[bigshot]

  The "transients" I am referring to are sudden impacts that result in very rapid increase (or decrease) in sound level... the strike of a drum, the crash of a cymbal, the initial bite of the bow on the string.  These are more amplitude transients, though indeed they are the accumulation of waveform transients across frequencies, aggregated into an instantaneous power level.

 
Not exactly instantaneous... If you look at the waveforms of even the sharpest drum hit, they take place over many many samples. When you talk about transient response as it relates to phase shift, you're talking about an error that is several orders of magnitude below the transient of any quick impulse that occurs in music. The big problem I see in the way time errors are discussed in audiophile circles is that no one bothers to compare the scale of the time slivers they are talking about. The discuss absurd concepts like PRaT as being something that affects the timing of the music, or claim that higher bitrates allow them to hear the ring out of reverb longer... or that transient response that affects square wave rendering somehow is in the same ballpark as the transient response of a drum hit. Scale matters a lot.
 
Originally, I posted a figure up there for the lowest JDT (just detectable threshold) for phase shift being 1 ms at 500kHz. My math isn't great, but with some struggle, I can wrap my head around how big that is. I can guesstimate that the amount of shift using rebooks sampling as a guide would have to be spread over 40 samples to become audible. That is pretty huge. And if you look at the listening test results in the second link I posted, even with headphones, even 8 ms wasn't clearly identifiable in any of the musical samples, and that would mean that it would cover 320 samples. Are there any headphones with phase problems that large?
 
Again, this isn't my area of expertise, so feel free to correct me if I'm missing something.

 

[Speedskater]

  Blind tests are most useful in determining whether a difference exists between two very similar sounds. Headphones rarely sound very similar. It seems to me that careful sighted listening with clear criteria for judging would do fine for headphones.

Seldom do I have a different view-point than bigshot, but this is one time that I do.
 
There are different types of blind tests.  On example is an ABX test to determine if there are JND/JDT differences. Another type of blind test would determine preference when the differences are larger.  Loudspeakers, headphones, mics & pre-amps and phono cartridges & pre-amps often have large enough differences to skip the ABX tests and move on to preference tests. But those components often come with cognitive bias baggage, so any testing must be blind.

 

[RRod]

 
Seldom do I have a different view-point than bigshot, but this is one time that I do.
 
There are different types of blind tests.  On example is an ABX test to determine if there are JND/JDT differences. Another type of blind test would determine preference when the differences are larger.  Loudspeakers, headphones, mics & pre-amps and phono cartridges & pre-amps often have large enough differences to skip the ABX tests and move on to preference tests. But those components often come with cognitive bias baggage, so any testing must be blind.

 
Except how do you blind test headphones; I've not met a pair across models that had identical feel on the head and ears.

 

[ruthieandjohn]

   
Not exactly instantaneous... If you look at the waveforms of even the sharpest drum hit, they take place over many many samples. When you talk about transient response as it relates to phase shift, you're talking about an error that is several orders of magnitude below the transient of any quick impulse that occurs in music. The big problem I see in the way time errors are discussed in audiophile circles is that no one bothers to compare the scale of the time slivers they are talking about. The discuss absurd concepts like PRaT as being something that affects the timing of the music, or claim that higher bitrates allow them to hear the ring out of reverb longer... or that transient response that affects square wave rendering somehow is in the same ballpark as the transient response of a drum hit. Scale matters a lot.
 
Originally, I posted a figure up there for the lowest JDT (just detectable threshold) for phase shift being 1 ms at 500kHz. My math isn't great, but with some struggle, I can wrap my head around how big that is. I can guesstimate that the amount of shift using rebooks sampling as a guide would have to be spread over 40 samples to become audible. That is pretty huge. And if you look at the listening test results in the second link I posted, even with headphones, even 8 ms wasn't clearly identifiable in any of the musical samples, and that would mean that it would cover 320 samples. Are there any headphones with phase problems that large?
 
Again, this isn't my area of expertise, so feel free to correct me if I'm missing something.

May not be your area of expertise, but you are sure educating me... thanks!
 
It is interesting to understand about the fact that most of these effects are taking effect in the sub-millisecond time regime, so would be hardly perceptible.  Also illuminating to realize that even what seems to be a large impact like a drum strike really permeates through several 10s of samples at CD sampling rates, so phase misalignment effects would be minor.
 
It would be fun, and possible with a tool such as Audacity (and maybe Matlab) to take a snippet of impact sound, Fourier transform it, rearrange the alignment of the phase portion in some frequency-dependent manner in line with measurements made on some systems (perhaps a total of 3 to 5 complete phase cycles over the audio frequency range), then put the signal back together with an inverse Fourier transform and see how the impact sound has changed, if at all.
 
(For control, with all this stuff, also better FT and IFT the signal WITHOUT messing with the phase to be sure it really goes back together OK).
 
Project for when I have lots of time!

 

[RRod]

It would be cleaner to make the phase adjustments for an impulse response; then you can convolve the waveform from the ifft with whatever files you want to apply the effect of the phase changes. Got any preference for a change you'd want to try to hear? I should be able to crack something out for you without much hassle. Just tell me how you'd want any of phase shift, phase delay, or group delay to be a function of frequency.

 

[bigshot]

  It is interesting to understand about the fact that most of these effects are taking effect in the sub-millisecond time regime, so would be hardly perceptible.  Also illuminating to realize that even what seems to be a large impact like a drum strike really permeates through several 10s of samples at CD sampling rates, so phase misalignment effects would be minor.

 
It really helps to compare specs to perception. You find out how forgiving the human ear actually is. But absolute phase is a bit trickier. There are certain types of non-symmetrical waveforms that can be distinguished from inverted phase by some people. But I seriously doubt that it would be a sound quality issue. More of a perceptual one.

 

[ruthieandjohn]

  It would be cleaner to make the phase adjustments for an impulse response; then you can convolve the waveform from the ifft with whatever files you want to apply the effect of the phase changes. Got any preference for a change you'd want to try to hear? I should be able to crack something out for you without much hassle. Just tell me how you'd want any of phase shift, phase delay, or group delay to be a function of frequency.

Wow thanks.
 
I saw a plot of system phase error vs frequency that went about four times through 360 of phase error, linearly with frequency from say 20 Hz to 20 kHz.  I'd suggest that. (probably logarithmic with frequency, perhaps 180 degrees per octave?  That would be 5 cycles over the 10 octaves from 20 Hz to 20 kHz.
 
Choose a signal that is an impact from say an orchestra, like a sudden loud entrance (one example is in the first part of Hadyn's Creation, where the chorus comes in "And Their Was LIGHT!" and the orchestra goes crazy.
 
Another possibility is the start of the fourth movement of Saint Saens Organ Symphony, which begins with a full organ and orchestra blast... I can send you a link to a drop box address where I have already posted that. (in fact I will PM it to you).
 
Compare the with and without phase error versions.

 

[RRod]

 
Wow thanks.
 
I saw a plot of system phase error vs frequency that went about four times through 360 of phase error, linearly with frequency from say 20 Hz to 20 kHz.  I'd suggest that. (probably logarithmic with frequency, perhaps 180 degrees per octave?  That would be 5 cycles over the 10 octaves from 20 Hz to 20 kHz.
 
Choose a signal that is an impact from say an orchestra, like a sudden loud entrance (one example is in the first part of Hadyn's Creation, where the chorus comes in "And Their Was LIGHT!" and the orchestra goes crazy.
 
Another possibility is the start of the fourth movement of Saint Saens Organ Symphony, which begins with a full organ and orchestra blast... I can send you a link to a drop box address where I have already posted that. (in fact I will PM it to you).
 
Compare the with and without phase error versions.

 
OK, I'll give it a shot. I can use my recordings of the Organ Symphony and Creation as a start.

 

[castleofargh]

my main problem with those phase talks or jitter talks, transient, PRAT... is the actual lag between what people seem to be able to notice in tests, and what people claim they can hear(while mostly doing no test at all). and sadly when you read about it on headfi it's often from the second group. so I tend to develop a kind of distaste for those subjects ^_^.
 
now I'm spending most of my time on headfi saying how the headphone is the problem and how bad they are at doing signal fidelity. so I won't claim that all is perfect and that we shouldn't concern ourselves about it.  it's not hard to look at a cascade graph of any headphone to understand that headphone perfection isn't for tomorrow.
when I see a frequency ringing pretty loud for 2ms, the school boy in me gets scared thinking "ok I can hear 16000hz(old schoolboy ^_^), so my ears are still ok with changes happening in 0.0625ms omagad omagad omagad!!!!!!!! that headphone is so crappy.
but in practice, the best I can get is to find that this headphone sounds annoying at a given frequency. I feel like it sounds bad but I have no idea if it's because that frequency is spiking? if the distortion levels are crazy high? or if the signal just went on for 6ms with almost no attenuation.
so I need mad values to just notice something is off, and even then I'm not so sure what it is. that makes me think that overall transient and damping are important for signal fidelity, but need to be real bad for me to notice them(without real hope to actually identify the problem by myself).
 
now when talking about an amp and it not being fast enough or whatever BS, the delays we're talking about are a joke compared to the movement of the headphone's drivers, to me it's pretty obvious that a great electrical or mechanical (or even better, both)damping will do a lot better for the sound than any amp with faster transient. once again the overall distortion values are so much higher into headphones, how can we pretend not to hear it but then claim to hear stuff much smaller from amps? in my brain that doesn't compute.
and by reverse engineering science like placebophiles love to do so much, I noticed that those headphones I find so good, are the ones that show a very fast and uniform attenuation in CSD graphs(electro ortho planar whatever stuff damped like mad). so somehow I feel like I could go saying that my favorite headphone is the one with the FR I like and the best damping. but transients stuff... I don't really notice those guys TBH.
do I know that a sound is a 5khz tone if I only hear 1/10th of a period? 1/50th of it? or do I need 10periods? I honestly don't know, but I would suspect that my brain needs a certain amount of data through time before it decides to interpret it as being meaningful. so I make the hypothesis (because I really don't have a clue), that my brain doesn't mind imperfect beginning as long as what follows is stable long enough to interpret the sound. and when it goes on too long, it starts masking the next sound making us feel like we're losing in detail and speed? something like that. so I care about the post stuff, not the pre.

 

[ruthieandjohn]

   
 
now when talking about an amp and it not being fast enough or whatever BS, the delays we're talking about are a joke compared to the movement of the headphone's drivers, to me it's pretty obvious that a great electrical or mechanical (or even better, both)damping will do a lot better for the sound than any amp with faster transient. once again the overall distortion values are so much higher into headphones, how can we pretend not to hear it but then claim to hear stuff much smaller from amps? in my brain that doesn't compute.
 

Am I correct in understanding that "amp delay" of transients is tiny compared to headphone delay, and so we do NOT need all that extra power (beyond the level needed to drive the headphone to say 115 dB SPL) for amp handling transients?
 
I've been trying to quantify the statement that others have made when they state that you need gobs more amp power than the level needed to achieve a desired listening level (for a given headphone efficiency).  The statement, as I understand it, is that you need "some amount" of extra power to handle transients.
 
But... how much extra??

 

[castleofargh]

 

   
 
now when talking about an amp and it not being fast enough or whatever BS, the delays we're talking about are a joke compared to the movement of the headphone's drivers, to me it's pretty obvious that a great electrical or mechanical (or even better, both)damping will do a lot better for the sound than any amp with faster transient. once again the overall distortion values are so much higher into headphones, how can we pretend not to hear it but then claim to hear stuff much smaller from amps? in my brain that doesn't compute.
 

Am I correct in understanding that "amp delay" of transients is tiny compared to headphone delay, and so we do NOT need all that extra power (beyond the level needed to drive the headphone to say 115 dB SPL) for amp handling transients?
 
I've been trying to quantify the statement that others have made when they state that you need gobs more amp power than the level needed to achieve a desired listening level (for a given headphone efficiency).  The statement, as I understand it, is that you need "some amount" of extra power to handle transients.
 
But... how much extra??

transient for me is how fast the signal can go from one value to another, not so much how high it needs to go. so while there is a logical relation, I'm not sure driving power does much to transient.
 
 to play a sound at 100db, a given headphone will need a known amount of current and voltage. you can't expect the headphone to "take in" more current than what it can let pass through. and if it were to "take in" more volts, then it would just be louder ^_^. so all we need is to estimate our own needs for loudness correctly, for the loudest signal on the record reaching 0db in the digital domain, to still be within the amp's ability.  it's a game of estimating right.
 sometimes we need more than the exact amount we estimate from the headphone specs, because specs are only for 1khz and things can be different at other frequencies, but the result is still the same. a source provides all the headphone can take, and the headphone is well driven. or it doesn't and the headphone isn't properly driven. there is no condition where the source will actually provide more. that doesn't exist.
 
to better improve the electrical damping and in the process the control of the driver, the impedance of the amp is also an important factor IMO. after that, I really can't see what more max power never used could have to offer.
 
I also go with the usual 115db target when I try to see if I can pair a headphone and an amp, because it's a very safe value to look for while not being a ludicrously high one. but I don't need that much myself and listen to music very very quietly. so my actual needs are lower than those of most people and if I were to end up with an amp able to only do 105 or 110db into a given headphone, I would most likely still buy it for my own use.

 

[bigshot]

  I've been trying to quantify the statement that others have made when they state that you need gobs more amp power than the level needed to achieve a desired listening level (for a given headphone efficiency).  The statement, as I understand it, is that you need "some amount" of extra power to handle transients. But... how much extra??

 
I'm not sure about tube amps, but I know for solid state amps, they perform pretty much perfectly all the way up to the point of clipping. Anything above and beyond power wise isn't used.