Oh, that depends. Of course on if that's been asked by someone with a folding ruler standing besides them or someone on FB, snapchat, Tinder or Grindr. 😉
Doctors office records are often taken as some of the best and most reliable evidence we have for certain measured physical characteristics. Most of the good doctors measure systematically, and or train their nurses, and or certified medical assistants, to do it.
Regarding some people being able to hear the quantizing distortion on undithered CD, it is considered noncontroversial. IIRC, Scott Wurcer mentioned a couple of papers on it some years ago.
Regarding some people being able to hear the quantizing distortion on undithered CD, it is considered noncontroversial. IIRC, Scott Wurcer mentioned a couple of papers on it some years ago.
A lot depends on the listening conditions. In the conditions in which I played with THD, I could not hear a difference of up to 5%.
I suppose, in very good conditions, you can hear 1%.
We are talking about this type of distortion.
.
Once I passed the online AB test, there I could find the difference in THD 0.02% but it was very difficult, I can say that I just guessed these 0.02%, in fact I could easily hear 0.1% and already with difficulty 0.05%. But I do not know what kind of distortion was used there.
Well, that might be part of the test. 😀 Though, I saw measurements of someone in a forum and he was bashing about how awful the 'series consistancy' of the drivers he'd ordered lately are. I asked him about the mic and then he said it was DIY with a Panasonic electret capsule. I looked it up and it was a 45ct one that was rated up to 14k. Uh, suuure, the tweeters are all trash...
I don't see how that applies here. I was referring to your comment earlier that some people can hear the difference between CDs recorded with dither and CDs recorded without dither:
That's a yes/no answer to the question, "can you hear if this CD is recorded with dither?", right? If the answer is a YES/NO answer many statistical analysis tools become unusable, including those that assume normality as they often require continuously distributed data rather than categories.
Tom
For all of you audiophiles. Go listen to Bohemian Rapsody by Queen and tell us where and when you hear harmonic and intermodulation distortion and what effect it had in the production you hear. The answer will be interesting
It’ s nonsense in theory world, of course.
Bot somewhere after 2k x3k you get to 5k + the 4k and 6k komponets thats kind of obvious as it is 2. harmonics.
But it all skows that as long harmonics<<fundemental for one fundemental it vill stay that way for natural harmonic signals also.
Its god of course.
But remember what I was told when I first met pro audio world
You can only judge equipment with recordings of acoustic instruments. You have no way of knowing what studio equipment should sound like
Before going to great lengths of achieving 0.00000 .... 1 % distortion try out these tests og different lossless vs. non lossless formats. Use the best headphones you can get.
I know it it not directly THD test, but I think it will show you how difficult it is to hear the difference, and I think it will be exactly the same with distortion.
https://abx.digitalfeed.net/list.html
What really matters is in room frequency response; frequency response of the speakers, directionality and the room treatment. Spend the money and time there.
(of course also matters much that the amp/speakers are capable of high dynamics .... but start with frequency)
My 2 cent 😉
I know it it not directly THD test, but I think it will show you how difficult it is to hear the difference, and I think it will be exactly the same with distortion.
https://abx.digitalfeed.net/list.html
What really matters is in room frequency response; frequency response of the speakers, directionality and the room treatment. Spend the money and time there.
(of course also matters much that the amp/speakers are capable of high dynamics .... but start with frequency)
My 2 cent 😉
The way I understand it, the mechanism that produces harmonic distortion is actually an IMD mechanism. The side bands just happen to be harmonically related. If you send H1 to the non-inverting input, and H2 to an inverting input, it's IMD that creates H3 etc. So in a processing sense it is recursive.
The loop speed of a 1MHz amplifier may be fast, but it's surprisingly tricky to look at say, the leading edge of a square wave test signal, looking at settling time, etc, and see the ripples with the resolution of a logarithmic scale at the same time!
But there are 2 different frequencies in my example: H1 and H2. How could they not undergo IMD?
You must consider the second intercept for harmonic distortion and third intercept for IMD. IMD rises much faster with power out than HD hence third intercept point is higher. Just for short, the intercept points are calculated when the harmonic or intermodulation product is equal to the fundamentals and they theoretically intercept with the fundamental. These phenomena are very well described in radio frequency design manuals.
In RF these are very important factors because you cannot allow two frequencies "mixing" in the front end creating new frequencies and contaminate everything else. This is exactly what jammers do. The Russians are masters at these techniques and place their jamming frequencies near enough the receiver frequency to garble the received signal to the extent that it is useless. You can just look at a scope and see the resulting amplitude (or frequency or phase) modulation placing two frequencies close to each other. This is what we try to avoid in audio as well, we don't want two signals mixing due to non-linearity and generate a third beat frequency that obscures to original. You can however have no distortion at all by making the supply voltage infinite in your simulator because the circuit won't blow up your computer and the result will be zero distortion of any kind because you are stretching the DUT linearity to infinity.
In amplification there are two non-linearities, one close to zero and the other close to supply, there is no, or should be no squiggles and wiggles between.
In amplification there are two non-linearities, one close to zero and the other close to supply, there is no, or should be no squiggles and wiggles between.
Last edited:
Its not a question, that some people can hear if a CD is undithered, is a statement which is already known to be true. However, if we wanted to do the research ourselves, if we found one person who could hear quantizing distortion on an undithered CD then what we would want to know is, whether that's a big problem or not. If its only one person in the world, then nothing to worry about. So the question we would then need to ask, what percentage of the population can hear it? For that we might use ABX and the usual statistical methods. And that's probably about what happened in the case of the originally published work.
All the above should be obvious.
What I don't understand is where you got the idea I wanted to apply statistics to a yes/no question related to the original statement? Reason I'm asking is I don't think I said that.
Last edited:
Maybe not everyone is used to look at IMD measurements with mid freq. At least not me. So often the f1 + f2 component are forgotten as it is out of band
Inspired by the 3 tone test i now did a 2 tone test with 1% 2. harmonic distortion,
I used 2 k and 3k tone. So not harmonically related.
Then i can write the whole expression for the distortion component of the signal
2k is a and 3k is b
(a+b)^2 = a^2 + 2ab+ b^2. I guess this is pretty well known formula
So then after substituting a with sin("2k") and b with sin("3k") we get the:
sin("2*2k") + sin("2*3k" + 2*sin("3k-2k") + 2*sin("3k+2k") components, excuse the non correctnes.
And loking at the phase of the distortion components you see I have made some shortcuts and not included phase change
'
Inspired by the 3 tone test i now did a 2 tone test with 1% 2. harmonic distortion,
I used 2 k and 3k tone. So not harmonically related.
Then i can write the whole expression for the distortion component of the signal
2k is a and 3k is b
(a+b)^2 = a^2 + 2ab+ b^2. I guess this is pretty well known formula
So then after substituting a with sin("2k") and b with sin("3k") we get the:
sin("2*2k") + sin("2*3k" + 2*sin("3k-2k") + 2*sin("3k+2k") components, excuse the non correctnes.
And loking at the phase of the distortion components you see I have made some shortcuts and not included phase change
'
I agree that you never said anything about how the experiment was conducted. See, this is where a reference to the original paper - or a paper looking at this question - would be handy.
If one wanted to determine whether CD dither is audible, it seems to me that the participant in the study would have to be presented with two stimuli: a) with dither and b) without dither. Then a question along the lines of, "which of these sound samples do you prefer or is there no difference?" That leaves three options: 1) A is better; 2) B is better; 3) there's no difference. So you will have categorical data to analyze.
All statistical analysis methods I've heard from you so far, including here, involves normality. Here for example:
That made me think that you were applying a statistical analysis that assumed normality in the analysis of the categorical data or figured that such an analysis would be the appropriate tool in this case.
Tom
I did say bell curve (as an assumptive model), but not necessarily a normal one. For reference: https://www.edge.org/response-detail/11715#:~:text=The Gaussian or normal bell,thus the most impulsive noise.