Alternative perspective on the Harman loudspeaker preference curve.

[youngho]

Which of these three curves is considered to be the most objectively accurate?

Short answer: none.

Longer answer: they were subjectively derived, and there was a high degree of variance overall (I don't have sufficient data, but if I'm correct about the three female listeners, then they were actually quite consistent with little variance; similarly, if I guess the two listeners who preferred the most bass were the younger males, then more variance for treble but little for bass, and then the rest with more variance but still less than overall).

Longer still: how do you define accurate, as well as objective when it comes to listener preference?

 

[Justdafactsmaam]

Short answer: none.

Longer answer: they were subjectively derived, and there was a high degree of variance overall (I don't have sufficient data, but if I'm correct about the three female listeners, then they were actually quite consistent with little variance; similarly, if I guess the two listeners who preferred the most bass were the younger males, then more variance for treble but little for bass, and then the rest with more variance but still less than overall).

Longer still: how do you define accurate, as well as objective when it comes to listener preference?

I separate preferences and accuracy as two different things. Accurate is defined as output matching the input on an audible level.

The question is where do preferences and accuracy overlap and where do they diverge

 

[youngho]

I separate preferences and accuracy as two different things. Accurate is defined as output matching the input on an audible level.

The question is where do preferences and accuracy overlap and where do they diverge

As discussed above, the curves were derived from individual preference and deviated from projected measurements (predicted in-room response, which also deviated from actual measurements above 400-500 Hz) of relatively linear loudspeakers (Revel F208) placed in the Harman reference room, hence my short answer above. What I assume was the female curve was probably closest below 100 Hz, while the overall curve was closest above.

However, just out of curiosity, how does your definition of accurate take into account equal loudness contours, since human perception or "audible level" is not linear with sound pressure level, or do you mean at about ~80 dB where hearing is supposedly most linear (https://www.theseus.fi/bitstream/handle/10024/749081/Kinnunen_Teemu.pdf)?

 

[Justdafactsmaam]

As discussed above, the curves were derived from individual preference and deviated from projected measurements (predicted in-room response, which also deviated from actual measurements above 400-500 Hz) of relatively linear loudspeakers (Revel F208) placed in the Harman reference room, hence my short answer above. What I assume was the female curve was probably closest below 100 Hz, while the overall curve was closest above.

However, just out of curiosity, how does your definition of accurate take into account equal loudness contours, since human perception or "audible level" is not linear with sound pressure level, or do you mean at about ~80 dB where hearing is supposedly most linear (https://www.theseus.fi/bitstream/handle/10024/749081/Kinnunen_Teemu.pdf)?

The accuracy of the playback system itself is independent of a human listener.

 

[youngho]

@sigbergaudio: you can find a lot more data from this paper (and others) here: https://www.aes-media.org/sections/pnw/pnwrecaps/2014/solive_sep14/

There are a number of points that might warrant further discussion:
1. Slide 14 and 15 seems to indicate that the response of all channels in the HRR is reasonably linear before equalization, though the smoothing hides a lot. I wonder why the side and rear channels don't show any rise in the bass relative to the left and right front channels when they appear to be much closer to the walls.
2. Slide 16 indicates that Target 1 was based on measurements of the JBL LSR (6332, though that was specified in the slide) in the HRR, but then slide 19 would imply a linear response. I don't think that's actually the case because of #4 below. Data for the LSR 6332 can be found at https://jblpro.com/en-US/site_elements/lsr6332-spec-sheet, can compare with slide 24
3. Slide 20: note that JBL Target 2 (basically Target 1 with slightly less bass and treble) was preferred over Target 1. That only makes sense if that Target 1 did not reflect the linear response in slide 15.
4. Slide 24: compare the Before EQ measurements of the Revel 208 with what we saw in slide 15. Here we see what appear to be significant modal and boundary effects below the presumed Schroeder frequency range. I have to assume that the JBL LSR 6332 measurements resulting in Target 1 above also reflected these effects, as well, so that could explain why slightly bass as in Target 2 was preferred over Target 1. In terms of the After EQ curve, I don't quite understand why the 80 Hz peak couldn't be flattened more, since the net result of averaging seems to result in a response that's a little bass-shy below 60 Hz relative to a little bright above 1 kHz.
5. Slide 30: the red arrows seem to be in the wrong position along the x-axis. Reading the discussion by Todd Welti at https://hometheaterhifi.com/technic...n-interview-with-todd-welti-and-kevin-voecks/ but ignoring the diagram, I would have thought that the range from 105 to 2.5 kHz was pretty linear, rather similar to https://tech.ebu.ch/docs/tech/tech3276.pdf. These tilts seem to be pretty smooth without the lumpiness/variability seen in Toole's curve from his paper: https://www.aes.org/e-lib/browse.cfm?elib=17839
6. Slide 33: here is some information on the listeners.
7. Slide 37 and 38: I think there's an error in one of these slides, since the untrained listeners preferred an average of -0.8 treble with speaker and headphone combined, but then they preferred an average of +1.8 on headphones and again on speakers when the data is separated? I'm going to guess that the headphones should have been -2.6, not +1.8. The right part of the graph goes towards the relative difference of treble and bass.
8. Slide 39: Wow, look at the huge variability between listeners! Who were the untrained listeners? Probably 400, 401, and 402. Maybe the three women were 4, 58, and 346.
9. Slide 40: again, a lot of variability in the delta
10. Slide 45: "The preferred bass and treble levels of the target function for loudspeakers and headphones varied among individual listeners (see Fig 14). For loudspeaker playback, the range of preferred bass and treble levels was 17 dB and 11 dB, respectively." That certainly seems like a lot of variability to me!

I am attaching slide 39 below to illustrate the huge variance. Again, I believe listeners 400, 401, and 402 are the three untrained listeners (Sean Olive's slide deck says that there were three untrained listeners, all under 30, but Toole's description said that there were four), as it looks like the averages of the bass and treble preferences come out close to the +11.2 and +1.8 respectively described in the paper for untrained listeners' settings for loudspeakers (and similarly for the average for headphones). A subsequent study with headphones (https://acousticstoday.org/he-perce...uality-what-do-listeners-prefer-sean-e-olive/, edit: also https://www.listeninc.com/wp/media/Perception_and_-Measurement_of_Headphones_Sean_Olive.pdf) suggested that "Females like less bass [+5.6 vs +6.5] and treble [-3.1 vs -1.2]...Listeners with more experience in critical listening prefer less bass and treble than listeners with less experience...With increasing age, listeners prefer less bass and less treble until 56+ years where they prefer much less bass and more treble possibly due to hearing loss." I wonder about what characteristics of listeners 4, 58, and 346 might have contributed to their preference for much less bass and treble than the overall average of +6.6 below 105 Hz and -2.4 above 2.5 kHz. There seems to be a small (3/11) Much Less Bass is Better (the headphone study indicated 2 dB less bass than the headphone target curve) group, but here it was much less), as well as a minority (3/11) More Bass is Better (>+4 dB) group. These numbers are relatively similar to the larger headphone study, but the underlying characteristics seem to differ.

In any case, the Harman target curve for loudspeakers in rooms averaged the data from 100% of listeners in this study, while the Trained Listeners curve averaged the data from 73% (8 out of 11) and the Untrained Listeners curve 27% (3/11). This is not the same thing as saying that 100% of listeners preferred the Harman target curve in the same way that the average height of 100% of the (edit: adult) US population may be ~5'6", while the average height of men may be ~5'9" and women 5'4", but it's not correct to say that 100% of the US population is 5'6" or that roughly half (just under) are 5'9" and roughly half (just over) are 5'4".

 

[sigbergaudio]

(...)In any case, the Harman target curve for loudspeakers in rooms averaged the data from 100% of listeners in this study, while the Trained Listeners curve averaged the data from 73% (8 out of 11) and the Untrained Listeners curve 27% (3/11). (..)

Just to keep beating the not quite dead horse, please do not call it a target curve. It is not a target curve. It is an average of the preference of 11 people listening to some decent speakers at Harman. No one concluded or suggested that this should be a target for people tuning their systems.

 

[sigbergaudio]

I separate preferences and accuracy as two different things. Accurate is defined as output matching the input on an audible level.

The question is where do preferences and accuracy overlap and where do they diverge

It is more complicated than that. The room affects the frequency response, the speakers dispersion characteristics affects the frequency response. So you cannot directly map the recorded frequency response in the listening position to the source response.

Are you asking which of the responses are a result of a speaker that is closest to flat anechoically?

 

[Justdafactsmaam]

It is more complicated than that. The room affects the frequency response, the speakers dispersion characteristics affects the frequency response. So you cannot directly map the recorded frequency response in the listening position to the source response.

Are you asking which of the responses are a result of a speaker that is closest to flat anechoically?

No. That is not what I am asking. Since the speakers aren’t being played in an anechoic chamber no one is reacting to that. The effect the room has on the sound is its own deviation from accuracy. So I would be asking which curve caused by the speaker and room in tandem would be deemed the most accurate to the source signal.

 

[restorer-john]

Too thin or too lean is not accurate. Without exception, all engineers I've collaborated or presented sound systems to have preferred a room curve with 6-8dB rise in the bass (50hz-10khz), and felt that this is the closest to listening live. Personally I've also never been to a live / acoustic concert where the instruments have sounded hard or lacking of bass. Quite the contrary.

This is due to the gradual and ongoing boosting of LF in all aspects of live/amplified/replayed music of any type in the last 30 years. In fact, I'd say it was digital and the ability to record any amount of LF without causing problems that created the problem.

The availability of vast amounts of clean power at LF and drivers that can absorb and reproduce it has resulted in an "arm's race" in the low frequencies.

When you are in a small club/venue with a live drummer only a few feet away and they feel the need to augment an acoustic with an electronic is where the problem lies IMO. It sounds stupidly disproportionate in situations like that or in 'live' situations where you are near the stage and the drum kit is just an excuse for LF electronic impulses. Drums don't sound like drums anymore on recordings- many never existed in the first place and the ones that did, have been amplified to ridiculous levels.

Most systems I hear where the unfortunate owners think they are "fully dialled in" sound putrid. Like the obsession with subwoofers to 'reproduce' sounds that never existed in the first place, except in the minds of the mixing guys (not engineers). Drums are mid bass, not seismic.

 

[sigbergaudio]

No. That is not what I am asking. Since the speakers aren’t being played in an anechoic chamber no one is reacting to that. The effect the room has on the sound is its own deviation from accuracy. So I would be asking which curve caused by the speaker and room in tandem would be deemed the most accurate to the source signal.

Well, again, since that (the "correct" curve) will look different in every room, how do you suggest I answer that?

 

[Justdafactsmaam]

Well, again, since that (the "correct" curve) will look different in every room, how do you suggest I answer that?

I guess I’d start by explaining how accuracy can vary based on rooms

 

[sigbergaudio]

I guess I’d start by explaining how accuracy can vary based on rooms

So to quote myself:

It is more complicated than that. The room affects the frequency response, the speakers dispersion characteristics affects the frequency response. So you cannot directly map the recorded frequency response in the listening position to the source response.

This means a speaker that is flat anechoically, will not measure the same in two different rooms.

This also means that another speaker that is also flat anechoically, may not measure the same as the first speaker, in any of those two rooms.


The above implies

A) that there is no one answer to "What does an accurate in-room curve look like" that is valid across two different rooms.

B) Two different speakers in the same room, may measure differently when measured by a microphone in the main listening position, but both may still sound accurate.

To make it even more fun (not all will agree with this one probably):

C) A speaker that is not flat anechoically from around Schroeder and down may still sound accurate in-room.


EDIT:
We also have the interesting

D) From around Schroeder and up, we have the interesting fact that if we have a curve that does not look right, and EQ it to look more right, it may end up sounding less right.

 

[Deleted member 21114]

Most systems I hear where the unfortunate owners think they are "fully dialled in" sound putrid.

Line up the family members. Turn the subs down until the teenage son is sulking, the husband is frowning and the wife is smiling or at least not saying "boomy".

 

[Justdafactsmaam]

So to quote myself:



This means a speaker that is flat anechoically, will not measure the same in two different rooms.

This also means that another speaker that is also flat anechoically, may not measure the same as the first speaker, in any of those two rooms.


The above implies

A) that there is no one answer to "What does an accurate in-room curve look like" that is valid across two different rooms.

B) Two different speakers in the same room, may measure differently when measured by a microphone in the main listening position, but both may still sound accurate.

To make it even more fun (not all will agree with this one probably):

C) A speaker that is not flat anechoically from around Schroeder and down may still sound accurate in-room.


EDIT:
We also have the interesting

D) From around Schroeder and up, we have the interesting fact that if we have a curve that does not look right, and EQ it to look more right, it may end up sounding less right.

So in the real world rooms cause a wide variety of inaccuracies as do speaker dispersion patterns which will be room dependent. So flat on axis frequency response is at best one element of objective system accuracy that will be corrupted by room interaction to varying degrees depending on the room, the speaker/ listener positions and speaker dispersion patterns.

But at the end of the day when we are considering true objective system accuracy would it not be the system as a whole whose measurements at the listening position are closest to the source signal? And if not doesn’t that ultimately make the very idea of objective system accuracy a bit ambiguous or at least complicated at best?

Still very much looking forward to JJs lecture on the subject. I suspect true objective system accuracy is a bit of a greased pig so to speak.

 

[sigbergaudio]

But at the end of the day when we are considering true objective system accuracy would it not be the system as a whole whose measurements at the listening position are closest to the source signal? And if not doesn’t that ultimately make the very idea of objective system accuracy a bit ambiguous or at least complicated at best?

I think one conclusion is that we cannot judge a system by measurements alone (audience gasps audibly).

As we know, the room typically adds a quite significant tilt on the frequency response, so measurements at the listening position is quite different from the source signal. So if you want it as close to the source as possible, you are saying you want as little tilt as possible. We know that is not necessarily what will be perceived as most correct. Little to no tilt will give too little low end.

 

[fpitas]

I just adjust my speakers so they sound natural, especially voices. I use headphones as a sanity check.

 

[Deleted member 21114]

I just adjust my speakers so they sound natural, especially voices. I use headphones as a sanity check.

Humans are great at telling when voices sound right and skin tones look good.

 

[DanielT]

Absolutely. A total of 11 participants is hardly sufficient for a reliable statement.
So the trend/result may be correct or not so much. Are there follow up studies?

Did you get an answer? Maybe I missed it.It is of course crucial.

Having 11 participants decide which is the best type of ice cream, you cannot generalize any results on what people in general think about that.

 

[fpitas]

Humans are great at telling when voices sound right and skin tones look good.

Yes, it took some doing to really get it right.

 

[thewas]

Just to keep beating the not quite dead horse, please do not call it a target curve. It is not a target curve. It is an average of the preference of 11 people listening to some decent speakers at Harman. No one concluded or suggested that this should be a target for people tuning their systems.

Exactly, like Toole says the curve is a result of a good loudspeaker in an acoustically good room but never a target as it changes depending on the directivity, placement, listening distance and room reverberation characteristics.

Here is a collection of Toole's quotes on that topic, wish more people would read it so we don't have this discussion several times a year all over again:

Some comments from Floyd Toole about room curve targets, room EQ and more

Room curve targets Every so often it is good to review what we know about room curves, targets, etc. Almost 50 years of double-blind listening tests have shown persuasively that listeners like loudspeakers with flat, smooth, anechoic on-axis and listening-window frequency responses. Those with...

www.audiosciencereview.com