April 2015 Toole video on sound reproduction

[dallasjustice]

One constant criticism of the automatic method is the inability to duplicate resulting response determined by the EQ tool via independent measurements, e.g., using REW. That is, however, an issue resulting from things like the inability to duplicate mike placements exactly in calibration vs. subsequent measurement. Also, many aspects of the algorithms used by the tool may be proprietary or unknown, such as multipoint averaging, time windowing, 1/x frequency smoothing, etc. So, independent followup measurement gets trickier with automated tools, unlike the more open and more deliberate completely manual calibration idea.

This criticism is certainly not true. I've done it many times with Dirac Live, Audiolense and Acourate. All three match up with predicted response in frequency and time domains. In the case of the Dirac Live, I even made the measurements from the exact same mic positions and averaged all of them together. I did a thread here on this and displayed the result. At that time, I used Fuzzmeasure instead of REW. For some reason the Dirac DAP messes with the sync in REW on windows and I couldn't get clean measurements through the DAP filter. It was nice and easy on a mac using Fuzzmeasure. The measurements I've done for Audiolense and Acourate were done with REW and they matched up very nicely also. DSP works well when done right.

 

[dallasjustice]

Many Auto EQ systems in contrast, create one FIR filter transform for the entire response and not a list of filter parameters that ARCOS generates. So there are no distinct filters to turn on and off. You can take their graphs and modify them but that transformation will always be there to some extent. Not ideal but heck of a lot better than not having the ability to change things.

To which software are you referring in this post?

 

[amirm]

To which software are you referring in this post?

Audyssey Pro. Dirac, etc. Indeed any system that doesn't auto generate PEQ parameters as they don't have individual correction filters.

 

[dallasjustice]

Taps

Audyssey Pro. Dirac, etc. Indeed any system that doesn't auto generate PEQ parameters as they don't have individual correction filters.

Yes, they do. It may appear to be one correction filter, but in reality the length of the filter determines the number of taps available for correction. Of course, the majority of the taps are used in the lower frequency.

Dirac uses around 30,000 taps. Audiolense usually uses around 65,000 taps and so does Acourate. The filter length can be increased if the user feels the need.

All of these softwares I've mentioned can do partial corrections, with Acourate being the most customizable. I have tried different partial corrections and I've found they never sound as good as a full frequency targeted correction.

How many taps or points of EQ/time can ARCOS do?

 

[amirm]

ARCOS uses IIR so its effective tap length is far more than FIR filters due to recursive nature of it. It uses digitized response of an analog filter so for people concerned about digitizing bits, it is the most analog of digital filters. Its effective frequency response resolution from what I recall is 1/3 of a Hertz.

What sets it apart though is that it segmented correction. The systems corrects the response using a bank of filters that are independently programmed. So any correction individually can be undone with just a check box.

With other systems you get a continuous correction curve and there is no distinct control over turning one filter say at 80 Hz on and off.

 

[dallasjustice]

ARCOS uses IIR so its effective tap length is far more than FIR filters due to recursive nature of it. It uses digitized response of an analog filter so for people concerned about digitizing bits, it is the most analog of digital filters. Its effective frequency response resolution from what I recall is 1/3 of a Hertz.

What sets it apart though is that it segmented correction. The systems corrects the response using a bank of filters that are independently programmed. So any correction individually can be undone with just a check box.

With other systems you get a continuous correction curve and there is no distinct control over turning one filter say at 80 Hz on and off.

Amir,
I think the difference between ARCOS and the other software mentioned herein, is that they are designed for different applications. I think ARCOS was meant to be used with the M2. I believe the M2's driver crossover is digitally corrected to have close to perfect step response. Therefore, ARCOS full frequency DSP may not be geared so much on trying to correct for time domain. Although Audiolense, and Acourate to a greater extent, can do the same thing with multiway active crossovers and time correction, the full range correction in each of these also compensates for time domain as well. It's just a theory, but I believe the reason I haven't liked the partial corrections on in any of these softwares is due to the time domain correction differences could muddy things up if not done for the entire bandwidth.

Michael.

 

[dallasjustice]

ARCOS uses IIR so its effective tap length is far more than FIR filters due to recursive nature of it.

This is a different topic, so you can't tell me to put a sock in it, if go on too long about this.

I prefer FIR filters. I understand that one needs many fewer taps with IIR filters because the steepness can be much greater with IIR. But there are compromises with IIR. There will always be a small phase shift with IIR filters. Whether that shift is audible is totally up for debate. FIR filters are linear phase. Of course, the downside to linear phase filters is that there's the possibility for audibile preringing. Therefore, folks using FIR filters need to carefully examine the filter's group delay (low frequency or high frequency ringing) and step response for serious pre-ringing problems. This is a judgment call, of course. One needs to listen after all to determine whether he likes the filter or not.

Dirac also uses mostly IIR. They call it mixed phase, but I think it's mostly IIR. I think IIR filters are more user friendly and can sound just as good as FIR filters, in some case.

 

[amirm]

The issue of filter type and time domain correction is highly complex and debated. In general food IIRs are much more difficult to design than FIR. High performance implementations of IIR are done and as I said that is what is in SDEC-4500 which ARCOS uses as its hardware platform. It is basically a high resolution digitized version of an analog filter.

FIR filters are expensive to implement as the number of taps increases. If too few taps are used, then the resolution in low frequencies drops. As I mentioned, it is essential that we have very high resolution for correction in bass frequencies. A FIR filter that takes the easy way out here will wind up compromising in the very thing it is asked to do: correct frequency response errors especially in bass frequencies. As you say, some systems, indeed many, used a mixed domain implementation such as IIR for low frequencies and FIR for higher.

In summary what matters is how well a filter is implemented, not what type it is.

As to effect of phase, and time domain correction, there is little objective data to back any of those techniques. Indeed, listening tests shows that we are exceptionally insensitive in this domain. In sharp contrast, we hear frequency response variations readily. It is for that reason that I so emphasize frequency response measurements and results. Time domain does not enter my vocabulary other than having the equalizer add sufficient delay to make the sound arrival the same with both speakers.

 

[andromedaaudio]

I watched the presentation , the things i find missing are :

You need a Industrial tolerance standard , to be able to compare measurements , for example a 1/6 octave smoothfactor, this is of utmost importance , like measuring distance in either meters or decimeters , he mentions 1/20 i tried them all if you look at my systems thread , you also see unsmoothed measurements , very hard to get a picture then of whats going on imo

The fact that one can achieve a flat FR at the listening spot with 2 or more types of membranes , one that sounds unnatural and one that sounds like " music" , here comes the ART thing into play, flat FR is not so hard to accomplish its merely a basic requirement in my view

Ive done all my measurments in room with different measurement locations and speaker locations in 2 different houses , on axxis , off axxis you name it ,various distances , they all boil down to be able to do accurate measurments above 300 -400 hz , he mentions that more or less at 33.00 in the vid

If i recall correctly YG doesnt measure in a acoustical chamber , they put the speaker high up in a large hall to make sure they measure direct sound , the fact whether it mathers is easibly checkable by doing some in room measurments after purchase , and compare them to the acoustical chamber measurments done by the LS company , at these price levels in high end that would be a very real thing to ask for wouldnt it
Then you know whether you have a good product (FR response wise there is obviously much more to it)or a product you happen to like (ART)


I agree with him (45.00 min) that there needs to be some sort of standarisation of the recording /reproducing process , and thus create a high end product standard and not a " REVIEWERS STANDARD"

There is another thing that i disagree with however , thats where he states that the bumps are resonances 1 hour 2 min , they could be but they also could be a variation in output of that particular driver , no driver is flat from itself , a midrange will not have exactly the same output at 1000 - 2000 or 3000 hz +- 2 or 3 db is normal /acceptable at 1/6 octave sm.
Ringing is uncontrolled resonances they usually happen at the end of the drivers useability

 

[Gregadd]

The notion that what measures better sounds better certainly is not something I want to challenge. How could anyone claim this study id unbiased or that the participants were picked at random. Talk about intellectual slight of hand. Your boss calls you on the carpet about performing poorly in a Consumer Reports test. And this what you come up with?

 

[amirm]

The notion that what measures better sounds better certainly is not something mI to challenge. How could anyone claim this study id unbiased or that the participants were picked at random. Talk about intellectual slight of hand. Your boss calls you on the carpet about performing poorly in a Consumer Reports test. And this what you come up with?

Here is the back story. They wrote a completely different report. But the day before, dog ate it so they had to cook up something else in a hurry and this is what they produced.

 

[Gregadd]

Here is the back story. They wrote a completely different report. But the day before, dog ate it so they had to cook up something else in a hurry and this is what they produced.

Apparently it worked. They're still working there. Are ou sure the dog ate it? Maybe he did something else.

 

[amirm]

Apparently it worked. They're still working there. Are ou sure the dog ate it? Maybe he did something else.

Not possible. They showed me the proof:

See how extensive the report was in that they had to make a book binding for it.

Honestly, these things are all scientific and have proof points like above. Let's not ?!?? on it please.

 

[Gregadd]

 

[Gregadd]

Nobody would refer to Consumer Reports as gullible audiophiles. So what was wrong their test?. If listeners reward you for the honoring certain principle. why has not Harmon been rewarded.
Let us suppose I started a thread telling people "you could listen around the room. Can you imagine the response?
What is his scientific explanation for why mono is better than stereo -"t just is"[/I

BTW= It is no question people are horrible at measurement. That is why we developed an arbitrary system of weights and measures and devices and machines to do it for us. Then we still get it wrong

 

[amirm]

Nobody would refer to Consumer Reports as gullible audiophiles. So what was wrong their test?

This was the problem with their test:

To save money (subscriptions were low that year), they used the same subjects as the painkiller test and Harman found out they were all half asleep when evaluating loudspeakers. Here is proof from AES paper, "A Multiple Regression Model for Predicting Loudspeaker Preference Using Objective Measurements: Part II - Development of the Model" (honest! not making up that title).

"The second part to research question three asks
whether models based on sound power measurements
can accurately predict loudspeaker preference ratings.
The CU [Consumer Reports] model tested in Part One is only one
example of a model based on sound power. First, we
re-visit the CU model to explain why it failed to
accurately predict loudspeaker preference."

When we use non-sleep induced listeners, we get to a new model from Harman based on different set of measurements that do predict listener preference to very high degree:

"A new model has been developed that accurately
predicts preference ratings of loudspeakers based on
their anechoic measured frequency response. Our
model produced near-perfect correlation (r = 0.995)
with measured preferences based on a sample of 13
loudspeakers reported in Part One. Our generalized
model produced a correlation of 0.86 using a
sample of 70 loudspeakers evaluated in 19 listening
tests. Higher correlations may be possible as we
improve the accuracy and resolution of our subjective
measurements,which is a current limiting factor."

 

[Gregadd]

This was the problem with their test:

To save money (subscriptions were low that year), they used the same subjects as the painkiller test and Harman found out they were all half asleep when evaluating loudspeakers. Here is proof from AES paper, "A Multiple Regression Model for Predicting Loudspeaker Preference Using Objective Measurements: Part II - Development of the Model" (honest! not making up that title).

"The second part to research question three asks
whether models based on sound power measurements
can accurately predict loudspeaker preference ratings.
The CU [Consumer Reports] model tested in Part One is only one
example of a model based on sound power. First, we
re-visit the CU model to explain why it failed to
accurately predict loudspeaker preference."

When we use non-sleep induced listeners, we get to a new model from Harman based on different set of measurements that do predict listener preference to very high degree:

"A new model has been developed that accurately
predicts preference ratings of loudspeakers based on
their anechoic measured frequency response. Our
model produced near-perfect correlation (r = 0.995)
with measured preferences based on a sample of 13
loudspeakers reported in Part One. Our generalized
model produced a correlation of 0.86 using a
sample of 70 loudspeakers evaluated in 19 listening
tests. Higher correlations may be possible as we
improve the accuracy and resolution of our subjective
measurements,which is a current limiting factor."

Toole never said anything about the panel being half asleep. He said he the test was wrong and that he could teach them to do it properly. Having "half asleep participants (whatever that means) would be a self evident error. Since you bring it up the make up of the test group was hardly iideal . It would not even remotely qualify as a random selection. That is what you would need to extrapolate the results to the general public.

 

[amirm]

Toole never said anything about the panel being half asleep. He said he the test was wrong and that he could teach them to do it properly. Having "half asleep participants (whatever that means) would be a self evident error.

Not if you yourself had dipped into the bottle. As to Toole, do you want me to continue to give you the inside scoop or not???

 

[Gregadd]

Not if you yourself had dipped into the bottle. As to Toole, do you want me to continue to give you the inside scoop or not???

Tnat would be an interesting variation on double blind.



Sure I always want insider information. Of course you realize proving someone else wrong does not make the other person right.

 

[Gregadd]

How does off axis response effect one party straight on monophonic listening>