Sennheiser HD650 Review (Headphone)

[amirm]

First track: Time: 2:54 to 3:06. Mid sharpness on the vocalist for the HD650.
Second track: Time: 3:40 to 4:00. Mid sharpness on the vocalist for the HD650.
Third tack: Time: the whole track. Lower resolution, half of the detail, worse bass, uneven balance of the tracks (in comparison to Focal Clear, otherwise it's fine headphone)

Finally getting to this . I cannot find a Tidal version of the first two to try with my EQ. I did try the third and EQ absolutely improves the HD650.

I did compare it to Focal Clear (with my EQ) and I agree it sounds better than HD650. It has a more open and clean sound although you have to mindful of the fact that it is more efficient so plays louder. You need to turn down the volume to make a fair comparison.

 

[Jimbob54]

Finally getting to this . I cannot find a Tidal version of the first two to try with my EQ. I did try the third and EQ absolutely improves the HD650.

I did compare it to Focal Clear (with my EQ) and I agree it sounds better than HD650. It has a more open and clean sound although you have to mindful of the fact that it is more efficient so plays louder. You need to turn down the volume to make a fair comparison.

An important point for readers of all headphone reviews (and measurements ) here and elsewhere. A recommendation for a $300 unit vs no recommendation for a $1500 unit does not necessarily mean if you applied EQ and listened to both the $1500 unit wouldnt be preferable to the $300. The recommendation/ panther includes relative value, build, comfort AND how they sound. Is the Clear worth $1000 more than the 650? Probably not but it most certainly can sound better . Until the driver pops

 

[Francis Vaughan]

Just to quickly reference your last sentence about averaging making things worse,

I didn't mean every possible use of averaging will make things worse. But there is a really important point. Not everything can be validly averaged. For a start, averaging assumes something about the probability distribution. Not all distributions have a useful average. Multiple measurements where a point is moving about will be telling you something important about the nature of the underlying physical basis of the peak. Just taking an average may mean you lose sight of something important. Why does it move? Just saying you don't know and taking an average is not always the best answer. You might be averaging away from an optimum since the best case is being equally weighted with all the poor options. It is invalid to assume the optimum is always the middle of the samples.
There is of course averaging inherent in what we are using. The HRTF in the Harman curve is an average. Average head dimensions. Average pinna. Although the word "average" is not well defined here . Some of that could be improved on on an individual basis. It is important to have some underpinning idea of what the physics of the meachanism being corrected is. Without that you are fighting with an arm tied behind your back, and possibly making invalid assumptions about what the numbers are telling you.

 

[Robbo99999]

??? Did he provide the no averaging EQ so you could determine that?

I see your point, in as much as you're saying I haven't compared an EQ based on his average measurement vs say his optimum position measurement (the latter being what you do). So, yes, I can't answer that on a direct level for comparison....but I would imagine that there is some variance to be seen even between two optimal mounting positions (say if you got an optimal mounting position and then decided to remove the headphone & then reposition it optimally again), so an average could be more representative even in that situation. This response can also be aimed at your post below too. I'm still not really sold on the one measurement approach as opposed to the average of measurements approach....even though I understand your approach.
EDIT: it would be pretty cool if I could compare EQ based on Oratory's average measurement curve vs your single optimal position measurement on the same headphone.....I'd actually do that if you ever measure a headphone I own (AKG K702, NAD HP50, Senn HD600, Hifiman HE4XX).

Not so here. Let's assume I know how to mount the headphone to get the most optimal response. I then random take the headphone off and put it on haphazardly as I think Tyll did. Now averaging these will get you worse results than the first one that was optimally done.

The variance here is not purely random so filtering could very well take away useful precision.

I didn't mean every possible use of averaging will make things worse. But there is a really important point. Not everything can be validly averaged. For a start, averaging assumes something about the probability distribution. Not all distributions have a useful average. Multiple measurements where a point is moving about will be telling you something important about the nature of the underlying physical basis of the peak. Just taking an average may mean you lose sight of something important. Why does it move? Just saying you don't know and taking an average is not always the best answer. You might be averaging away from an optimum since the best case is being equally weighted with all the poor options. It is invalid to assume the optimum is always the middle of the samples.
There is of course averaging inherent in what we are using. The HRTF in the Harman curve is an average. Average head dimensions. Average pinna. Although the word "average" is not well defined here . Some of that could be improved on on an individual basis. It is important to have some underpinning idea of what the physics of the meachanism being corrected is. Without that you are fighting with an arm tied behind your back, and possibly making invalid assumptions about what the numbers are telling you.

@Francis Vaughan Given what you've said, how would you go about applying your approach to the following HD600 measurements from Oratory?

 

[Francis Vaughan]

how would you go about applying your approach to the following HD600 measurements

To be honest, I have no clue what to make of that one graph. I can't even see how many measurements have been made, and there is no information as to what the measurements are. What conditions changed and which line corresponds to what.
However, in general I would be looking to try to identify individual physical mechanisms that were apparent in moving the response from the optimal. And do this for each measurement individually. Look to see which were common between measurements and decide on possible compensation parameters. If averaging is needed I would be looking to average the parameter settings for each given compensation, not averaging the measurements before calculating compensation. I might want to pick the modal value (for instance) for compensation, not the average. Those variations between measurements carry important information. Just averaging them away is not a great idea.

 

[pkane]

To be honest, I have no clue what to make of that one graph. I can't even see how many measurements have been made, and there is no information as to what the measurements are. What conditions changed and which line corresponds to what.

Here’s another one. I made that one with in-ear mics and HD650. The only thing changed between measurements was that I took off the headphones and then repositioned them for best fit and comfort — in other words every one of those measurements was from a likely position of HD650 on my head:

 

[solderdude]

at least you got a consistent seal.
You can safely EQ 20Hz to 3kHz.

Mine in the same scale (2dB/div) HD650, HD600, HD660S

 

[Francis Vaughan]

in other words every one of those measurements was from a likely position of HD650 on my head

Yup. This is exactly where blindly averaging is invalid. What does it mean to calculate the "average" position on your head? Even with these measurements we can see that the blue trace is an outlier. I would not be including it in the correction for the dip at 2kHz. There seems to be a bimodal effect above 5kHz. You can't average that. Each trace tells a story. Maybe some bits can be usefully averaged, but until you do further analysis you can't know if it can.

 

[pkane]

Yup. This is exactly where blindly averaging is invalid. What does it mean to calculate the "average" position on your head? Even with these measurements we can see that the blue trace is an outlier. I would not be including it in the correction for the dip at 2kHz. There seems to be a bimodal effect above 5kHz. You can't average that. Each trace tells a story. Maybe some bits can be usefully averaged, but until you do further analysis you can't know if it can.

But that’s not a reason to pick just one position and then correct for it, is it? With this much variation, it’s best not to touch the variable part of the curve with any EQ.

 

[pkane]

at least you got a consistent seal.
You can safely EQ 20Hz to 3kHz.

Mine in the same scale (2dB/div) HD650, HD600, HD660S

Yeah, I made a serious attempt at each measurement to position the phones at what felt like was the best position I could find, similar to what I’d do when using them normally. The amount of variation surprised me.

 

[solderdude]

You can hear the amount of treble varying when moving the HD650 (small drivers) around on the head.
I do see some variation but as you are using a pinna the variations will be bigger.

 

[Francis Vaughan]

But that’s not a reason to pick just one position and then correct for it, is it? With this much variation, it’s best not to touch the variable part of the curve with any EQ.

Absolutely. That may well be the best call of all. Again why averaging is not always the right answer.

 

[Robbo99999]

To be honest, I have no clue what to make of that one graph. I can't even see how many measurements have been made, and there is no information as to what the measurements are. What conditions changed and which line corresponds to what.
However, in general I would be looking to try to identify individual physical mechanisms that were apparent in moving the response from the optimal. And do this for each measurement individually. Look to see which were common between measurements and decide on possible compensation parameters. If averaging is needed I would be looking to average the parameter settings for each given compensation, not averaging the measurements before calculating compensation. I might want to pick the modal value (for instance) for compensation, not the average. Those variations between measurements carry important information. Just averaging them away is not a great idea.

I think I'd have the same approach, I'd want to see how the frequency response changed with headphone position and then make decisions on what to average from seeing that. Yes, it's hard to comment when you don't have the machine in front of you & are doing the measurements, seeing the effects.

 

[Robbo99999]

Here’s another one. I made that one with in-ear mics and HD650. The only thing changed between measurements was that I took off the headphones and then repositioned them for best fit and comfort — in other words every one of those measurements was from a likely position of HD650 on my head:

View attachment 103098

That's pretty cool, and I think a very good representation on what can happen in the real world in terms of always placing a headphone for optimal position....you'd think they'd be some correlation with that too when trying to place optimally on a simulated head too (ie the GRAS gear of Oratory or Amir).

EDIT: it's pretty interesting that the bottom two lines don't follow the pattern of the other 3 lines in the say 4.5kHz+ area. If this was a GRAS measurement I'd probably do some more measurements to see if the bottom 2 lines were outliers, and if they were then I'd probably discount them and average the rest of the lines.....because it's those bottom 2 lines that are really throwing out the variance.....there's a big gap there between them & the others in the +4.5kHz area with no measurements lying inbetween.

 

[pkane]

That's pretty cool, and I think a very good representation on what can happen in the real world in terms of always placing a headphone for optimal position....you'd think they'd be some correlation with that too when trying to place optimally on a simulated head too (ie the GRAS gear of Oratory or Amir).

EDIT: it's pretty interesting that the bottom two lines don't follow the pattern of the other 3 lines in the say 4.5kHz+ area. If this was a GRAS measurement I'd probably do some more measurements to see if the bottom 2 lines were outliers, and if they were then I'd probably discount them and average the rest of the lines.....because it's those bottom 2 lines that are really throwing out the variance.....there's a big gap there between them & the others in the +4.5kHz area with no measurements lying inbetween.

That was just five measurements. I assume there will be more falling in between as I do more tests.

 

[Robbo99999]

That was just five measurements. I assume there will be more falling in between as I do more tests.

If those were GRAS measurements, and you had more falling inbetween, then I'd do an average line in that area (rather than discounting the bottom two measurements) to represent a general trend and then for EQ just use only very broad filters or shelf filters above 4.5kHz with the exception that if they all exhibited a peak at 7kHz as your current measurements do then I'd consider using a sharper filter on that point along with the broader filters and/or high shelf filters in that 4.5kHz+ area.

 

[Feljor]

por favor revise ThieAudio Monarch, Moondrop S8 y MoonDrop Blessing 2

 

[amirm]

por favor revise ThieAudio Monarch, Moondrop S8 y MoonDrop Blessing 2

I have a couple of Moondrop IEMs which I will test in the future.

 

[pkane]

If those were GRAS measurements, and you had more falling inbetween, then I'd do an average line in that area (rather than discounting the bottom two measurements) to represent a general trend and then for EQ just use only very broad filters or shelf filters above 4.5kHz with the exception that if they all exhibited a peak at 7kHz as your current measurements do then I'd consider using a sharper filter on that point along with the broader filters and/or high shelf filters in that 4.5kHz+ area.

Right. I'm just concerned that variation due to position has more to do with the actual function of the ear than with any deficiencies in reproduction. I assume that the ear physiology creates these peaks/valleys including variations in frequency response to help with detecting the direction of sound. If we spend too much time trying to "correct" these variations to some specific, static target, we may just be destroying or distorting information that the ear expects to receive to perceive the sound as being more natural. The area where the measurements change drastically with tiny position changes are perhaps better not corrected at all.

 

[Robbo99999]

Right. I'm just concerned that variation due to position has more to do with the actual function of the ear than with any deficiencies in reproduction. I assume that the ear physiology creates these peaks/valleys including variations in frequency response to help with detecting the direction of sound. If we spend too much time trying to "correct" these variations to some specific, static target, we may just be destroying or distorting information that the ear expects to receive to perceive the sound as being more natural. The area where the measurements change drastically with tiny position changes are perhaps better not corrected at all.

Well, we got the Headphone Harman Curve to rely on as a Target Curve....which is kinda all we got when it comes to measurments & targets. That doesn't show a peak at 7kHz, therefore if you're EQ'ing to the target curve then you'd remove that peak if it was consistent across all your measurements. I can't really speculate on the finer points you make. More specifically re your measurements, I'm not sure you can really use them to EQ your headphone, because it's not like you'd overlay the Headphone Harman Curve on your measurements and then EQ to that, because the Headphone Harman Target Curve is only specific to that particular GRAS unit and not your own head & HRTF....so that's not valid. If you've got in ear microphones (which you do), then you could have a stab at creating your own Target Curve......I've loosely thought about doing that in my own room with my flat anechoic EQ'd speakers which are at equidistant triangle from my listening position....do a REW sweep & measure from those in ear microphones, that would be a simulation target of your speakers in your room.....don't know too much about the specifics of making this happen as not researched it due to not really needing to as Harman Headphone Target sounds accurate to me. Could be an interesting project, sure people have done it before. You'd then measure your headphones on your head like you have done, and then EQ to your determined Target Curve. I'm certain there's lots of practical pitfall details associated with trying to do this!