Figure 9 is a great example of how your metric shows the low importance of a standard THD measurement.
The industry should really pick this up, they owe the customer as well as themselves.
My bet is that some companies start using it to display the better results than their competitors. Or AP should get a license from you(?). What's to be lost for the industry? Nothing. It would be similar to a SNR measurement with a-weighting.
Perhaps we, as customers, could ask for the Gedlee metric in the comment section of audio reviews.
Dr. Geddes, have you contacted Audiomatica (CLIO)?
Single-box hardware+software integrated solution (CLIO) would have much bigger impact on both industry and DIY, than Virtins Technology Multi Instrument software alone.
I've mentioned this before - I think there should be TWO numbers to characterize harmonic distortion, one for the amount of lower harmonics (maybe first three to five), and the other for the amount of higher harmonics. This would easily distinguish between "euphonic" distortion (generated by SETs, woofers) and "undesirable, harsh" distortion (clipping, zero-crossing).
Did you leave out the relation to level on purpose?
This can be sound pressure, voltage, whatever you like.
Because the perceived distortion is also dependent on absolute level played.
Iow, there's more than just some lower and some higher harmonics at play.
Right after Lidia and I gave those papers, I contacted AP about implementing it in their systems. The basic response was :"Our customers like THD. It suites their purposes." Sad, but true.
There’s only a few people who are doing listening first AND then takes measurements afterwards, which (s)he then tries to correlate and finding meaning for.
How do you implement it Earl? Perhaps you could talk to @bikinpunk who runs erinsaudiocorner.com , his own YouTube channel and Patreon club. He has stated that THD is not a good measure of how transducers or speaker sound, and he continues to devise different ways, the latest being the Klippel multi-tone test, but that doesn’t take into account psychoacoustic models.
Yevgeniy Kozhushko @HiFiCompass is another who takes extensive data, although he doesn’t have many drivers of 10-15” that interests you and many of us. but your GedLee metric might be able to be applied retrospectively to the data he has already collected (H2 to H5 at different voltage drive levels, FR at different drive levels etc). That might help sort out the better 5-8” drivers from the rest.
Everyone is going smaller these days…
How do you implement it Earl? Perhaps you could talk to @bikinpunk who runs erinsaudiocorner.com , his own YouTube channel and Patreon club. He has stated that THD is not a good measure of how transducers or speaker sound, and he continues to devise different ways, the latest being the Klippel multi-tone test, but that doesn’t take into account psychoacoustic models.
Yevgeniy Kozhushko @HiFiCompass is another who takes extensive data, although he doesn’t have many drivers of 10-15” that interests you and many of us. but your GedLee metric might be able to be applied retrospectively to the data he has already collected (H2 to H5 at different voltage drive levels, FR at different drive levels etc). That might help sort out the better 5-8” drivers from the rest.
Everyone is going smaller these days…
Last edited:
No, I wasn't even thinking about it.
If that's so (and yes, I believe it is, but ...) then the only absolute level that could be effectively used for volume-dependent distortion is dbSPL. These others are just relative, and all you can say is "perceived distortion increases with volume level."
There's always more at play, the idea isn't fully fleshed out, but IMHO it's less gross (all puns intended) and more useful than a single THD figure.
THD is a fine figure of merit. I have never experienced SQ going up with increasing distortion. I only experienced SQ going up with decreasing distortion. This is also born out by loudspeaker industry investigations into the matter. Unfortunately, I am not aware of any published work in this regard.
In order of importance for loudspeaker quality: frequency response, power response, and third, harmonic distortion. For a system to be top notch, all three will have to be optimized.
There might be reasons why the Gedlee approach to distortion is not finding wide acceptance. For one, I have not seen his results validated by other independent research. For another, they conflict with the experiences of those who have been successful in lowering distortion in transducers. People generally like speakers better when they distort less.
In order of importance for loudspeaker quality: frequency response, power response, and third, harmonic distortion. For a system to be top notch, all three will have to be optimized.
There might be reasons why the Gedlee approach to distortion is not finding wide acceptance. For one, I have not seen his results validated by other independent research. For another, they conflict with the experiences of those who have been successful in lowering distortion in transducers. People generally like speakers better when they distort less.
"perceived distortion increases with volume level."
You could watch the YT podcast and find out it's different from what you think ;-)
I'm not sure what you're getting at.
You seem to say THD is fine, let's not apply any research done that deepens the meaning of the spectrum, suited best for the ears. The industry has a tool that transforms the meaning of the raw data (THD) to the actual meaning for the receiver(ears and brain).
It shouldn't be applied to bats or radar installations, it does e.g. however, make r&d more efficient as it gives insights in what budget should be used for what research, as it can be that now research is being done on no perceivable benefits. Etc.
No, I am not saying that. I am saying that Dr. Geddes is overunderestimating the impact of distortion in loudspeakers. The fact that different harmonics have different impact is well known. THD is just fine as a figure of merit with me and apparently with others as well.
In simpel words, all harmonic distortions are evil and should be combatted. Some even more than others.
In simpel words, all harmonic distortions are evil and should be combatted. Some even more than others.
Most loudspeaker engineers agree with my conclusions - Floyd Toole, Sean Olive, Alex Vishvillo. I think that it is your position that is in the minority, not mine.
There is unquestionably a point below which THD becomes inaudible. Without a good metric it's just a guess where that point is.
..especially because THD is often mostly composed of 2nd order distortion, and it takes a lot of it in a driver for the listener to actually become aware of 2nd order distortion (..particularly if you are any *significant distance from the driver). (..and of course THD isn't telling you what the higher order products are like relative to the average.)
*most of the subjective non-linear testing was with extreme nearfield headphone testing, and a loudspeaker in-room is orders of magnitude more difficult to assess.
Still, I think that it's fine to continue on with 2nd though 5th order and various mult-tone tests because it does show fundamental limitations of the driver that can be then be further assessed against some metric with regard to audibility. It is typically however of significantly less importance than many other aspects of a driver's design/results (mostly linearity both on and off-axis), and I don't think enough people interested in this put enough "weight" into those other aspects - instead **spending way to much time/interest into non-linear effects.
**it's understandable though, because for a given driver/type linearity is often very similar (or can be somewhat corrected) for similar drivers - at that point then you are left with "nit-picking" non-linear effects.
*most of the subjective non-linear testing was with extreme nearfield headphone testing, and a loudspeaker in-room is orders of magnitude more difficult to assess.
Still, I think that it's fine to continue on with 2nd though 5th order and various mult-tone tests because it does show fundamental limitations of the driver that can be then be further assessed against some metric with regard to audibility. It is typically however of significantly less importance than many other aspects of a driver's design/results (mostly linearity both on and off-axis), and I don't think enough people interested in this put enough "weight" into those other aspects - instead **spending way to much time/interest into non-linear effects.
**it's understandable though, because for a given driver/type linearity is often very similar (or can be somewhat corrected) for similar drivers - at that point then you are left with "nit-picking" non-linear effects.
Last edited:
More difficult, yes, but not "orders of magnitude".
Many people have hung their hat on the assumptions that I made in producing this work, like non-hysteresis, .... Over the years I've thought a lot about the assumptions and each and every one is almost trivial to resolve (now.) I just didn't do that at the time because "simple" experiments seem to always work better than a massive comlex one (been there, done that.) It's curious that no one has ever asked me how "hysteresis" should be handled, just not interested I guess. The lights better over there.
Simply by doing Gm(f), i.e. calculating the value of Gm at each frequency. Modern stepped sign analyzers can do that no sweat.
The original Gm data used an assumption that G(f) was flat. Hence they are the same on average, like SPL versus frequency spectrum. Hence, if you want a single number then Gm is right. The Gm(f) spectrum may tell you a lot more, but I've never looked at it.
The original Gm data used an assumption that G(f) was flat. Hence they are the same on average, like SPL versus frequency spectrum. Hence, if you want a single number then Gm is right. The Gm(f) spectrum may tell you a lot more, but I've never looked at it.