Unfortunately it just explores topics like square wave response and time-coherency among other things - without referencing proper testing supporting their reliable audibility. It reads as something written by someone who makes empirical goals based on EE dogma, but doesn't want to usefully weigh their significance. Yes, I realize the benefits of omni speakers he was advocating, but he seems to disregard their drawbacks. He mentions how great the old Quad electrostats used to be in certain categories but ultimately fails to tell me how great the value in those categories actually is.
Basically, he told me a lot of "most loudspeakers are really bad at x! Only cabinet makers would ignore X!".
I don't mean to discredit him, but I'm offended at how vehemently he pushes his priorities without substantiating their relevance to audibility.
And don't get me wrong - I'm all for exploring different areas. But don't pre-suppose that improving X or Y - probably at the cost of A or B - is more scientific unless you can show me the perceptual research backing the position. Because all that I've read tends to disagree with him more often than not.
Last edited:
The "grills" are just something like a stocking BTW. He wanted to build a speaker that is looking like an ordinary box speaker to Joe average, because people are used to the looks of square boxes. But the actual "boxes" used are cylinders.
Regarding transient-response I'd say that it is better to improve it as much as possible as long as there isn't any degradation in another area.
Maybe some day someone will come up with a reliable model - describing the thresholds of audibility of group delay - that will allow us to make the right compromises in terms of transient response while still achieving a "blameless" loudspeaker.
Theoretically most modern DSP crossover platforms (also the cheaper ones) would be able to support transient-perfect crossover solutions. Unfortunately the GUIs delivered with them don't generally support such crossover topologies.
Quoting John Watkinson: "A highpass and a lowpass are not a crossover - they are just a pair of filters !"
Regards
Charles
Regarding transient-response I'd say that it is better to improve it as much as possible as long as there isn't any degradation in another area.
Maybe some day someone will come up with a reliable model - describing the thresholds of audibility of group delay - that will allow us to make the right compromises in terms of transient response while still achieving a "blameless" loudspeaker.
Theoretically most modern DSP crossover platforms (also the cheaper ones) would be able to support transient-perfect crossover solutions. Unfortunately the GUIs delivered with them don't generally support such crossover topologies.
Quoting John Watkinson: "A highpass and a lowpass are not a crossover - they are just a pair of filters !"
Regards
Charles
Last edited:
I think transient perfect is not the solution. There needs to be some way to integrate with the limitations of the drivers to provide a transient perfect solution. There are still many trade off considerations in the design process. But it is an interesting process when you start implementing an accumulation set of solutions one by one and see how the sound evolves.
There are transient-perfect crossovers that can be built in the analog domain without problem. These don't have steep slopes in the vicinity of the crossover frequency and they also have quite some overlap. It is possible to build transfer functions that achieve such a crossover function together with the natural responses of the drivers involved. It is depending on the drivers used whether this setup is feasible. Generally spoken: It would be quite easy to use them with a FAST topology but it would be difficult to use them with a woofer and dome tweeter combination.
Then there are those transient perfect crossovers that are very difficult to implement in the analog domain but theoretically very easy in the digital domain. These offer steeper slopes than the aforementioned analog ones. With those steeper slopes it is quite easy to take the driver's natural response into consideration: They do just have to be EQed over a reasonable frequency range. Because lowpass filters have fairly constant group delay within their passband the natural lowpass behaviour of a woofer can aproximately be compensated for by delaying the tweeter accordingly. Delay is some very simple task for DSP.
Regards
Charles
Then there are those transient perfect crossovers that are very difficult to implement in the analog domain but theoretically very easy in the digital domain. These offer steeper slopes than the aforementioned analog ones. With those steeper slopes it is quite easy to take the driver's natural response into consideration: They do just have to be EQed over a reasonable frequency range. Because lowpass filters have fairly constant group delay within their passband the natural lowpass behaviour of a woofer can aproximately be compensated for by delaying the tweeter accordingly. Delay is some very simple task for DSP.
Regards
Charles
Last edited:
Charles - those thresholds are known - see the work of Brian Moore. The problem is that they are dependent on the absolute SPL, thus making the thresholds nonlinear. This make them very hard to deal with.
But transient response and group delay are different issues.
Some years back we did blind subjective tests showing that the thresholds were level dependent but we did not deal with absolute numbers at that point. What we wanted to show was that group delay could sound like nonlinear distortion because of this effect. This makes claims about nonlinear distortion audibility highly suspect.
Earl - I am aware that transient response and group delay distortion is not the same but they are interrelated.
A transient perfect speaker will not have any group delay distortion. OTOH a system with group delay distortion can't be transient-perfect.
But it would IMHO be easier to tackle the problem from the group-delay point-of-view because it is easier to specify group delay distortion than transient non-perfectness (or whatever one would call this).
I will have a look at the figures Brian Moore came up with.
Regards
Charles
A transient perfect speaker will not have any group delay distortion. OTOH a system with group delay distortion can't be transient-perfect.
But it would IMHO be easier to tackle the problem from the group-delay point-of-view because it is easier to specify group delay distortion than transient non-perfectness (or whatever one would call this).
I will have a look at the figures Brian Moore came up with.
Regards
Charles
I just had a look at his JAES paper. He used high-Q all pass filters to generate the group delay distortion. It is basically the same method as used by Blauert and Laws. This gives basically a small but fairly constant group delay below the pole frequency with a narrow peaking group delay at the pole frequency itself. Basically it has the same shape as a high-Q lowpass filter's amplitude response. But with these high Q values involved one could easily approximate this as a small ( i.e. narrow-band) peak in group delay.
What I would be interested in is the audibility of the kind of group-delay distortion that is introduced by ordinary crossovers. These do also have "constant" group delay below the pole frequency but with considerably higher figures than the all pass filters used for Moore's tests combined with a much lower peak group delay than the test all pass filters.
Generally one could say that the frequency range below the crossover frequency is delayed against the frequencies above the crossover frequency. While the group delay above crossover is strongly frequency dependant.
But I have not found a definition of audibility of this crossover caused kind of group-delay distortion so far.
Regards
Charles
What I would be interested in is the audibility of the kind of group-delay distortion that is introduced by ordinary crossovers. These do also have "constant" group delay below the pole frequency but with considerably higher figures than the all pass filters used for Moore's tests combined with a much lower peak group delay than the test all pass filters.
Generally one could say that the frequency range below the crossover frequency is delayed against the frequencies above the crossover frequency. While the group delay above crossover is strongly frequency dependant.
But I have not found a definition of audibility of this crossover caused kind of group-delay distortion so far.
Regards
Charles
Agreed.
But there is still no evidence that "transient perfect" is audible (that I am aware of. I'd love to read about it if there is.) There some evidence that certain forms of excess group delay are, like excess group delay in a waveguide. I know that is audible, I've tested it in controlled conditions.
But there is still no evidence that "transient perfect" is audible (that I am aware of. I'd love to read about it if there is.) There some evidence that certain forms of excess group delay are, like excess group delay in a waveguide. I know that is audible, I've tested it in controlled conditions.
How much phase-shift could be indistinguishable from a transient-perfect system?
I remember from somewhere (sorry, anecdotal, can't seem to find the paper) that some signals can be distinguishable from the original if the phase-shift inside a critical bandwidth exceeds 15deg. So you could have 15?*24=360deg phase shift from low to high and be indistinguishable from the original even with special test signals.
Earls findings supersedes this - abrupt phase/gd changes get unmasked at higher listening levels.
I remember from somewhere (sorry, anecdotal, can't seem to find the paper) that some signals can be distinguishable from the original if the phase-shift inside a critical bandwidth exceeds 15deg. So you could have 15?*24=360deg phase shift from low to high and be indistinguishable from the original even with special test signals.
Earls findings supersedes this - abrupt phase/gd changes get unmasked at higher listening levels.
Exactly - me too. But we can't assume it is audible just because it is there.
The rePhase software might be able to help there. It allows phase correction without level changes. I've done a few blind tests and it does seem (just audible. But a lot more testing would be needed before I'd say for sure.
But only at a single point in space. All other points are made worse.
I'm glad you said that. Just make sure they are blind or there isn't any point in doing more than one.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.