[GoldenSound]

As many of you already know, there are Convolution filters by Mitch from Accurate Sound for SR-1 and CA-1 headphones.

You could have a perfectly flat (within reason) response of those two headphones, but how close that flat FR has brought them together, it's up to each listener to decide, but it didn't make them sound the same.
Also, neither the other 20 different headphones that Mitch made the filters for, sound the same after the filter is applied.

It's important to note that these will not make the headphones flat, and so substantial differences (subjectively and objectively) are to be expected even after testing two headphones with these filters.

Mitch's filters only correct up to about 5khz. Anything above that will not be altered.
Additionally, they are based on measurements of the headphones on his own head with in-ear mics. Even if the headphones were both perfectly corrected to flat or a specific target on his own head, they'd still then be different once listened to or measured on any other head or rig since the HRTF would not be the same.

Currently we do not have a good way to actually correct a headphone to flat for any particular listener. Below 2khz differences between listeners are a lot less substantial and so can generally be brought quite close to a common response for most listeners. But once you go above 2khz or so the differences between HRTFs get so large that without an actual accurate picture of one's own specific HRTF it'd be impossible to correct it even if just trying to get within a few dB.

And that's before any HPTF, positional variation or unit variation effects are considered which can all also be quite substantial

EDIT: See post a few messages down for a practical demonstration

 

[Resolve]

It's important to note that these will not make the headphones flat, and so substantial differences (subjectively and objectively) are to be expected even after testing two headphones with these filters.

Mitch's filters only correct up to about 5khz. Anything above that will not be altered.
Additionally, they are based on measurements of the headphones on his own head with in-ear mics. Even if the headphones were both perfectly corrected to flat or a specific target on his own head, they'd still then be different once listened to or measured on any other head or rig since the HRTF would not be the same.

Currently we do not have a good way to actually correct a headphone to flat for any particular listener. Below 2khz differences between listeners are a lot less substantial and so can generally be brought quite close to a common response for most listeners. But once you go above 2khz or so the differences between HRTFs get so large that without an actual accurate picture of one's own specific HRTF it'd be impossible to correct it even if just trying to get within a few dB.

And that's before any HPTF, positional variation or unit variation effects are considered which can all also be quite substantial

Yup, and this also explains why two headphone's EQ'd to match on the graph don't actually sound the same in practice. You have different FR at the ear drum due to coupling effects, positional changes and HpTF behavior variation. Different ears do be wild. I am actually very curious how consistent the Immanis would be across heads, given the unique nature of its design. I would expect it to be more consistent than others but it hasn't been measured on enough heads/ears yet.

Its acoustic impedance is one thing, but I wonder if the multi-ribbon concept may actually be influential there as well. I didn't personally have any coupling issues that I could detect, and I have a big head.

I will note, a lot of this stuff we talk about isn't something the community has come to learn about yet. Blaine's presentation at CanJam was on exactly this, so I hope folks give that a watch once it's been published. It goes a long way to improving the discourse and solving some of the confusion on this topic. People seem to get particularly entrenched in their camps on the subject of measurements, and that's likely due to the existing paradigm of data visualization not being sufficient or even properly understood.

 

[MadEconomist]

It's important to note that these will not make the headphones flat, and so substantial differences (subjectively and objectively) are to be expected even after testing two headphones with these filters.

Mitch's filters only correct up to about 5khz. Anything above that will not be altered.
Additionally, they are based on measurements of the headphones on his own head with in-ear mics. Even if the headphones were both perfectly corrected to flat or a specific target on his own head, they'd still then be different once listened to or measured on any other head or rig since the HRTF would not be the same.

Currently we do not have a good way to actually correct a headphone to flat for any particular listener. Below 2khz differences between listeners are a lot less substantial and so can generally be brought quite close to a common response for most listeners. But once you go above 2khz or so the differences between HRTFs get so large that without an actual accurate picture of one's own specific HRTF it'd be impossible to correct it even if just trying to get within a few dB.

And that's before any HPTF, positional variation or unit variation effects are considered which can all also be quite substantial

I should also note that this doesn't account for errors introduced by the measurement method, which can be quite meaningful if the microphones are improperly placed in a real ear measurement.

 

[GoldenSound]

It's important to note that these will not make the headphones flat, and so substantial differences (subjectively and objectively) are to be expected even after testing two headphones with these filters.

Mitch's filters only correct up to about 5khz. Anything above that will not be altered.
Additionally, they are based on measurements of the headphones on his own head with in-ear mics. Even if the headphones were both perfectly corrected to flat or a specific target on his own head, they'd still then be different once listened to or measured on any other head or rig since the HRTF would not be the same.

Currently we do not have a good way to actually correct a headphone to flat for any particular listener. Below 2khz differences between listeners are a lot less substantial and so can generally be brought quite close to a common response for most listeners. But once you go above 2khz or so the differences between HRTFs get so large that without an actual accurate picture of one's own specific HRTF it'd be impossible to correct it even if just trying to get within a few dB.

And that's before any HPTF, positional variation or unit variation effects are considered which can all also be quite substantial

As a practical example of this, here is a measurement of a Hifiman Susvara on a rig using a KB5000 pinnae (GRAS 43AG for example)

Now here's that measurement again, but overlaid with the DF HRTF (10dB tilt) of this rig in green, and the susvara measured again using Mitch's convolution filter in purple:

The convolution doesn't make it flat, either in terms of 'absolute flat' (which would be bad anyway), or vs the actual DF HRTF of this rig, or Harman, or any other target that one might argue as being neutral.
Some areas like the 2khz dip are much closer to neutral, others like the 4-5khz region are actually now further away, and the midrange peaks seem to have been worsened a bit too.

Why? Well could be any number of things

- Mitch's head and other heads won't have the same HRTF
- Nothing above 5khz is corrected, even just at a basic shelf level (which may have been beneficial here)
- Differing HpTF effects
- Positional variation (this can be huge on some headphones)
- In-Ear microphones are quite finnicky in terms of placement/positioning so that can often be a factor.

To be clear I'm not saying "you shouldn't like Mitch's convolutions". Just that it's important to know what a product is/isn't doing, and it will not actually give you a flat FR for your headphone as this isn't currently possible with the technology we have.

Just to be sure, here's the same thing but with the HD650

Measurement of HD650 on this rig vs DF HRTF:

Now also overlaid with a measurement with Mitch's convolution applied:
Purple: original
red: with mitch's convolution (simulated)
dark green: with mitch's convolution (actual)
lime: DF HRTF of rig

Bass is definitely more linear and up to 1khz ish we see a generally better conformity to neutral, but above that some areas like around 1500hz or 5khz are definitely much further away now.

Again, whether or not this filter is preferable to you is entirely up to you, but I feel it's important to explain that these corrections are NOT actually making your headphone 'flat' and you shouldn't be surprised when you still hear big differences between headphones even after applying them.

Here are susvara and HD650 measurements both after applying the convolutions for example:

Still very substantial differences across the board.
You can also just make these same adjustments (or adjustments that are more accurate for your own head) with EQ

EQ'ed susvara in purple:

 

[nfandi]

As a practical example of this, here is a measurement of a Hifiman Susvara on a rig using a KB5000 pinnae (GRAS 43AG for example)



Now here's that measurement again, but overlaid with the DF HRTF (10dB tilt) of this rig in green, and the susvara measured again using Mitch's convolution filter in purple:




The convolution doesn't make it flat, either in terms of 'absolute flat' (which would be bad anyway), or vs the actual DF HRTF of this rig, or Harman, or any other target that one might argue as being neutral.
Some areas like the 2khz dip are much closer to neutral, others like the 4-5khz region are actually now further away, and the midrange peaks seem to have been worsened a bit too.

Why? Well could be any number of things

- Mitch's head and other heads won't have the same HRTF
- Nothing above 5khz is corrected, even just at a basic shelf level (which may have been beneficial here)
- Differing HpTF effects
- Positional variation (this can be huge on some headphones)
- In-Ear microphones are quite finnicky in terms of placement/positioning so that can often be a factor.

To be clear I'm not saying "you shouldn't like Mitch's convolutions". Just that it's important to know what a product is/isn't doing, and it will not actually give you a flat FR for your headphone as this isn't currently possible with the technology we have.

Just to be sure, here's the same thing but with the HD650

Measurement of HD650 on this rig vs DF HRTF:




Now also overlaid with a measurement with Mitch's convolution applied:
Purple: original
red: with mitch's convolution (simulated)
dark green: with mitch's convolution (actual)
lime: DF HRTF of rig


Bass is definitely more linear and up to 1khz ish we see a generally better conformity to neutral, but above that some areas like around 1500hz or 5khz are definitely much further away now.

Again, whether or not this filter is preferable to you is entirely up to you, but I feel it's important to explain that these corrections are NOT actually making your headphone 'flat' and you shouldn't be surprised when you still hear big differences between headphones even after applying them.

Here are susvara and HD650 measurements both after applying the convolutions for example:


Still very substantial differences across the board.
You can also just make these same adjustments (or adjustments that are more accurate for your own head) with EQ

EQ'ed susvara in purple:

Where’s the moderator now? This is the Immanis thread, why are you showing us graphs of the Susvara and HD650?

 

[GoldenSound]

Where’s the moderator now? This is the Immanis thread, why are you showing us graphs of the Susvara and HD650?

It was relevant to the discussion at hand. Aleksandar brought the convolutions up for the Immanis and so I wanted to chime in and figured others may be interested. This isn't something most people are too aware of

 

[nfandi]

It's important to note that these will not make the headphones flat, and so substantial differences (subjectively and objectively) are to be expected even after testing two headphones with these filters.

Mitch's filters only correct up to about 5khz. Anything above that will not be altered.
Additionally, they are based on measurements of the headphones on his own head with in-ear mics. Even if the headphones were both perfectly corrected to flat or a specific target on his own head, they'd still then be different once listened to or measured on any other head or rig since the HRTF would not be the same.

Currently we do not have a good way to actually correct a headphone to flat for any particular listener. Below 2khz differences between listeners are a lot less substantial and so can generally be brought quite close to a common response for most listeners. But once you go above 2khz or so the differences between HRTFs get so large that without an actual accurate picture of one's own specific HRTF it'd be impossible to correct it even if just trying to get within a few dB.

And that's before any HPTF, positional variation or unit variation effects are considered which can all also be quite substantial

Your statement about Mitch’s filters is factually incorrect. The DSP does actually cut-off around 5-6Khz but Mitch uses his own hearing of his own recorded music to address the high frequencies above the cut-off. So it is not factually correct to say anything above 5khz is not altered.

 

[GoldenSound]

Your statement about Mitch’s filters is factually incorrect. The DSP does actually cut-off around 5-6Khz but Mitch uses his own hearing of his own recorded music to address the high frequencies above the cut-off. So it is not factually correct to say anything above 5khz is not altered.

They do cut off above 5-7khz. It's just flat from there on out

One of the convolutions for example:

To be clear though, this is the RIGHT way to do it. You don't want to be making fine adjustments at high freqs that may not only not work on other heads or with changes in positioning, but could make things much worse.

My point was just that it can't be argued that 'FR doesn't mean anything because two headphones corrected to flat still sound different' when they were never actually corrected flat to begin with (and practically speaking cannot be)

 

[GoldenSound]

So chime in about the convolution filters for the Raal phones, not Susvara or HD650.

Everything said above applies to the Raal headphones. It's not model specific.

I just don't happen to have a Raal headphone here at the moment but have in the past

 

[nfandi]

I should also note that this doesn't account for errors introduced by the measurement method, which can be quite meaningful if the microphones are improperly placed in a real ear measurement.

This is why Mitch takes 30 separate measurements of different placements within the ear.

 

[nfandi]

Yup, and this also explains why two headphone's EQ'd to match on the graph don't actually sound the same in practice. You have different FR at the ear drum due to coupling effects, positional changes and HpTF behavior variation. Different ears do be wild. I am actually very curious how consistent the Immanis would be across heads, given the unique nature of its design. I would expect it to be more consistent than others but it hasn't been measured on enough heads/ears yet.

Its acoustic impedance is one thing, but I wonder if the multi-ribbon concept may actually be influential there as well. I didn't personally have any coupling issues that I could detect, and I have a big head.

I will note, a lot of this stuff we talk about isn't something the community has come to learn about yet. Blaine's presentation at CanJam was on exactly this, so I hope folks give that a watch once it's been published. It goes a long way to improving the discourse and solving some of the confusion on this topic. People seem to get particularly entrenched in their camps on the subject of measurements, and that's likely due to the existing paradigm of data visualization not being sufficient or even properly understood.

If I’m coming across as negative towards you please rest assured I’m really not. I’ve learned so much from reading your various threads and I really loved that statement you made recently on one of the Immanis threads when you said any measurement made that involves an individual’s HRTF is only valid for that one single condition. This is so important but so many people are not recognising that this is a fact. Someone posts a measurement using a clone or dummy head and suddenly these become the definitive measurements that people use. Please keep doing what you’re doing.

 

[betula]

If I’m coming across as negative towards you please rest assured I’m really not. I’ve learned so much from reading your various threads and I really loved that statement you made recently on one of the Immanis threads when you said any measurement made that involves an individual’s HRTF is only valid for that one single condition. This is so important but so many people are not recognising that this is a fact. Someone posts a measurement using a clone or dummy head and suddenly these become the definitive measurements that people use. Please keep doing what you’re doing.

Yepp. We just all need to walk that fine line of knowledge and arrogance carefully.

 

[nfandi]

Yup, and this also explains why two headphone's EQ'd to match on the graph don't actually sound the same in practice. You have different FR at the ear drum due to coupling effects, positional changes and HpTF behavior variation. Different ears do be wild. I am actually very curious how consistent the Immanis would be across heads, given the unique nature of its design. I would expect it to be more consistent than others but it hasn't been measured on enough heads/ears yet.

Its acoustic impedance is one thing, but I wonder if the multi-ribbon concept may actually be influential there as well. I didn't personally have any coupling issues that I could detect, and I have a big head.

I will note, a lot of this stuff we talk about isn't something the community has come to learn about yet. Blaine's presentation at CanJam was on exactly this, so I hope folks give that a watch once it's been published. It goes a long way to improving the discourse and solving some of the confusion on this topic. People seem to get particularly entrenched in their camps on the subject of measurements, and that's likely due to the existing paradigm of data visualization not being sufficient or even properly understood.

This is why Alex measures his phones with no coupling effect against a flat surface. Speakers are measured in an anechoic chamber to avoid any interactions with reflective surfaces, so why are so many measurements made for headphones using dummy heads? As soon as you use a dummy head you are making that measurement specific to that single condition.

 

[Resolve]

This is why Alex measures his phones with no coupling effect against a flat surface. Speakers are measured in an anechoic chamber to avoid any interactions with reflective surfaces, so why are so many measurements made for headphones using dummy heads? As soon as you use a dummy head you are making that measurement specific to that single condition.

Ah, that one probably needs to be addressed a bit. I think the latter point about measuring on any HATS or dummy head that has a pinna making it specific to the condition of that head and ears is a good thing to consider. However... you do always have to measure using some pinna because that's how it's actually used in practice. So there's an ear physically inside the space, and it's part of the system created by coupling the headphone to the side of the head - what gets created when actually wearing the headphone. This is also why industry standard measurement equipment is designed to be anthropometric - you're giving an indication of how a human could hear these products. Is it every human? no. Is it the average human? not necessarily. And there are valid questions about how great the variation is for pinna effects across a wide population. But at the very least it's humanlike. It wouldn't make sense to put any other ear in there, nor would it make sense to put no ear in there.

 

[MadEconomist]

This is why Alex measures his phones with no coupling effect against a flat surface. Speakers are measured in an anechoic chamber to avoid any interactions with reflective surfaces, so why are so many measurements made for headphones using dummy heads? As soon as you use a dummy head you are making that measurement specific to that single condition.

Because that single condition is the condition of use. Speakers are affected by air pressure and temperature as well, but we traditionally do not consider 0K vacuum measurements the gold standard for assessing their performance

This is why Mitch takes 30 separate measurements of different placements within the ear.

This does not necessarily help. It's more ideal to use a single, highly consistent placement for the microphone. Possibly as a result of issues here, here's the predicted FR using GRAS 45CA/KB501x measurements and Mitch's EQs:

Not very surprising that one might hear a difference there, I would say.