I agree with that. Note: I Saw a finnish research paper pointing that 0.5 millisecond delay is audible under circumstances. Have to find the link though.
You're probably talking about this Audibility of Group-Delay Equalization
I'm not worried that some feature such as deviation in group delay is audible. Audibility does not mean that speaker is unacceptably bad. That limit depends on other features, personal preference and interests such as favorite music. GD deviation is one of them, and some other features could either increase or decrease requirements for GD. For example directivity and dynamics of the drivers could be so fantastic at low mid that GD of 1.5 ms at 100 Hz is easily acceptable. On the other hand, 5 ms at 100 Hz is so much that for example I would never accept such a product. 0.5 ms is not easy to achieve down to 100 Hz as conventional 3-4-way with IIR XO so we better design speakers otherwise so well that close to ideal timing is not mandatory.
In addition, headphones are limited for detecting problems related to timing. Consequences are more than just audible by ear.
Could you please explain a bit more about how we can identify dynamics of drivers.
Thanks
Right, just reading along.
It's important to put everything in perspective, as always (which seems something 99.8% of people just simply neglect
)
First of, it's not just group delays, it's about small (local) band group delays on certain impulse like signals.
Totally different animal.
Next to that;
"Group-delay audibility was tested at the frequencies 500 Hz, 1 kHz, 2 kHz, 3 kHz, and 4 kHz. The results indicate that the audibility thresholds for local group-delay variation are less than ±1 ms for the most critical signals, and approximately 1.5 ms to 4.5 ms for a local positive group-delay peak and between −1.0 ms and −2.3 ms for a local negative group-delay peak for real-life signals."
"The thresholds found in our study for the real-life signals were either the same or slightly higher than the ones reported in other studies."
Positive group-delays don't really happen practically speaking.
So why 100Hz is being called up here is totally strange, because that was never tested in this paper to begin with.
Conclusions with those kind of signals are not allowed to be made.
100Hz signals also are generally (in music) not considered as impulse signals anymore.
I have to look up the data they used in this experiment, but they mention something about going with only 1 sigma.
Which is rather sad to begin with, that's just slightly more than pure gambling.
In cancer treatments they are super happy to go along with it, but the rest of the world 2 or 3 sigma is the (bare) minimum.
But again I have to read the further context of it.
Also don't forget those signals were tested in an ideal environment with headphones.
Which can't be simply extrapolated to a practical listening room (with all the background noise and other issues)
Last of all, unless I accidentally skipped over it, the experiment didn't test something more important, what was more audible.
The hump in frequency response that has to be corrected vs the slightly noticeable artifact of a small band group delay that goes with it.
It's important to put everything in perspective, as always (which seems something 99.8% of people just simply neglect
)First of, it's not just group delays, it's about small (local) band group delays on certain impulse like signals.
Totally different animal.
Next to that;
"Group-delay audibility was tested at the frequencies 500 Hz, 1 kHz, 2 kHz, 3 kHz, and 4 kHz. The results indicate that the audibility thresholds for local group-delay variation are less than ±1 ms for the most critical signals, and approximately 1.5 ms to 4.5 ms for a local positive group-delay peak and between −1.0 ms and −2.3 ms for a local negative group-delay peak for real-life signals."
"The thresholds found in our study for the real-life signals were either the same or slightly higher than the ones reported in other studies."
Positive group-delays don't really happen practically speaking.
So why 100Hz is being called up here is totally strange, because that was never tested in this paper to begin with.
Conclusions with those kind of signals are not allowed to be made.
100Hz signals also are generally (in music) not considered as impulse signals anymore.
I have to look up the data they used in this experiment, but they mention something about going with only 1 sigma.
Which is rather sad to begin with, that's just slightly more than pure gambling.
In cancer treatments they are super happy to go along with it, but the rest of the world 2 or 3 sigma is the (bare) minimum.
But again I have to read the further context of it.
Also don't forget those signals were tested in an ideal environment with headphones.
Which can't be simply extrapolated to a practical listening room (with all the background noise and other issues)
Last of all, unless I accidentally skipped over it, the experiment didn't test something more important, what was more audible.
The hump in frequency response that has to be corrected vs the slightly noticeable artifact of a small band group delay that goes with it.
Last edited:
Ok, well it's written on a slightly confusing way (or for me at least).
But I was missing the statistical data of the amount of "guesswork" , aka be able to successfully hear (or guess) the ABX test. (some call that the success rate I think)
I already got the sense the the researchers were actually just interested in an threshold number, not how easy/difficult it is to hear such a thing.
Two things clarify this;
"The ABX listening test is a Bernoulli trial [52], i.e., an
experiment with exactly two outcomes: “success” or “failure”.
The binomial proportion confidence interval [53] can be used
to estimate the interval of a success probability, and only the
number of experiments and number of successes are required
for this."
But maybe more important the statement of the practical test itself;
"Finally, we consider some of the verbal comments given
by the test subjects. The test was generally considered hard,
since there were many test cases where no differences could
be heard. At the same time, most test subjects said that they
learned to distinguish the audible cues after a certain number
of sounds, which helped in the test. They commented that the
most noticeable clue was the chirp-like property of some sounds.
Many test subjects commented that the test felt binary: one either
heard the difference immediately or did not hear it at all. They
noticed that listening to a test sound multiple times did not help
them. There was no consensus about the hardest test signal, since
each of the four test signals was named as the hardest one by
different subjects. The comments from the few subjects who
took the test twice did not change between the rounds. The test
was still considered hard, but due to the familiar test sounds
and the knowledge about the binary nature of the test, the time
required for the second test decreased considerably, which was
noted in the comments."
Also, the amount of testers (12 only in total, one was excluded) is statistically not really significant (although this is often the case unfortunately). The age group is pretty tight as well.
In the end I read this article as;
We found the same results as other research showed before, except for a few specific signals were those threshold differ slightly (their own words btw)
Most important!!! All in the context of a super hyper controlled environment seen from a point of view to determine those threshold values (and nothing more) NOT how audible they are in a bigger sense or statistical setting.
So it doesn't hold much value in the sense of speaker building at all.
But I was missing the statistical data of the amount of "guesswork" , aka be able to successfully hear (or guess) the ABX test. (some call that the success rate I think)
I already got the sense the the researchers were actually just interested in an threshold number, not how easy/difficult it is to hear such a thing.
Two things clarify this;
"The ABX listening test is a Bernoulli trial [52], i.e., an
experiment with exactly two outcomes: “success” or “failure”.
The binomial proportion confidence interval [53] can be used
to estimate the interval of a success probability, and only the
number of experiments and number of successes are required
for this."
But maybe more important the statement of the practical test itself;
"Finally, we consider some of the verbal comments given
by the test subjects. The test was generally considered hard,
since there were many test cases where no differences could
be heard. At the same time, most test subjects said that they
learned to distinguish the audible cues after a certain number
of sounds, which helped in the test. They commented that the
most noticeable clue was the chirp-like property of some sounds.
Many test subjects commented that the test felt binary: one either
heard the difference immediately or did not hear it at all. They
noticed that listening to a test sound multiple times did not help
them. There was no consensus about the hardest test signal, since
each of the four test signals was named as the hardest one by
different subjects. The comments from the few subjects who
took the test twice did not change between the rounds. The test
was still considered hard, but due to the familiar test sounds
and the knowledge about the binary nature of the test, the time
required for the second test decreased considerably, which was
noted in the comments."
Also, the amount of testers (12 only in total, one was excluded) is statistically not really significant (although this is often the case unfortunately). The age group is pretty tight as well.
In the end I read this article as;
We found the same results as other research showed before, except for a few specific signals were those threshold differ slightly (their own words btw)
Most important!!! All in the context of a super hyper controlled environment seen from a point of view to determine those threshold values (and nothing more) NOT how audible they are in a bigger sense or statistical setting.
So it doesn't hold much value in the sense of speaker building at all.
I don't have any clear plan or suggestion how compression should be measured. Common method is to measure SPL spectrum or spectrum of fundamental and few harmonics with multiple levels. Normalized by the response measured with the lowest level shows response linearity. This kind of measurements can be done to single driver or complete speaker. Main problem is that signal with constant level does not represent spectrum while typical use. It does not reveal how much for example LF drops compared to HF with some music such as classical orchestra, death metal or RnB. Result is usually too optimistic compared to disaster in real life where about 15" vented woofer is compatible with 1" alu dome. HF with compression driver needs few 18" woofers to be balanced at high SPL without bleeding ears.
M-Noise would be more practical though spectrum is even more demanding for LF than limited pink. It's just noise so measurement is simple with spectrum analyzer. Reality with casual muzak is something between previous two.
Identifying dynamic drivers without measurements is partly guessing. Drivers with high efficiency produce less heat per Pascal so relative thermal compression should be low if motor has also proper ventilation. High efficiency is produced with large and light enough cone, strong motor and low mechanical and electrical losses. High maximum SPL requires also some excursion capacity though cone area or possible horn compensates that.
You're the only one who is "discussing" about that paper. I just linked the document.
Thanks a lot Kimmosto.
If possible, could you suggest any example of a driver having good dynamics/specs as mentioned above.
I can only think about pro audio drivers hearing the criteria.
Why linking a paper if it doesn't have any benefits to a discussion?
You might as well share some pictures of some cute cats?
That being said, it's still very valuable, if in a hyper controlled environment at those frequencies those values are barely noticeable, we automatically can conclude they won't at lower frequencies in a practical environment.
Another GD study: Audibility of Loudspeaker Group-Delay Characteristics
It's few years older and also listened with HD-650, but there are some differences. Results are visualized as number of ways and slope orders of multi-way speakers. Not entirely the same thing as listening with speakers including also other senses, but much better than nothing.
It's few years older and also listened with HD-650, but there are some differences. Results are visualized as number of ways and slope orders of multi-way speakers. Not entirely the same thing as listening with speakers including also other senses, but much better than nothing.
That is safe assumption because hifi-drivers are focusing to LF extension and resonance handling
Some manufacturers prefer low mechanical losses which is good but BL could be low, voice coil quite small and not so perfectly ventilated.
Than it would be at least adequate to let people know that those studies only have something to do with finding the absolute theoretical minimum values.
As said above, they don't hold any practical value in building loudspeakers or sound reproduction at all.
So discussing the numbers also doesn't hold any ground. (except for pure believing in the idea that it will be beneficial)
It would be interesting to see those absolute values against some practical tests in a real environment with trained/skilled listeners as well as casual listeners. Still there is a big difference between detectable and having a negative impact on the sound impression (especially when the listener isn't aware of it)
Well i would not know. If Griesinger mentions that interaural time differences of 5-6 usec are noticable, i find it strange that group delay of 1 msec of more is about observable.
I still believe we have not yet understood the role of time in discrimination of sounds with low frequencies.
Anyhow i cannot change this.
Member
Joined 2003
Bohdan of Bodzio software published this paper a few years back regarding low frequency timing and “punchy bass”. It may be a bit biased paper however since it is in promotion of the FIR filtering ability of his Ultimate EQ software.
https://bodziosoftware.com.au/Perfecting_Punch.pdf
(of course, we have determined earlier here that the filtering ability of UE can be covered by VituixCAD + APO EQ quite easily as well)
https://bodziosoftware.com.au/Perfecting_Punch.pdf
(of course, we have determined earlier here that the filtering ability of UE can be covered by VituixCAD + APO EQ quite easily as well)
You may conclude automatically what ever you want, but at least I disagree. Sound is local pressure deviation from ambient as a function of time, and music is not continuous/unchanging signal so it's perfectly natural that time includes more than just frequency after Fourier Transform. Deviation in timing causes for example energy diffusion and probable reduction to transients' peak pressure, change in perceived sound balance and location of sound images (HF scattering and locating to speakers). Speakers with long excess group delay at LF have been lame crap showing all those problems very clearly. Most of 3-ways have some difficulties depending on other features such as XO orders though comparing different models and types is not controlled due to other differences. The same features are almost flawless in 2-ways. My conclusion is that Genelec has (finally) selected exactly right path by rather testing than just believing someone else believer.
Content is just perfect though paper is written to justify the feature.
I use piano instead of percussion for testing because many piano recordings have clear hammer and frequency range of fundamental is wide. Strength and force of the hammer is easy to sense on the skin and eye balls. Easier than by ears in my opinion (which is the reason why I don't care much tests using headphones). Strong right hand but weak left hand indicates that GD is okay at HF, but probably not at LF. Both hands weak is multi-way with all slopes very steep. This kind of speakers suck no matter how perfect spinorama, off-axis and on-axis magnitude responses are. Quite small speaker with minimum phase response (by FIR or TP) or conventional 2-way with high XO such as Kef Q100 can easily beat big 3-way main monitors with 15" woofer and very steep IIR XO slopes.
Last edited:
Member
Joined 2003