Few things have changed my views on audio as sitting in this room and voting for speakers "blind." Here is what the setup looks like and nice history of it from Dr. Sean Olive. Only 11 minutes and "must watch:"
-
WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
Dr. Sean Olive Interview on Harman Blind Speaker Testing system
- Thread starter amirm
- Start date
OP
- Thread Starter
- #2
Longer interview by Scott Wilkinson, includes such topics as expert listeners, room correction, etc.
I was OK listening at 1.5X real-time (click on the gear symbol).
I was OK listening at 1.5X real-time (click on the gear symbol).
OP
- Thread Starter
- #3
One of the excuses for not doing double blind test of speakers is the high cost of the speaker shuffler room Harman uses above. But this need not be the case. Their multi-channel room has a simple rotating triangle that can swap out speakers and audio devices quickly to same end:
Detlof
Member
Fascinating! Does anybody know according to what criteria the listeners are trained?
Phelonious Ponk
Addicted to Fun and Learning
- Joined
- Feb 26, 2016
- Messages
- 859
- Likes
- 218
Amir does. He has done the training. Toole and Olive have probably done more to advance home audio (and automotive) than almost anyone out there. And of course, with the exception of the expensive Revel line, they are dismissed by the audiophile community. JBL kicks ass. Nobody in the hobby manages to believe it.
Tim
Tim
OP
- Thread Starter
- #6
They give away the tool. It is called appropriately, "how to listen."
Here is the web site: http://harmanhowtolisten.blogspot.com/
In a nutshell, the training software allows you to:
1. Tell frequency response variations
2. Be able to identify accurately the nature of the response variation
Frequency response variations by far is the most accurate predictor of loudspeaker preferences by listeners. And where speakers vary substantially from each other.
We are all pretty good about #1 but poor on #2. Training software helps to improve #2.
But it is trivial for a machine to give us a precise, objective measurement of #2. Why do humans need to be able to do it?
One day the machines will deceive us.. We must never become complacent ....
I would be rejected (I assume) as a listener at 9:00 in the first video.
Dr Olive: First, we are going to test your hearing.
Me: What?
Dr Olive: Next!
Me: Huh?
---
I'll have to bring my P-363's back into the Research Kitchen here at Neverland East and do some more rigorous measurements/comparisons versus my crappy panels. I'm not satisfied with the quality of the measurements I took before, having learned a little more about measurement pitfalls, and now having automated correction capability.
Dr Olive: First, we are going to test your hearing.
Me: What?
Dr Olive: Next!
Me: Huh?
---
I'll have to bring my P-363's back into the Research Kitchen here at Neverland East and do some more rigorous measurements/comparisons versus my crappy panels. I'm not satisfied with the quality of the measurements I took before, having learned a little more about measurement pitfalls, and now having automated correction capability.
- Joined
- Feb 23, 2016
- Messages
- 21,960
- Likes
- 40,216
There is merit in what you wrote here.
OTOH, I think they want to test humans trained to the fullest extent possible of human perception. Then use those listeners to determine just what response variations bother the most trained humans available. Then if they build a product that these humans hear nothing wrong with the assumption is their product is beyond reproach for virtually all humans.
Sort of like saying most humans can hear distortion once it reaches about 1-2%. Some highly trained humans can hear it at .1%. Build gear with less than .1% and no one will hear it.
Of course that doesn't fix the situation when some humans prefer a little distortion to none, or some humans prefer a response accent in some ranges vs none or when recording you want response variations to pretty up an otherwise drab sound vs none.
Looking at:
https://www.harman.com/sites/default/files/white-paper/12/11/2015 - 05:54/files/AudioScience.pdf
And seeing:
Page 8: "High-frequency loss, by itself, was not a clearly correlated factor. Reference 4 covers this in detail."
4. Toole, F.E., “Subjective Measurements of Loudspeaker Sound Quality and Listener Performance”, J. Audio Eng. Soc., vol. 33, pp. 2-32 (1985 January/February)
Not having access nor $33 worth of interest (since there is nothing I can do about my (possibly) genetic disposition to not hearing HF), could you (Amir) post an excerpt or summarize whatever is said on this topic, that is (as I interpret what I read), HF loss but normal LF acuity?
https://www.harman.com/sites/default/files/white-paper/12/11/2015 - 05:54/files/AudioScience.pdf
And seeing:
Page 8: "High-frequency loss, by itself, was not a clearly correlated factor. Reference 4 covers this in detail."
4. Toole, F.E., “Subjective Measurements of Loudspeaker Sound Quality and Listener Performance”, J. Audio Eng. Soc., vol. 33, pp. 2-32 (1985 January/February)
Not having access nor $33 worth of interest (since there is nothing I can do about my (possibly) genetic disposition to not hearing HF), could you (Amir) post an excerpt or summarize whatever is said on this topic, that is (as I interpret what I read), HF loss but normal LF acuity?
Last edited:
- Joined
- Feb 23, 2016
- Messages
- 21,960
- Likes
- 40,216
Yes I would be interested in that too.
I do have a guess as to why it would be so. You would find the level of sound at frequencies above 10 khz and usually 8 khz aren't very high with the great majority of music -60 dbfs or less being quite common. Plus above 5 khz your ear's sensitivity is dropping by some 10 db or so in that range. So my guess would be hearing still good to 8 khz would only change results slightly.
An interesting thing to do in Audacity is to do a steep EQ to high pass everything above 9 or 10 khz. Then slow the speed by 50% which doubles your ear's frequency range in a sense. Even then when such a track is played back there isn't all that much going on in almost all music. What's left is just not very loud or impressive. Take into account you are now hearing it with no masking from lower and louder frequencies and it isn't hard to think much of what is there is not terribly meaningful or heard when listening normally to music.
OP
- Thread Starter
- #13
The measurement is half the battle. You then need to correlate that with human preference. We have two ears which cannot be represented by one microphone. You could perform binaural measurements but then you have two graphs and you have to decide how the brain would interpret the different signals. It is more practical to use humans to get the data. Once there, you work backward to what measurements contributed to that. And that is precisely what is done now.
Mine is lower than that. Maybe I'm lucky! Can't be annoyed by HF defects...
My "perception" when shopping back in 1995 was the highs were present in the panels (no cross in the range) when they weren't on three ways... Is this tweeter working? At all? Hello?
Interestingly (to me), I can't even seem to imagine frequencies higher than I can hear. Can you?
- Joined
- Feb 23, 2016
- Messages
- 21,960
- Likes
- 40,216
That just mean you aren't a good audiophile. Apparently some can imagine hearing up to 100khz at least.
In reality I would answer your question as no I can't imagine higher frequencies.
NorthSky
Major Contributor
- Joined
- Feb 28, 2016
- Messages
- 4,998
- Likes
- 965
Me I like solid bass. But I don't know if my ears can hear below 22Hz.
...It don't matter...as long as the speakers measure a clean 20Hz and only 2 decibels down @ 16Hz.
I'm an old listener, so speakers with good response @ 12kHz are good enough for me. No need for super tweeters.
_______
What I'd like to see from HK is those same blind tests from audiophiles living in Germany, China, Russia, Spain, Brazil and France testing speakers in the $100,000 range and above...up to the max. They'll need a switching platform with one billion horsepower.
...It don't matter...as long as the speakers measure a clean 20Hz and only 2 decibels down @ 16Hz.
I'm an old listener, so speakers with good response @ 12kHz are good enough for me. No need for super tweeters.
_______
What I'd like to see from HK is those same blind tests from audiophiles living in Germany, China, Russia, Spain, Brazil and France testing speakers in the $100,000 range and above...up to the max. They'll need a switching platform with one billion horsepower.
But it has already been stated that the assumption is that frequency response variations are the strongest predictor of speaker preference. In other words, we seem to be training people in order for them to confirm what a machine can already tell us.
What I am getting at is that if the aim is "high fidelity", the need for a flat or smooth frequency response is (to quote Monty Python) "bleedin' obvious". Why go through the rigmarole of training people to confirm what you already know? (or at least already assume)
Hi Blumlein. I would agree with that if it was clear that achieving a smooth response (or low distortion) was expensive. As it is, I don't think there is a problem in achieving any arbitrary response we want, as long as we are prepared to spend more than a few hundred dollars i.e. audiophile pocket money.
An unusual aspect to the Harman approach seems to be that they do their tests using commercial rivals' products, rather than building a series of speakers with controlled, defined variations in specific variables - so they seem to be making the results a bit less definitive than they might be.
While I think that the goal of fidelity to the source signal is an aim of high fidelity, when it comes to the big variable, the consumers, we do not all agree on what set of speakers sounds the best to us, and its those individual characteristics that allow for hundreds of speakers out there to try. Speakers as I see them (and the room) are where folks tend to personalize for a sound that pleases their preferences, so the literally most accurate speaker playing plain old stereo might actually not please as many folks as we might think.
Its kind of like the mixing and mastering, if you could hear three or four engineers mix and master of the same mic feeds performance, we would all not choose the same rendition as the most pleasing to us.
That's why I think the human factor is important in the Harman tests although of course a measurement can tell you strictly about accuracy of the what comes out of each speaker in a speaker box or panel or whatever.
Reproduction of an event with plain old stereo in the most accurate sense does not convey to our complex hearing and ear/brain interface the real thing, so that opens the door for folks to make speakers sound whichever way that tickles folks fancy.
Its kind of like the mixing and mastering, if you could hear three or four engineers mix and master of the same mic feeds performance, we would all not choose the same rendition as the most pleasing to us.
That's why I think the human factor is important in the Harman tests although of course a measurement can tell you strictly about accuracy of the what comes out of each speaker in a speaker box or panel or whatever.
Reproduction of an event with plain old stereo in the most accurate sense does not convey to our complex hearing and ear/brain interface the real thing, so that opens the door for folks to make speakers sound whichever way that tickles folks fancy.
Similar threads
