harman's Double-blind Listening Tests.

First, or Early Reflections
Most of the sound we hear in rooms is reflected. The second-loudest sound, after the direct sound, is the first reflected sound from the loudspeakers. In fact, Harman research has discovered that the first reflection from side walls, both from the wall adjacent as well as the opposite side wall are critically important. The acoustic output of a loudspeaker far off-axis horizontally is very significant, and should match the response of the Listening Window as much as possible. This goal is technically challenging, but is essential for optimum timbre, as well as to provide a sense of seamless coherency. Revel's waveguides, along with optimum engineering choices such as crossover points and slopes, relatively small midranges and tweeters that can be safely used to lower frequencies than typical designs contribute to far off-axis responses that are close to the direct sound, as seen in Listening Window measurements. http://www.aes.org/e-lib/browse.cfm?elib=6079

Sound Power
Sound Power is composed of the weighted average of all 70 measurements. Each measurement is weighted to properly represent an area of a sphere. Sound Power is a measure of the total sound radiated by a loudspeaker. One of its uses is to detect resonances, since response aberrations seen in both the Sound Power and other curves are likely true resonances. Harman research has determined the threshold of audibility of resonances. All Revel speakers are designed to keep resonances below the threshold of human audibility using the most sensitive stimulus. http://www.aes.org/e-lib/browse.cfm?elib=5163

Directivity Indices
The Directivity indices describe how the speaker's radiation changes as a function of frequency. The Sound Power Directivity Index is defined as the difference between the Listening Window curve and the Sound Power, while the First Reflection Directivity Index utilizes the Listening Window curve and the First Reflections curve. Both curves should be smooth and change gradually. A DI of 0 dB indicates an omnidirectional speaker, while a larger DI indicates greater directivity.
Whenever I come back to this thread it's clear to me that there's a lot of gold in a few key posts most of which I don't yet full understand. One nagging question is whether or not there are situations in which highly directive speakers are preferred over speakers with "better" listening window measurements. Or, put differently, are their situations in which narrow dispersion wave-guides work better than wide dispersion wave-guides? I'm thinking here of small rooms in which first reflections are close in time to direct sounds and, thus, minimising these reflections might be better.
 
#4 · (Edited)

Attachments

#7 ·
Thank you Gooddoc. We do not have the personnel available to put speakers through our testing by request. However, feel free to make requests, and if we have them available, I will add them. I will be adding both Revel and other competitors as time permits, so you might find your request fulfilled without even asking.
 
#9 ·
would be awesome if we had the measurements of the speakers after the users got to adjust eq...ya know to see what they really liked the sound to be
 
#11 ·
Since the 228BEs show better measurements through most of the FR range, particularly off-axis compared to the Salon 2, yet the Salon2 was preferred on average, should the explanation primarily be that the highs are rolled off (8K+?) in comparison? It seems to me the old NRC research showed a preference for rolled off highs for most people, so the only downside to the PerformaBe series is the tweeter is a little hot in the higher frequencies?
 
#50 ·
I don't know where the idea of the rolled off highs came from. I cannot remember ever saying it verbally or in print - I don't believe it. This in spite of the fact that the earliest listening tests used LPs that are not at all well behaved at any frequency, much less very high frequencies.

The evidence is published, and summarized in the attached figure: Figure 13.1 from my book. All that has happened in 30 years is that we now have better engineered drivers and systems. The performance targets are the same. FYI the "early refections" curves are very good predictors of steady-state room curves in normally reflective listening rooms.

It will be noted that the "idealized" steady-state room curve has eliminated the common dip around 2kHz caused by a slight directivity mismatch at the woofer/mid-to-tweeter crossover in most forward-firing systems. Ideally it should not be there, but because humans place the greatest importance on the direct sound for judgments of timbre it is not a major factor. Also, the off-axis radiation that includes the dip is delayed and attenuated compared to the direct sound and two ears and a brain figure it out - "dumb" omnidirectional mics do not. Equalizing the dip out of in-room steady-state measurements is a mistake - it degrades the direct sound.

The origin of the "idealized" curve is explained in Figure 12.4 in the book. It is the kind of steady-state room curve that is measured for loudspeakers that rate high in double-blind tests. It should not be assumed that equalizing a flawed loudspeaker to match the curve is an assurance of equally good sound. Directivity flaws in loudspeakers cannot be corrected by equalization, and resonances are difficult to identify in room measurements - it is in the book.
 

Attachments

#12 ·
science probably says that those that can afford 10k + speakers cant hear abover 15khz if lucky...
 
#13 ·
I'm surprised by the choice of Magico to compare to the Revels, it seems most people in the high end community prefer TAD Evolution or KEF Blades over the Magicos.

On the other hand, I sort of understand it from a marketing perspective, since KEF and TAD are using concentric designs which have much better polar response measurements than a typical multi-way dome speakers. Perhaps it would be harder to sell promotional material for Revel when the speakers you are comparing to measure better than what you are selling using your preferred measurement techniques...
 
#16 ·
On the other hand, I sort of understand it from a marketing perspective, since KEF and TAD are using concentric designs which have much better polar response measurements than a typical multi-way dome speakers. Perhaps it would be harder to sell promotional material for Revel when the speakers you are comparing to measure better than what you are selling using your preferred measurement techniques...
I guess we'll just have to see if any comparos or measurements have been done. It was clearly stated there was more to follow.
 
#15 ·
Ah, I think I see why the Salon2 might be preferred. It's got a large dip at 2KHz in the off-axis measurements as well. I know many high end headphone companies (like HFM) intentionally engineer 2KHz dips with high end electrostatic headphones to give a larger sense of staging, and this is also where British brands traditionally put a frequency dip (also known as the BBC dip) because their own research found people often prefer a dip in that region.
 
#19 ·
after diving head first into headphone game I hear alot more...nothing close to my stereo fullrange home setup...but way easier to focus and listen and hear stuff different.
 
#22 ·
if people dont like a 2k+ bookshelve its cause they picky/have a sound they after...I havent heard a poor bookshelf yet at 2 grand
 
#25 ·
I posted a published paper that dismisses the op's statement by the guy that wrote the paper that op is using...confusing.
 
  • Like
Reactions: BufordTJustice
#27 ·
well I guess so...most people know what they like and dont need science to tell them
 
  • Like
Reactions: Scotth3886
#28 ·
Food companies and restaurants spend enormous amounts of time and money to determine the likes and dislikes of consumers via taste testing that is conducted using a scientific method. The data from those findings are used to develop new or alter preexisting products. The intent is to have a product that is favored by the majority. If this well known and accepted discipline is used for taste, why not do the same for hearing?

It isn't about what you like or what I like. Rather, it is about finding out what the majority of consumers like so you can maximize product adoption. If you build a speaker according to the data you gather, you have a statistical probability that the when the consumer hears that speaker they will not only like it but buy it. The science simply shows what the majority of people prefer. It doesn't mean that everyone will like it. What Harmon does is smart business and my ears agree with their findings. I guess I fall in the statistical majority although my mother always told me I was "special";)
 
#29 ·
many, many smart people have lost their shirts/bank rolls on science...I thought we all could agree flat sounds like shiit but guess not
 
#32 ·
I would assume that is why there is a Harmon curve because their findings agree with you. I want a speaker that measures flat because it shows accuracy. I will always season to taste regardless of what others tell me. My money, my home, my sound. I find the science interesting, but much of it is over my head. I didn't buy JBL for the science, but I am confident their science had a lot to do with me ending up as a customer. Likewise, McDonalds french fries are my favorite. I am sure there was some science that took place behind the scenes that keeps me coming back.
 
#31 ·
I would like op to show measurements dont mean jack diddly
 
#33 ·
well not getting too far over my head...revel salon 2's were favored over a better measuring jbl m2...so that alone throws this topic underwater
 
#53 ·
Even though John et al. took great pains for a blind comparison, I wouldn't make such a well defined and artless conclusion especially when it's used as evidence to wholesale dismiss peer-reviewed studies. Preferences weren't overwhelming and "better measuring" wasn't in all aspects of performance. There might be something to the "BBC dip" for off-axis while maintaining a non-dip on axis.

Yet I bought M2s after that contest when I could have bought Salons. I must be the dull ax in the shed. Do you think the results may have been different in a home theater venue vs mono audio? Both of those speakers are world class. I don't think there was a loser in the room. With that said, I will probably put Revels downstairs where I listen to audio more. I do prefer a dome tweeter for music most of the time but I want dynamics in my theater room.
 
#35 ·
yup and I have run many sweeps/test tones...but what is weird is that Im deaf at 15khz on my system but 17khz in pro sound booth/dr office.
 
#276 ·
If your deaf at 15KHz, why do you care if the highs are rolled off or not?
 
  • Like
Reactions: zencarver
#36 ·
well maybe not weird...but i aint no spring chicken celebrating 22khz anymore
 
#37 ·
and Im not getting mad at not hearing air on my 5 grand speakers so I can get madder at not hearing air on 20 grand speakers
 
#43 ·
I posted a published paper that dismisses the op's statement by the guy that wrote the paper that op is using...confusing.
How are you arriving at that conclusion? The link you posted, at least according to the introduction(couldn't read it all), was comparing speaker response to headphone response. The OP in his first post linked a blog by Sean Olive, one of the authors of the paper in your link, where he says:

There are clear visual correlations between listeners' loudspeaker preferences and the set of frequency graphs. Both trained and untrained listeners clearly preferred the loudspeakers with the flattest, smoothest and most extended frequency response curves...
and

It is both satisfying and reassuring to know that both trained and untrained listeners recognize and prefer accurate loudspeakers, and that the accuracy can be characterized with a set of comprehensive anechoic measurements
 
#46 · (Edited)
I’ve been curious about Harman’s research relating to the following variables.

1) Take a speaker with ideal/accurate anechoic measurements and place it in the extremes of a live room and a dead room. Assuming the live vs. dead room will impact the frequency response at the listener’s ears, and assuming that the brain prefers a certain frequency response, could that same ideal/accurate speaker (from anechoic measurements) produce a different frequency response at the listener’s ears, based on the reflective and absorptive characteristics of the room? If the brain prefers a certain frequency response curve, how could the same ideal/accurate speaker (from anechoic measurements) be preferred in both the dead room and live room?

Said another way, if the brain prefers a specific curve at the listening position, is it possible that a non-ideal/accurate speaker (from anechoic measurements), could deliver closer to the preferred curve because of the absorptive or reflective characteristics of the room?

2) Now add in the variable of one's ears and the high frequency hearing loss that males of my age typically have :))). If the brain prefers a certain frequency response, but the ears have high frequency loss, could one prefer speakers (especially in an absorptive room) that are not ideal/accurate (from anechoic measurements) because they “compensate” for hearing loss (and room absorption), and deliver closer to the ideal/flat response that the brain prefers?

Not trying to start any arguments here – just looking to learn and discuss.


-
 
#48 ·
I’ve been curious about Harman’s research relating to the following variables.

1) Take a speaker with ideal/accurate anechoic measurements and place it in the extremes of a live room and a dead room. Assuming the live vs. dead room will impact the frequency response at the listener’s ears, and assuming that the brain prefers a certain frequency response, could that same ideal/accurate speaker (from anechoic measurements) produce a different frequency response at the listener’s ears, based on the reflective and absorptive characteristics of the room? If the brain prefers a certain frequency response curve, how could the same ideal/accurate speaker (from anechoic measurements) be preferred in both the dead room and live room?

Said another way, if the brain prefers a specific curve at the listening position, is it possible that a non-ideal/accurate speaker (from anechoic measurements), could deliver closer to the preferred curve because of the absorptive or reflective characteristics of the room?

2) Now add in the variable of ones ears and the high frequency hearing loss that males of my age typically have :))). If the brain prefers a certain frequency response, but the ears have high frequency loss, could one prefer speakers (especially in an absorptive room) that are not ideal/accurate (from anechoic measurements) because they “compensate” for hearing loss (and room absorption), and deliver closer to the ideal/flat response that the brain prefers?

Not trying to start any arguments here – just looking to learn and discuss.


-
Read this carefully https://www.amazon.com/Sound-Reproduction-Psychoacoustics-Loudspeakers-Engineering/dp/113892136X/ref=pd_lpo_sbs_14_t_0?_encoding=UTF8&psc=1&refRID=BQSVBMH9KGG28WWH1EV5

The answers are in there
 
#47 ·
You guys who keep referring to the "OP" do realize who the OP is in this case, right? He's not just some generic know nothing poster spewing nonsense.
Yes, and Kevin's time and participation here is much appreciated. You could make the case that he, Sean Olive, and Floyd Toole among many others are our version of rock stars. :)


Not trying to start any arguments here – just looking to learn and discuss.
Exactly what this forum is for. There are a lot of variables in the playback chain, so if you know going in what the speakers are doing as expressed by anechoic measurements, you can, through placement, setup, room treatments, EQ, etc..., get the response you want in whatever room you're in.
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top