• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Question on in-ear microphones

Reverb

Member
Joined
Feb 2, 2022
Messages
5
Likes
0
Hi all,
Mostly for curiosity's sake, I'm considering getting some in-ear microphones that I've seen mentioned in several threads: SP-TFB-2. However, in those posts, I see several mentions that these are good for "relative" and not "absolute" measurements. And also, implications that responses at higher frequencies should be ignored. For example, these posts:
and

I'm most interested in comparing the general shape of the FR curve between headphones. Also to see what effect, if any, various EQ's have on the curve. I don't care so much whether the absolute volumes at given frequencies are the same between headphones. Are these microphones good enough to give a rough picture of the curve?

Also, in the second post it mentions that only the LF curves (e.g. below 800 Hz) are valid. The mics are supposedly flat from 20Hz to 20KHz. Why are measurements at higher frequencies less accurate or not valid? Is this due to the wavelengths and geometry of the ear?
Thanks!
 

pozz

Слава Україні
Forum Donor
Editor
Joined
May 21, 2019
Messages
4,036
Likes
6,827
These "in ear" microphones are not quite that. They block your ear canal, which determines the shape of high frequencies. You can still get some useful midrange information for headphones since the concha effects from the shape of your ear will be preserved. And provided you are consistent in how you put headphones on, you can get good bass information. But your measurements will look different from those of any reviewer. And you will still have to mostly listen to make corrections to the highs.

That models looks fine, but it's hard to say outright since the specs are so scant. It would be better if they included graphs of unmounted frequency response, polar pattern, channel deviation and self noise.

Longer explanation here:
Take the measurements you see on ASR. 1kHz and below there are effects due to seal. Above that there effects due to the shape and material of the simulated ear (or flat plate, or whatever kind of rig is used). The only rig which specifically models the ear accurately at the highest frequencies is the B&K 5128, but as far as I know there are no published target curves which present what ideal IEM or headphone responses should look like on it and why. And the more common research level rigs, like those by GRAS, do not provide meaningfully accurate data beyond 10kHz.

The further issue is that the curves you are seeing are the effects of a system: the headphone + the simulated ear. Same with speakers: the speakers + room form one system. If, for example, you inserted perfect probe microphones far enough for them to sit at your eardrums and ran a sweep, each headphone curve would look very different. This is, like you said, due to geometry and wavelengths. The list goes on (compensations, scaling, smoothing, averaging, fitment). It makes it difficult to compare measurements between reviewers since the approaches will always differ in some important respect.

There is no way to extrapolate, from a given set of data on one rig, what it would look like on another. You may see some people scrape data from graphs from various reviewers and then plot them together. Typically those comparisons aren't meaningful, in the sense that the differences are hard to definitively explain. For example, the Harman curve that everyone talks here about is only valid for the GRAS 45CA.

It's still better to have this mess of information than none. Where the details matter is if you're shooting for the best. Despite all the work done so far, you can't just follow the EQ suggestions of a reviewer without some experimentation of your own. Generally speaking, the only EQing you should be doing is EQing you can hear. As I'm sure you know, the goal isn't a textbook curve and tiny, meticulous corrections.
 
OP
R

Reverb

Member
Joined
Feb 2, 2022
Messages
5
Likes
0
These "in ear" microphones are not quite that.
This is great info, thanks very much! I'm not trying to match curves published by anyone else; just curious how they perform when actually mounted on my head. I have two sets of headphones (AudioTechnica ATH-M50 and Sony WH-1000XM4) and would like to see their "objective" FR vs my subjective opinion. And also to see what effect EQ has on either of them. The circular clips that fit into your ear when wearing them probably changes the measured response compared to the ear without those clips. Still, curious to see what kinds of measurements I get with these.

There are a few specs on their website (copied below). In the text description they claim they have a "flat" frequency response. It would be nice if they defined what they mean by flat...
Frequency Response – 20-20,000 Hz
Signal To Noise Ratio – Standard sensitivity 58dB; High sensitivity 60dB
Open Circuit Sensitivity – Standard Sensitivity -42dB; High sensitivity -30dB
Maximum Input Sound Level – 120dB
Dynamic Range – 96dB
Polar Pattern – Omnidirectional

They block your ear canal, which determines the shape of high frequencies. You can still get some useful midrange information for headphones since the concha effects from the shape of your ear will be preserved. And provided you are consistent in how you put headphones on, you can get good bass information. But your measurements will look different from those of any reviewer. And you will still have to mostly listen to make corrections to the highs.

That models looks fine, but it's hard to say outright since the specs are so scant. It would be better if they included graphs of unmounted frequency response, polar pattern, channel deviation and self noise.

Longer explanation here:
Take the measurements you see on ASR. 1kHz and below there are effects due to seal. Above that there effects due to the shape and material of the simulated ear (or flat plate, or whatever kind of rig is used). The only rig which specifically models the ear accurately at the highest frequencies is the B&K 5128, but as far as I know there are no published target curves which present what ideal IEM or headphone responses should look like on it and why. And the more common research level rigs, like those by GRAS, do not provide meaningfully accurate data beyond 10kHz.

The further issue is that the curves you are seeing are the effects of a system: the headphone + the simulated ear. Same with speakers: the speakers + room form one system. If, for example, you inserted perfect probe microphones far enough for them to sit at your eardrums and ran a sweep, each headphone curve would look very different. This is, like you said, due to geometry and wavelengths. The list goes on (compensations, scaling, smoothing, averaging, fitment). It makes it difficult to compare measurements between reviewers since the approaches will always differ in some important respect.

There is no way to extrapolate, from a given set of data on one rig, what it would look like on another. You may see some people scrape data from graphs from various reviewers and then plot them together. Typically those comparisons aren't meaningful, in the sense that the differences are hard to definitively explain. For example, the Harman curve that everyone talks here about is only valid for the GRAS 45CA.

It's still better to have this mess of information than none. Where the details matter is if you're shooting for the best. Despite all the work done so far, you can't just follow the EQ suggestions of a reviewer without some experimentation of your own. Generally speaking, the only EQing you should be doing is EQing you can hear. As I'm sure you know, the goal isn't a textbook curve and tiny, meticulous correct
These "in ear" microphones are not quite that. They block your ear canal, which determines the shape of high frequencies. You can still get some useful midrange information for headphones since the concha effects from the shape of your ear will be preserved. And provided you are consistent in how you put headphones on, you can get good bass information. But your measurements will look different from those of any reviewer. And you will still have to mostly listen to make corrections to the highs.

That models looks fine, but it's hard to say outright since the specs are so scant. It would be better if they included graphs of unmounted frequency response, polar pattern, channel deviation and self noise.

Longer explanation here:
Take the measurements you see on ASR. 1kHz and below there are effects due to seal. Above that there effects due to the shape and material of the simulated ear (or flat plate, or whatever kind of rig is used). The only rig which specifically models the ear accurately at the highest frequencies is the B&K 5128, but as far as I know there are no published target curves which present what ideal IEM or headphone responses should look like on it and why. And the more common research level rigs, like those by GRAS, do not provide meaningfully accurate data beyond 10kHz.

The further issue is that the curves you are seeing are the effects of a system: the headphone + the simulated ear. Same with speakers: the speakers + room form one system. If, for example, you inserted perfect probe microphones far enough for them to sit at your eardrums and ran a sweep, each headphone curve would look very different. This is, like you said, due to geometry and wavelengths. The list goes on (compensations, scaling, smoothing, averaging, fitment). It makes it difficult to compare measurements between reviewers since the approaches will always differ in some important respect.

There is no way to extrapolate, from a given set of data on one rig, what it would look like on another. You may see some people scrape data from graphs from various reviewers and then plot them together. Typically those comparisons aren't meaningful, in the sense that the differences are hard to definitively explain. For example, the Harman curve that everyone talks here about is only valid for the GRAS 45CA.

It's still better to have this mess of information than none. Where the details matter is if you're shooting for the best. Despite all the work done so far, you can't just follow the EQ suggestions of a reviewer without some experimentation of your own. Generally speaking, the only EQing you should be doing is EQing you can hear. As I'm sure you know, the goal isn't a textbook curve and tiny, meticulous corrections.
These "in ear" microphones are not quite that. They block your ear canal, which determines the shape of high frequencies. You can still get some useful midrange information for headphones since the concha effects from the shape of your ear will be preserved. And provided you are consistent in how you put headphones on, you can get good bass information. But your measurements will look different from those of any reviewer. And you will still have to mostly listen to make corrections to the highs.

That models looks fine, but it's hard to say outright since the specs are so scant. It would be better if they included graphs of unmounted frequency response, polar pattern, channel deviation and self noise.

Longer explanation here:
Take the measurements you see on ASR. 1kHz and below there are effects due to seal. Above that there effects due to the shape and material of the simulated ear (or flat plate, or whatever kind of rig is used). The only rig which specifically models the ear accurately at the highest frequencies is the B&K 5128, but as far as I know there are no published target curves which present what ideal IEM or headphone responses should look like on it and why. And the more common research level rigs, like those by GRAS, do not provide meaningfully accurate data beyond 10kHz.

The further issue is that the curves you are seeing are the effects of a system: the headphone + the simulated ear. Same with speakers: the speakers + room form one system. If, for example, you inserted perfect probe microphones far enough for them to sit at your eardrums and ran a sweep, each headphone curve would look very different. This is, like you said, due to geometry and wavelengths. The list goes on (compensations, scaling, smoothing, averaging, fitment). It makes it difficult to compare measurements between reviewers since the approaches will always differ in some important respect.

There is no way to extrapolate, from a given set of data on one rig, what it would look like on another. You may see some people scrape data from graphs from various reviewers and then plot them together. Typically those comparisons aren't meaningful, in the sense that the differences are hard to definitively explain. For example, the Harman curve that everyone talks here about is only valid for the GRAS 45CA.

It's still better to have this mess of information than none. Where the details matter is if you're shooting for the best. Despite all the work done so far, you can't just follow the EQ suggestions of a reviewer without some experimentation of your own. Generally speaking, the only EQing you should be doing is EQing you can hear. As I'm sure you know, the goal isn't a textbook curve and tiny, meticulous corrections.
These "in ear" microphones are not quite that. They block your ear canal, which determines the shape of high frequencies. You can still get some useful midrange information for headphones since the concha effects from the shape of your ear will be preserved. And provided you are consistent in how you put headphones on, you can get good bass information. But your measurements will look different from those of any reviewer. And you will still have to mostly listen to make corrections to the highs.

That models looks fine, but it's hard to say outright since the specs are so scant. It would be better if they included graphs of unmounted frequency response, polar pattern, channel deviation and self noise.

Longer explanation here:
Take the measurements you see on ASR. 1kHz and below there are effects due to seal. Above that there effects due to the shape and material of the simulated ear (or flat plate, or whatever kind of rig is used). The only rig which specifically models the ear accurately at the highest frequencies is the B&K 5128, but as far as I know there are no published target curves which present what ideal IEM or headphone responses should look like on it and why. And the more common research level rigs, like those by GRAS, do not provide meaningfully accurate data beyond 10kHz.

The further issue is that the curves you are seeing are the effects of a system: the headphone + the simulated ear. Same with speakers: the speakers + room form one system. If, for example, you inserted perfect probe microphones far enough for them to sit at your eardrums and ran a sweep, each headphone curve would look very different. This is, like you said, due to geometry and wavelengths. The list goes on (compensations, scaling, smoothing, averaging, fitment). It makes it difficult to compare measurements between reviewers since the approaches will always differ in some important respect.

There is no way to extrapolate, from a given set of data on one rig, what it would look like on another. You may see some people scrape data from graphs from various reviewers and then plot them together. Typically those comparisons aren't meaningful, in the sense that the differences are hard to definitively explain. For example, the Harman curve that everyone talks here about is only valid for the GRAS 45CA.

It's still better to have this mess of information than none. Where the details matter is if you're shooting for the best. Despite all the work done so far, you can't just follow the EQ suggestions of a reviewer without some experimentation of your own. Generally speaking, the only EQing you should be doing is EQing you can hear. As I'm sure you know, the goal isn't a textbook curve and tiny, meticulous corrections.
 

MayaTlab

Addicted to Fun and Learning
Joined
Aug 15, 2020
Messages
924
Likes
1,512
Also, in the second post it mentions that only the LF curves (e.g. below 800 Hz) are valid. The mics are supposedly flat from 20Hz to 20KHz. Why are measurements at higher frequencies less accurate or not valid?


For relative measurements between headphones above 1kHz or so, I personally feel a lot more confident about the results I obtain when I am able to criss-cross the results from different measurement methods, all preferably at least at the ear canal entrance point or further, and repeat the results across multiple headphones seatings and sessions.

Among these method the easiest to DIY yourself are blocked ear canal entrance mics, which may show some relative differences errors above 1kHz. For me so far I believe that these errors are mostly concentrated in the 2-3kHz band (because of the blocked canal) and above 7kHz (no idea why :D). This is where I'd start, while acknowledging their limitation.

They block your ear canal,

The SP-TFB-2 mics leave the ear canal open :D.
But Sound Professionals sell mics that may be used for blocked ear canal measurements, and the SP-TFB-2 can be modified for that application as well.

Take the measurements you see on ASR. 1kHz and below there are effects due to seal. Above that there effects due to the shape and material of the simulated ear (or flat plate, or whatever kind of rig is used).

Resolve often employs the term "coupling", which I like quite a lot as it avoids pinpointing an exact cause for some of the observations made, which often aren't easy to determine.
Here's an example of some interesting results below 1kHz that can't explained by either seal or sample variation :
 
OP
R

Reverb

Member
Joined
Feb 2, 2022
Messages
5
Likes
0
More great stuff! The posts on the 5909 thread are very interesting. I guess I shouldn't be surprised that other folks have already gone really far down this road. Apologies if you documented it elsewhere, but can you describe the test setup you used for your measurements? It appears that a closed-ear mic is preferable. I believe my soundcard will capture a single channel at 48K 16b resolution. For just capturing general shape of FR curves, that seems adequate to me. The SP-TFB-2 is stereo, and requires power, so may not be compatible with my soundcard's mic in. And as you mentioned, it's open ear canal.
Thanks!
Edit: I just found your poste here: https://www.head-fi.org/threads/how...guarantee-a-better-sound.958201/post-16405751
Is this the mic setup you used in measurements referenced in these posts?
For relative measurements between headphones above 1kHz or so, I personally feel a lot more confident about the results I obtain when I am able to criss-cross the results from different measurement methods, all preferably at least at the ear canal entrance point or further, and repeat the results across multiple headphones seatings and sessions.

Among these method the easiest to DIY yourself are blocked ear canal entrance mics, which may show some relative differences errors above 1kHz. For me so far I believe that these errors are mostly concentrated in the 2-3kHz band (because of the blocked canal) and above 7kHz (no idea why :D). This is where I'd start, while acknowledging their limitation.



The SP-TFB-2 mics leave the ear canal open :D.
But Sound Professionals sell mics that may be used for blocked ear canal measurements, and the SP-TFB-2 can be modified for that application as well.



Resolve often employs the term "coupling", which I like quite a lot as it avoids pinpointing an exact cause for some of the observations made, which often aren't easy to determine.
Here's an example of some interesting results below 1kHz that can't explained by either seal or sample variation :
 
Last edited:

MayaTlab

Addicted to Fun and Learning
Joined
Aug 15, 2020
Messages
924
Likes
1,512
Is this the mic setup you used in measurements referenced in these posts?

In the posts in the 5909 thread, I used blocked ear canal entrance microphones. This is what I'd recommend to start with.

I think that it's a good idea to start on your end with an experimental mindset, I don't think that I'm qualified to tell you "this is what you should use, this is best" as I'm also in the continuous process of finding out what works best for me. But for these mics this is what I use :
- SP-TFB-2 mics with the wings cut off, wrapped in silicone foam tape around the circumference. For me this turned out to be exactly what ensured good repeatability from measurement session to measurement session (mics positioned in the same place, visibly flush with the ear canal entrance - but IDK if that is exactly equivalent to the "acoustic" location of the ear canal entrance). For you another setup might work better to achieve that goal (I've often seen people shoving the electret inside a foam earplug for example).
- Zoom H1N as the preamp / USB ADC.
- Room EQ Wizard.
I also use a UMIK-1 and a DIY near-field "compensation rig" to get a rough idea of whether or not the SP-TFB-2 mics's FR is decently accurate (and occasionally compensate the measurements). I recommend spending quite a bit of time characterising them to check for potential problems.
An advantage of blocked ear canal entrance mics is that you can use fairly loud signals which can help with noise / environment.

I'm experimenting with open ear canal entrance mics, which honestly would have my preference I guess as a baseline, but I'm not satisfied with the designs I've made so far. So I tend to use another set of SP in-concha mics to check the response with open ear canals besides my DIY probe mics, but, particularly for smaller / closed earcups I don't think that I'm getting particularly accurate relative results past 3kHz with these.

I've also made DIY probe mics, regularly updated, the current ones I'm starting to use these days have little in common with David Griesinger's approach above. But this is quite another kettle of fish and comes with lots of drawbacks and difficulties, I don't recommend starting with these at all.

And I also use a set of very small electret mics (ex : https://soundprofessionals.com/product/MS-CB-900/) that I can shove in the ear canal while keeping it somewhat open. I've started to try to understand how the Airpods 3's feedback mechanism works and these mics, I think, are quite ideal for that.

So I use the following types of mics :
- DIY blocked ear canal entrance mics
- DIY open ear canal entrance mics
- DIY probe mics
- in concha mics, either with the ear canal open, or closed with foam plugs
- small electrets, that I can place anywhere I want (ear canal entrance, inside the canal, in concha, at various points in the front volume).
It's the use of them all that's making me a little more confident in the results I get, each one of them has advantages and drawbacks over the others.

Note that each one of these methods will produce different absolute values and should not be compared with each others. What I look at is, if I use a pair of headphones as a baseline (ex : my pair of HD650), are they recording the same relative difference between them and other headphones, or not ?

I think that it's a good idea to read some of the available literature on the subject of in-ear measurements as well. I understand it only very partially, but that was enough to steer my experimentations in more productive way. You can search for anything published by "Hammershøi" and "Møller" for example, they've written several articles touching on that subject.
 
OP
R

Reverb

Member
Joined
Feb 2, 2022
Messages
5
Likes
0
In the posts in the 5909 thread, I used blocked ear canal entrance microphones. This is what I'd recommend to start with.
Thank you! This gives me a place to start. I understand that these are all relative (i.e. 0dB is not really 0dB). Looks like I've got enough hear to keep me busy for awhile :)
I really appreciate yours and @pozz inputs!
I think that it's a good idea to start on your end with an experimental mindset, I don't think that I'm qualified to tell you "this is what you should use, this is best" as I'm also in the continuous process of finding out what works best for me. But for these mics this is what I use :
- SP-TFB-2 mics with the wings cut off, wrapped in silicone foam tape around the circumference. For me this turned out to be exactly what ensured good repeatability from measurement session to measurement session (mics positioned in the same place, visibly flush with the ear canal entrance - but IDK if that is exactly equivalent to the "acoustic" location of the ear canal entrance). For you another setup might work better to achieve that goal (I've often seen people shoving the electret inside a foam earplug for example).
- Zoom H1N as the preamp / USB ADC.
- Room EQ Wizard.
I also use a UMIK-1 and a DIY near-field "compensation rig" to get a rough idea of whether or not the SP-TFB-2 mics's FR is decently accurate (and occasionally compensate the measurements). I recommend spending quite a bit of time characterising them to check for potential problems.
An advantage of blocked ear canal entrance mics is that you can use fairly loud signals which can help with noise / environment.

I'm experimenting with open ear canal entrance mics, which honestly would have my preference I guess as a baseline, but I'm not satisfied with the designs I've made so far. So I tend to use another set of SP in-concha mics to check the response with open ear canals besides my DIY probe mics, but, particularly for smaller / closed earcups I don't think that I'm getting particularly accurate relative results past 3kHz with these.

I've also made DIY probe mics, regularly updated, the current ones I'm starting to use these days have little in common with David Griesinger's approach above. But this is quite another kettle of fish and comes with lots of drawbacks and difficulties, I don't recommend starting with these at all.

And I also use a set of very small electret mics (ex : https://soundprofessionals.com/product/MS-CB-900/) that I can shove in the ear canal while keeping it somewhat open. I've started to try to understand how the Airpods 3's feedback mechanism works and these mics, I think, are quite ideal for that.

So I use the following types of mics :
- DIY blocked ear canal entrance mics
- DIY open ear canal entrance mics
- DIY probe mics
- in concha mics, either with the ear canal open, or closed with foam plugs
- small electrets, that I can place anywhere I want (ear canal entrance, inside the canal, in concha, at various points in the front volume).
It's the use of them all that's making me a little more confident in the results I get, each one of them has advantages and drawbacks over the others.

Note that each one of these methods will produce different absolute values and should not be compared with each others. What I look at is, if I use a pair of headphones as a baseline (ex : my pair of HD650), are they recording the same relative difference between them and other headphones, or not ?

I think that it's a good idea to read some of the available literature on the subject of in-ear measurements as well. I understand it only very partially, but that was enough to steer my experimentations in more productive way. You can search for anything published by "Hammershøi" and "Møller" for example, they've written several articles touching on that subject.
 
Top Bottom