I just came across an impressive Power-Point from Geddes and Lee.
Recognizing that auditory masking (by your brain) hides the noise products adjacent to a test tone, those noises really don't count as much towards your annoyance. So they devised a means of calculating a sound quality metric which downplays adjacent noises.
Their tests revealed far better correlation to human judgments of annoyance compared to rather poorish relation to human judgments using THD and IMD.
The evidence looked good to me. But how would you apply their method to one's own speakers within, say, REW? Would you attend to 3 HD more than 2 HD? Or is that exactly why people have forever found that 3 HD is more annoying than 2 HD... even if they love the sound of an English horn?
Sorry, no link. Can somebody please supply one?
What is the SOTA on better methods of characterizing sound quality?*
B.
* other elderly scholarly enthusiasts will recall the discussions over the introduction of IMD measurements 80 years ago.
Recognizing that auditory masking (by your brain) hides the noise products adjacent to a test tone, those noises really don't count as much towards your annoyance. So they devised a means of calculating a sound quality metric which downplays adjacent noises.
Their tests revealed far better correlation to human judgments of annoyance compared to rather poorish relation to human judgments using THD and IMD.
The evidence looked good to me. But how would you apply their method to one's own speakers within, say, REW? Would you attend to 3 HD more than 2 HD? Or is that exactly why people have forever found that 3 HD is more annoying than 2 HD... even if they love the sound of an English horn?
Sorry, no link. Can somebody please supply one?
What is the SOTA on better methods of characterizing sound quality?*
B.
* other elderly scholarly enthusiasts will recall the discussions over the introduction of IMD measurements 80 years ago.
Last edited:
I haven't seen the paper, but I have some ideas about masking.
If you think about that threshold between 2 closely spaced notes, versus the sound of one note with a tremolo, then it becomes clear that the 'masked' frequencies manifest as amplitude modulation. However, since there's a bit of debate about how (in?)sensitive people are to AM, where doubling (or halving) the power is considered to be 'just' 3dB, this suggests that that's where most of the distortion goes when it comes to streaming formats.
But what does that all mean in our quest for better sound quality? (Not just the corporations' quest to squeeze more value for them out of every byte that they stream).
When it comes to THD, I have reasons to think that the amplifier and speaker components should not be treated as separate. I'm actually in the middle of a project, and I'm finding reactive speaker loads can have a huge effect on amplifier THD in simulations. It's possible to fine-tune the amplifier to fit the speaker, but if you change the speaker then you're back to square one. So, in that sense, active speakers have more potential than components that can be freely mixed.
Yes, sound quality assessment methods can not be separated from compression algorithms, as with Dolby, they are twins. The smartest compression wins the race.
Do you believe a streaming system can separate 2-harmonics from 3-harmonics?
B.
When you speak of "reactive loads" do you have in mind the dozen little components in a passive crossover in addition to the phase tricks of the VC?
For sure, that and a whole bunch of other reasons make it compelling to use multi-amping.
B.
Hello,
GedLee tells us that that the major contributors, 2nd Harmonic and 3rd Harmonic, to THD. are masked by the test signal. He says that 2ndH and 3rdH are not contributors to what our ear / brains perceive as distortion. Our ear / brains do not hear care or give weight to 2ndH and 3rdH distortions. GedLee has published AES papers that prove this.
Things that do not fit under the umbrella of the masking primary signal are processed by our ear / brain as not part of the illusion of music and sound really bad from a distortion point of hearing.
Thanks DT
GedLee tells us that that the major contributors, 2nd Harmonic and 3rd Harmonic, to THD. are masked by the test signal. He says that 2ndH and 3rdH are not contributors to what our ear / brains perceive as distortion. Our ear / brains do not hear care or give weight to 2ndH and 3rdH distortions. GedLee has published AES papers that prove this.
Things that do not fit under the umbrella of the masking primary signal are processed by our ear / brain as not part of the illusion of music and sound really bad from a distortion point of hearing.
Thanks DT
With amplifier testing and measurements, the spectrum that shows an amp that has higher 3rd order or even the slight presence of higher odd orders such as 5th, 7th, 9th even at minuscule levels will generally correlate with an amp that is fatiguing to listen to. An amp with only 2nd order and a small amount of 3rd order and minoronoxally descending higher orders will sound more pleasing even if distortion is in the 0.05% range vs amp with low overall THd under 0.001% but dominant 3rd order and with contribution from higher order odd harmonics. It’s is very measurable and predictable.
Without disputing your anecdote what so ever, the question Geddes and Lee are working on is what kind of physical measurement has a conceptual foundation (say, related to masking) and then, how to prove the measurement relates to fatigue or sonic displeasure otherwise.
As I hinted in post #1, as long ago as I can recall, folks regarded odd harmonics as yucky and even harmonics were kind of music-like. In days of yore, like with my old Heath THD tester, it was a simple matter of filtering out the fundamental - and magic - the rest is THD and noise. With modern fast sine analysis, you can display the separate harmonics.
Based on your observations, what recipe for a measurement would you propose?
B.
There are 4 key aspects that are important: frequency response flatness, relative amplitudes of 2nd, 3rd, 4th, 5th, 7th harmonics in distortion measurement with 1kHz sine wave excitation, transient reproduction ability, and flatness of phase of the measured response.
The first, flatness of the frequency response within +/-2.5dB or 5dB total variation is probably most audible and if not correct or having excessive peaks, will not sound good. Also sometimes having flat but tilted response higher in bass and slowly falling off by -5dB from 20Hz to 20kHz is not a bad thing.
The second, look for dominant 2nd order and at reference level of say 90dB at listening position, should not be more than -40dB for a speaker (1%) and should be monotonicall descending with higher order harmonics. 3rd order should be st least 12dB lower than 2nd order and so on. However, on a speaker with only 2nd and 3rd order, and nothing higher, they can be within 6dB and sound quite good.
The third is the transient ability. It should reproduce crisp drum rim shots, bongo, and piano strikes or string plucks. This can be measured easily by looking for a right triangle shaped step response function. Transient perfect speakers like Dunlavey SC-IV will do this perfectly. Look at step response of s typical 4th order LR speaker and it looks like a jumbled mess. Until you hear a transient perfect speaker you won’t know what it sounds like. Fortunately, all single driver full range speakers will be transient perfect for the most part. It should have measured group delay variation from 100Hz to 50Hz of less than 7ms, ideally 2ms.
Fourth is flatness of phase of the measured response - especially in the “telephone” band from circa 500Hz to 5kHz (as high as 10kHz). This is where imaging and soundstage come from. Mess up the phase and then imaging is imprecise. Should be within about +/-15deg (ideally +/-5deg) over 500Hz to 5kHz for the speaker.
So there you have 4 measurements that are quantifiable and indicative of good sounding speakers.
The first, flatness of the frequency response within +/-2.5dB or 5dB total variation is probably most audible and if not correct or having excessive peaks, will not sound good. Also sometimes having flat but tilted response higher in bass and slowly falling off by -5dB from 20Hz to 20kHz is not a bad thing.
The second, look for dominant 2nd order and at reference level of say 90dB at listening position, should not be more than -40dB for a speaker (1%) and should be monotonicall descending with higher order harmonics. 3rd order should be st least 12dB lower than 2nd order and so on. However, on a speaker with only 2nd and 3rd order, and nothing higher, they can be within 6dB and sound quite good.
The third is the transient ability. It should reproduce crisp drum rim shots, bongo, and piano strikes or string plucks. This can be measured easily by looking for a right triangle shaped step response function. Transient perfect speakers like Dunlavey SC-IV will do this perfectly. Look at step response of s typical 4th order LR speaker and it looks like a jumbled mess. Until you hear a transient perfect speaker you won’t know what it sounds like. Fortunately, all single driver full range speakers will be transient perfect for the most part. It should have measured group delay variation from 100Hz to 50Hz of less than 7ms, ideally 2ms.
Fourth is flatness of phase of the measured response - especially in the “telephone” band from circa 500Hz to 5kHz (as high as 10kHz). This is where imaging and soundstage come from. Mess up the phase and then imaging is imprecise. Should be within about +/-15deg (ideally +/-5deg) over 500Hz to 5kHz for the speaker.
So there you have 4 measurements that are quantifiable and indicative of good sounding speakers.
Last edited:
Hello,
Audio Precision in the business of measuring audio. They have the test capability to perform most any audio test you can dream up and then some; except the tests that GedLee speaks to. A good place to start is the basic 6 audio measurements.
https://www.ap.com/download/technot...d=AebYSMKKkMHWS-YsbIHpgplY0ZlMbICx1Tmo3vuoKoc
GedLee is off the commercial mainstream path, he is not taking any business away from Audio Precision.
I believe GedLee to be correct about what he says, however he is a bit short in the personality skills to convince most people to change their minds about 1 or 10% THD not being audible.
Thanks DT
Audio Precision in the business of measuring audio. They have the test capability to perform most any audio test you can dream up and then some; except the tests that GedLee speaks to. A good place to start is the basic 6 audio measurements.
https://www.ap.com/download/technot...d=AebYSMKKkMHWS-YsbIHpgplY0ZlMbICx1Tmo3vuoKoc
GedLee is off the commercial mainstream path, he is not taking any business away from Audio Precision.
I believe GedLee to be correct about what he says, however he is a bit short in the personality skills to convince most people to change their minds about 1 or 10% THD not being audible.
Thanks DT
Sure, it could. I don't know if it has been done yet, but it would be interesting. Say you've already got a compression system that takes sliding window snapshots of a wave stream, does an FFT, fiddles with the values using a super-secret recipe, and a playback system that does an IFT to reconstruct the wave.
We can start off by looking at a really old, crude method of compression: changing a 16 bit .wav file to 8 bits. Trouble is, it was kind-of the opposite of what was needed. A loud sine wave (7-8 bits) would still be extremely accurate, to within a small fraction of a dB, but progressively quieter sounds would rapidly lose resolution at a rate of 1 bit per 6 dB.
What about logarithmic weighting, so there is more resolution at lower volumes? Well, if you've got that sine wave as a loud carrier signal, and some smaller, more subtle signals riding on top of it, then the sine wave will modulate the bit depth of the smaller signal. The best accuracy will be at the zero crossing points, while the peaks will strongly distort it.
The break-through was the frequency conversion. Then you could make much better use of logarithmic ratios and lossless compression of "white space". For instance, if the sine wave sticks out by 20dB above the next-loudest tone, you could save it as a special value, delete it from the graph, and scale the remainder. We could stop there, save everything with 8 bits, but get 68dB of dynamic range instead of 48dB. So it would be lossless compression with 11-12 bit accuracy but only taking up 8 bits.
You could make the process recursive, so you save and subtract the loudest frequencies multiple times, until you reach some limit where there's so much noise in the remainder that you can no longer glean anything meaningful from it. Depending on whether you keep the remainder or throw it away, you'd have either lossy or lossless compression (although I wouldn't guarantee that lossless won't take up *more* space than the original
).If you want to prioritise odd harmonics, first, find them on the graph, and make sure the compressor doesn't throw them away.
Isn't there a certain teleological problem there where you need to know the "true" sound in order to compress the harmonics not in the true sound and later expand them?
B.
At the moment it's just 7: series RL + parallel RCL, for a minimalist model of low frequency resonance and rising impedance, and an RC zobel network.
It uses a MOSFET output stage in a single-ended class-A topology, with a current sense resistor, and just a little voltage feedback to stabilise DC.
To reduce THD, I can add more gain stages and use more feedback, or I can tune the phases to match the speaker. Because it's such a sensitive part of the circuit, "future-proofing" it with adjustable knobs is out of the question. That's why I'm starting to think the "next level" is mating speakers with with specially matched amplifiers, and although swapping them out can be done, it's neither easy nor desirable.
Let's say it's a recording of a guitar amp. Excepting a little distortion in the recording process, we can say that it's all part of the true sound.
One thing I've noticed is that it's sometimes possible to hear a "lack of sound". For instance, create a broad spectrum step or impulse. Then subtract a frequency to create a sharp notch in the FR. Then what you 'don't' hear becomes what you do hear! And you can see it on recordings. The frequency that is muted creates ringing, which is both audible and visible. One could argue that strictly speaking it's interference between adjacent frequency bands etc, etc, but the point is that you can hear it.
So with that in mind, indeed, it might be more important to keep low-level odd harmonics in a recording, because 'errors' caused by deleting them could be just as audible as the harmonics themselves.
To paraphrase Lord Kelvin, if you can't quantify a feature, you can't talk intelligently about it. Or looking at it from a DIY perspective, just how do you use REW to fine-tune your system without some metric to guide you.
B.
I think the thd has little to do with how fatiguing an amp is.
The main listening fatigue factor, if you have ears, is the over-damping/or under, high feedback and varying phase shifts.
You can listen to s.e.t. with 10% thd and high order harmonics without any fatigue, however some SS amps are fatiguing because they have that big waterfall pattern of harmonics up to 17H order, you can actually hear that in women voices and trumpets.
Standing waves are exaggerated in the cone by over-damping which is very annoying. High feedback Bass is only a dry rumble with 0 definition, decay, tone, all this is missing, if you listen to real instruments of course, if you listen to electric guitar or already amplified sound with PA then it doesn't really matter.
Measuring THD is a quality factor in building/designing an amplifier and it has almost nothing to do with how good it will actually sound.
IMD into a reactive/capacitive load with multiple signals is a sound quality factor. Even rise of harmonics is a sound quality factor.
The main listening fatigue factor, if you have ears, is the over-damping/or under, high feedback and varying phase shifts.
You can listen to s.e.t. with 10% thd and high order harmonics without any fatigue, however some SS amps are fatiguing because they have that big waterfall pattern of harmonics up to 17H order, you can actually hear that in women voices and trumpets.
Standing waves are exaggerated in the cone by over-damping which is very annoying. High feedback Bass is only a dry rumble with 0 definition, decay, tone, all this is missing, if you listen to real instruments of course, if you listen to electric guitar or already amplified sound with PA then it doesn't really matter.
Measuring THD is a quality factor in building/designing an amplifier and it has almost nothing to do with how good it will actually sound.
IMD into a reactive/capacitive load with multiple signals is a sound quality factor. Even rise of harmonics is a sound quality factor.
Last edited:
Depends on what you mean by "new". The idea of using a weighted sum of the harmonics in order to get a quantitative measure for "sound quality" (whatever that might be) has been around for a long time. The GedLee metric is just one of those weighting methods. The point seems to be that there are not many software packages out there that will output the GedLee metric (or other weighted sum) straight away, and engineers/DIYers/software users are too lazy to work out the GedLee metric (or other weighted sums) on their own.
Anyone can apply whatever metric they like. Different metrics will have different merits. One needs to understand how a specific metric works, and what it means.
I am not sure I get this right. Are you saying that the Ged Lee approach is just an effort by Earl Geddes, and noone else cares? If that's what you're saying, I'd like to disagree (a lot)! Someone already mentioned that Earl Geddes might have a "strong personality" (he seemed like a pretty normal guy when I communicated with him a while ago). Even if that were true, one would still need to distinguish between his scientific / engineering ideas and his personality.
Last edited:
Dolby DBX ��
Some of these discussions are a little, well....blindfold.
I don't doubt the Wrong distortion can be heard.
I'm far more sceptical about being able to hear the 17th Harm of a female note centred at 500Hz, and at a level of -40dB.
THD measurement and audibility is one of those subjects that will never be resolved here with logic.
Remember that thread started a few months back?
Can you hear the 3rd harmonic of 5kHz?
Or the 7th harmonic of 2 kHz?
Or the 4th harmonic of 4kHz?
Perhaps if the distortion is gross (> -30dB absolute), otherwise, are you sure it's not another distortion? IMD for eg
The whole premise of audibility of THD at higher freqs is flawed, as is the measurement and specification of THD at 10kHz -meaningless when the 'worst offending' harmonics are beyond the limits of human hearing.
Most my opinion is the same WRT phase distortion/So called time alignment.
The few times I know I have heard IMD it has been awful, just like stepping on two bass pedals a semi tone apart, although maybe the beating freq makes a big difference here?
(
I can't hear the one cycle phase wrap of a 4th order system crossing at 1kHz.
But the group delay of a 6th BP (maybe 30ms or more) is objectionable in the extreme.
(My hearing extends to 17kHz, with a scooped BBC dip in one ear, of -3 to -5dB, according to my OH hearing tests, and I'm 40.
Most my age can't even hear that well.)
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.