I cannot trust the Harman speaker preference score

[Presently42]

A smooth DI cannot predict sound quality, but it can allow you to equalize Speaker B (Bad) to make it sound more like Speaker A.

Forgiveness if this is a bit off topic, but might you elucidate this point a bit more? I thought I recalled @Floyd Toole writing, that eq only really affects the ON curve: if so, then one cannot use the DI to eq a speaker, as any attempt to do so will merely alter the ON curve, rather than the DI. Indeed, no curve at all can be used, other than the ON curve (though perhaps the ER curve can be useful: I've had great success, for example, using an amalgamation of the ON and ER curves - something I'd like to investigate more rigourously).

 

[Bjorn]

What happens between about 100 and 500 Hz is indeed messy, being a combination of adjacent boundary "radiated power" issues and acoustical interference. There is no perfect solution to that, but EQ definitely helps. Above 500 Hz detailed EQ is risky without anechoic data on the loudspeakers. Fortunately this seems to be something that humans can somewhat adapt to, as it happens in real life - even conversation.

There are much better speaker designs that flatten out this area and there's of course acoustical treatment. Both needed for a great result.

We are more sensitive to bass/low mid frequencies at 100-250 Hz than sub-frequencies, though we obviously want to have an excellent result for both ranges.

 

[Newman]

Forgiveness if this is a bit off topic, but might you elucidate this point a bit more? I thought I recalled @Floyd Toole writing, that eq only really affects the ON curve: if so, then one cannot use the DI to eq a speaker, as any attempt to do so will merely alter the ON curve, rather than the DI. Indeed, no curve at all can be used, other than the ON curve (though perhaps the ER curve can be useful: I've had great success, for example, using an amalgamation of the ON and ER curves - something I'd like to investigate more rigourously).

I took his comment NOT to mean that you can set a target DI and equalise to meet it, but to mean that if a speaker has a bad LW and a bad DI there is no point to EQ it, but if it has a bad LW and a smooth DI, you can anticipate good results from equalising it (to a target LW curve).

 

[fineMen]

I was just looking through some other speakers (not Dutch&Dutch or Kii), and there are few that seem to control directivity down to about 200 Hz. ....

Looks great!

Well, there is a solution, and it gives performance good enough that the behavior in 100-500 range will be similar to the rest of the spectrum. ...

A more narrow dispersion would sparse out reflections, which is more or less just an implicated objective.

Please explain this implication - why is it beneficial to have it, as You say, "similar to the rest of the spectrum"?

( There are so many 'mantras' in audio, that are considered obvious, then taken for granted, then without thinking get replicated over and over. Once we stand at the doors of big science, questions arise. )

Why is it, that below the => Schroeder frequency (see link) the somehow controlled dispersion is given up? If it is a trade-off, what parameters are compromised in favour of the other to which degree? If so, why the discrimination below vs. above the Schroeder frequency. Room dependence?

Eventually I wonder why a topic, explicitly neglected by "the score" is discussed, only to mention some successful doing ( TimVG, kvalsvoll, both: cardioid ), but never giving rigid reason for it. The latter being the only legit motivation and purpose for entering the thread. (... take it with a grain of salt, I'm not that square ... )

 

[Cote Dazur]

This interview with Sean Olive, helped me understand better how the score was implemented and what the limitations may or may not be.
Also helped me also put a face on a name, what a nice person Sean seems to be!

My biggest take away is that it looks like not much has been done since, from anyone, to improve on Sean's original work, which is a petty as more subjective evaluations/listening test, scientifically conducted, in different rooms, would have make the rating more accurate. As said in the video, all this work, need, among other factors, a lot of money.

Who knows how much better speakers available to us would be if such an improve rating existed?

 

[-Matt-]

Thanks for posting the interview - I'm part way through but have a hypothetical question...

Please forgive my naivety...

It is stated that a speaker that produces perfectly flat sound power across the frequency range does not yield the most favourable listening experience (as determined by blind listener preference tests).

This is understandable, but as a layman it seems to me that the non-flat target curve could be applied in the content rather than via speaker design.

I.e. If all speakers were designed with flat spl in mind (something that can be measured fairly readily and accurately), and if content was also produced on such speakers, then wouldn't the preference curve (bass boost etc) be included in the content?

The flip side of this argument is that all recordings that have been produced to-date (on traditional speakers) have a sound power curve encoded into them which is actually what makes speakers with flat spl less favourable. Hence, the content used in blind listening tests biases the listener preference.

Of course, I doubt this post will cause the world to rip up all previously recorded music content and start from scratch! ...But, hypothetically, could music be produced specifically for speakers that are designed to achieve flat sound power? If so, perhaps this could be a way of removing some of the vagaries of sound reproduction?

 

[Cote Dazur]

I'm part way through but have a hypothetical question...

My advice to you is keep reading and listen to the scientific evidence we have been provided by the luminaries that we are so fortunate to have to support our quest and all will be revealed, even the answer to your question.

 

[levimax]

Thanks for posting the interview - I'm part way through but have a hypothetical question...

Please forgive my naivety...

It is stated that a speaker that produces perfectly flat sound power across the frequency range does not yield the most favourable listening experience (as determined by blind listener preference tests).

This is understandable, but as a layman it seems to me that the non-flat target curve could be applied in the content rather than via speaker design.

I.e. If all speakers were designed with flat spl in mind (something that can be measured fairly readily and accurately), and if content was also produced on such speakers, then wouldn't the preference curve (bass boost etc) be included in the content?

The flip side of this argument is that all recordings that have been produced to-date (on traditional speakers) have a sound power curve encoded into them which is actually what makes speakers with flat spl less favourable. Hence, the content used in blind listening tests biases the listener preference.

Of course, I doubt this post will cause the world to rip up all previously recorded music content and start from scratch! ...But, hypothetically, could music be produced specifically for speakers that are designed to achieve flat sound power? If so, perhaps this could be a way of removing some of the vagaries of sound reproduction?

Flat anechoic response (or flat near field response) will appear downward sloping at distance / listening position.

 

[-Matt-]

Flat anechoic response (or flat near field response) will appear downward sloping at distance / listening position.

Yes, thank you. I do realise that, but my point was:

If the expectation was that all speakers were designed to have flat sound power (measured anechoically I guess) couldn't music just be produced with higher volume at higher frequencies in order to compensate? (And therefore sound just as good as current speakers with current content).

As I said, I'm a layman, but flat sound power vs frequency sounds like a more logical starting point for speaker design than something based on subjective listener preference (which is dependent on the frequency balance of existing sound recordings) (which is in turn likely based on some historical limitations that I'm unaware of, and may no longer be present).

Edit: Actually my question is a bit tangential to the discussion on preference ratings (sorry for the diversion). Apart from the slope parameter, wouldn't speakers with a perfectly flat sound power get quite a good preference score?

 

[levimax]

Yes, thank you. I do realise that, but my point was:

If the expectation was that all speakers were designed to have flat sound power (measured anechoically I guess) couldn't music just be produced with higher volume at higher frequencies in order to compensate? (And therefore sound good).

As I said, I'm a layman, but flat sound power vs frequency sounds like a more logical starting point for speaker design than something based on listener preference (which is dependent on the frequency balance of existing sound recordings) (which is in turn likely based on some historical limitations that I'm unaware of, and may no longer be present).

Recordings are mixed to sound good to the mastering engineer listening in his mastering studio... how it sounds to anyone else on their system in their room is anyone's guess.... hence the "circle of confusion" and the reason for tone controls per Floyd Toole.

 

[-Matt-]

...and if he mixed his recording using speakers with flat sound power and sold them to listeners with the same...?

 

[-Matt-]

What factors have led to the current frequency balance of speakers and recordings?

Was it perhaps the specific response curve of the earliest microphones and drivers?

 

[Kvalsvoll]

Please explain this implication - why is it beneficial to have it, as You say, "similar to the rest of the spectrum"?

First; it may not be obvious what this difference is, between "rest of the spectrum" and outside of this "rest..". At higher freq, the decay in the room is diffuse, there are no distinct resonances, frequency distribution is spread out, there is low directionality in the sound field. As frequency goes down, wavelengths get larger compared to dimensions and distances inside the room, and the decay moves towards discrete reflections and resonances. This can be observed in measurements as peaks in the frequency response and resonances in the decay.

When the room is fixed, all this goes away. The room now behaves in a similar way across the whole frequency range, where there are no longer any resonances left, just decaying energy.

This has huge impact on the sound. Tonality and timbre of instruments no longer colored by room resonances, presence of rendered instruments is much better, like they have a physical presence in the room, clarity in bass and lower midrange is much better, bass instruments now gets a rendering more similar to higher frequencies, where placement is defined and precise.

 

[dominikz]

Yes, thank you. I do realise that, but my point was:

If the expectation was that all speakers were designed to have flat sound power (measured anechoically I guess) couldn't music just be produced with higher volume at higher frequencies in order to compensate? (And therefore sound just as good as current speakers with current content).

As I said, I'm a layman, but flat sound power vs frequency sounds like a more logical starting point for speaker design than something based on subjective listener preference (which is dependent on the frequency balance of existing sound recordings) (which is in turn likely based on some historical limitations that I'm unaware of, and may no longer be present).

Edit: Actually my question is a bit tangential to the discussion on preference ratings (sorry for the diversion). Apart from the slope parameter, wouldn't speakers with a perfectly flat sound power get quite a good preference score?

IMHO in principle you are right that any EQ curve built into a loudspeaker with good DI could be inverted by a corresponding recording EQed with the exact opposite curve to get a perceived flat/balanced tonality - if such a standard encoding/decoding curve existed.

However above modal region people seem to naturally perceive tonality based mainly on the direct sound (and not sound power), so it makes sense to start with flat direct sound when designing a speaker - which then results in a downward sloping sound power (for conventional front-firing loudspeaker designs).

Note that this also means that we can keep the recordings encoded with natural/flat tonality and hear them the same.
E.g. white noise is actually encoded with the expected flat spectrum and be perceived as white noise.

If we used speakers with flat sound power instead, to hear white noise the noise spectrum on the recording would need to be EQed to be downward sloping - i.e. looking more similar to pink noise.

 

[-Matt-]

IMHO in principle you are right that any EQ curve built into a loudspeaker with good DI could be inverted by a corresponding recording EQed with the exact opposite curve...

Thanks, so in principle the target response for speaker designs and the complimentary EQ curve applied to recordings could be chosen arbitarily.

...so it makes sense to start with flat direct sound when designing a speaker - which then results in a downward sloping sound power (for conventional front-firing loudspeaker designs).

Note that this also means that we can keep the recordings encoded with natural/flat tonality and hear them the same.
E.g. white noise is actually encoded with the expected flat spectrum and be perceived as white noise.

Ok, thanks for the explanation, I accept the argument that tuning speaker response to give flat direct sound (for recorded white noise) is rational.

...So why not standardise the targeted speaker response to be exactly that (flat direct sound). A slight modification from this based on subjective listener preference doesn't seem to be a good idea, especially if that modification could instead be incorporated into the recorded content.

I'm not knocking the science, I understand that the formulas for the preference score provided a good fit for one particular set of blind listening data (and have been of use to many others since) however, it is still based on a fairly small group of listeners and speakers.

I'm instead trying to understand, hypothetically, what is the optimal speaker response to target for accurate sound reproduction.

 

[dominikz]

...So why not standardise the targeted speaker response to be exactly that (flat direct sound). A slight modification from this based on subjective listener preference doesn't seem to be a good idea, especially if that modification could instead be incorporated into the recorded content

There is no 'standard' loudspeaker design that I know of, but subjective preference research seems to imply that indeed loudspeakers with flat direct sound (LW) and good directivity are in general preferred to others - i.e. I see no conflict or discrepancy there TBH. Not sure what do you mean by 'modification', if you could clarify?

Are you maybe thinking about often discussed 'target' in-room response instead? That will be downward-sloping for conventional front-firing loudspeakers with flat LW. The so-called 'Harman target curve' is actually an average in-room response of such well-measuring loudspeakers - so more a 'result' of good design rather than a 'target'.

 

[-Matt-]

There is no 'standard' loudspeaker design that I know of, but subjective preference research seems to imply that indeed loudspeakers with flat direct sound (LW) and good directivity are in general preferred to others - i.e. I see no conflict or discrepancy there TBH.

I'm wishing there was a simple standard, for frequency response, towards which all speaker designs could aspire and against which all could be tested; and I guess I'm naively suggesting perhaps the most obvious candidates.

Not sure what do you mean by 'modification', if you could clarify?

Are you maybe thinking about often discussed 'target' in-room response instead? That will be downward-sloping for conventional front-firing loudspeakers with flat LW. The so-called 'Harman target curve' is actually an average in-room response of such well-measuring loudspeakers - so more a 'result' of good design rather than a 'target'.

Yes, indeed, I think that in my mind I am conflating the preference score and the Harman in-room target curve. I previously thought that to achieve the highest possible preference score a speaker would have to have a frequency response that deviates from having totally flat direct sound, perhaps I was wrong?

So, summarising... Perhaps we already have a suitable target frequency response for speaker designs (flat direct sound). Such speakers produce (on average) an in-room response close to the Harman target. Such speakers should also achieve high preference scores.

If the above is correct then of course the preference score should be a good metric that helps; although the actual frequency response and spin data can give much more information.

 

[Holmz]

I'm wishing there was a simple standard, for frequency response, towards which all speaker designs could aspire and against which all could be tested; and I guess I'm naively suggesting perhaps the most obvious candidates.
…

Like the ones that score the highest?

…
Yes, indeed, I think that in my mind I am conflating the preference score and the Harman in-room target curve. I previously thought that to achieve the highest possible preference score a speaker would have to have a frequency response that deviates from having totally flat direct sound, perhaps I was wrong?
…

Flat in the anechoic chamber, slopes down in a room.

…
So, summarising... Perhaps we already have a suitable target frequency response for speaker designs (flat direct sound). Such speakers produce (on average) an in-room response close to the Harman target. Such speakers should also achieve high preference scores.

If the above is correct then of course the preference score should be a good metric that helps; although the actual frequency response and spin data can give much more information.

Yep.

 

[Multicore]

So, to hear a section of strings on far left and another section of double bass on far right, is perversion? You seems like just a poor guy who can't appreciate spices of life....I will ignore you going forward. Bye bye!

Well, that's one example. Another is all those early stereo recordings of small jazz ensembles where you can literally silence half the band by turning off one channel.

 

[Holmz]

Well, that's one example. Another is all those early stereo recordings of small jazz ensembles where you can literally silence half the band by turning off one channel.

We can also do it visually by selecting the news channel(s).