9 ms·

AutoEQ: Automatic headphone equalization from frequency responses

184 points by ishitatsuyuki 5y ago

serverholic 5y ago
FYI, EQs will distort audio. Especially if the adjustments are large. Your average EQ will introduce phase shifts to the various frequencies that make up a sound. Linear-phase EQs don't phase shift but they also introduce pre-ringing and post-ringing effects.
- bitbang 5y ago
  Depends largely on how it's implemented. There is a big difference between the quality of a well implemented convolution engine and a crappy biquad filter.
  - l33tbro 5y ago
    Where would something like Fabfilter fall in the scale? I imagine there's some pretty complex stuff going on under the hood.
    - serverholic 5y ago
      Fabfilter isn't really doing anything magical. Its default mode is a pretty normal EQ with phase shifting. Linear-phase mode is like any other linear-phase mode with the same drawbacks.
      - l33tbro 5y ago
        Probably the UI that made me think it was a Wonka creation. It's really not sonically much different to a stock EQ.
  - nitrogen 5y ago
    Note that not all biquad filters are crappy.
- ishitatsuyuki 5y ago
  Analog distortion of signals also affects the phase. A digital filter might or might not compensate for it, but it's not always bad. Also, human ears tend to be more sensitive to amplitude than to the phase. Especially if the same filter is applied on both channels, which leaves the phase difference unchanged.
  - serverholic 5y ago
    It's more than just human ear sensitivity. Phase shifting changes the relationship between frequencies in a sound. For example, phase shifting can change the amplitude of a sound. Analog hardware is especially sensitive to changes in amplitude so you might be introducing distortion just by shifting phase.
- h2odragon 5y ago
  At headphone sizes, does your DAC have a good enough clock to even speak about phasing?
- praash 5y ago
  Phase shifts make no practical difference when you're equalizing headphones. You really shouldn't need to use notch filters there. EQ phase is relevant if the dry signal has a chance to mix with the EQ output. Changes in phase is not "distortion". When you end up making large boosts (especially to make sub bass audible), that's when your headphone drivers might start to distort sound. This is fixed with less volume, obviously.
  - serverholic 5y ago
    An EQ that cuts frequencies can actually increase the amplitude of the waveform due to phase shifts. It's not just boosting. What I'm talking about is more relevant on analog gear which is more sensitive to volume.
- willis936 5y ago
  The primary measurement is called group delay, which is the derivative of phase with respect to frequency (since a linear phase is just a uniform delay). Fortunately we have studies from the 70s publicly available that establish a group delay audibility threshold. FIRs add vanishingly small amounts of distortion. Their real drawback is in the added delay. For asynchronous music playback this is nothing to worry about, but for video or anything interactive (communications or gaming) it's going to be a real tough pill to swallow.
thih9 5y ago
It even has settings for Apple’s Ear Pods [1] (the wired earbuds). I remember being surprised how applying these values makes them resemble regular headphones. [1]: https://github.com/jaakkopasanen/AutoEq/tree/master/results/oratory1990/harman_in-ear_2019v2/Apple%20EarPods https://github.com/jaakkopasanen/AutoEq/tree/master/results/...
kohlerm 5y ago
it's great project. No idea why some people do not want to eq their headphones at least a little bit. It can make a huge difference. Also frequency response is IMHO at least partially a matter of taste.
- h2odragon 5y ago
  > a matter of taste Taste, yes; and individual circumstances. My hearing is so good it might count as a disability, but I know where I have peaks and valleys in my sensitivity and a good EQ can help with the spikes taken by tinnitus ringing from youthful big boom car stereo work and explosives. I checked their suggestions for my headphones against my EQ profile and theirs is pretty good. I like more lowfreq and much less high freq than the "flat" they're correcting for. They're offering a great resource for skipping the "what does this set of cans sound like" stage. I probably spent 60hr or more dialing these in when i got them.
  - mckirk 5y ago
    In which sense is your hearing 'so good it might count as a disability'?
    - h2odragon 5y ago
      https://en.wikipedia.org/wiki/Hyperacusis https://en.wikipedia.org/wiki/Hyperacusis I don't hear "notes" and "chords", I hear frequencies. My range (at near 50) still goes up to where i can hear the bats talking. In a room with people, i cannot not hear their pulse, breathing, and other biological proceedings...
      - O_H_E 5y ago
        > i cannot not hear their pulse, breathing, and other biological proceedings Oh WOW, that is ...... wild and interesting
    - pvarangot 5y ago
      I had it as a kid, went away in my teen years and now I only have tinnitus. For me it was only on certain frequencies that were boosted and they are near the principal harmonic of my always-there ringing tinnitus or in the frequency of another noise I sometimes hear that's more like a pure sinewave. As far as I know it's more of a brain thing and not like a super-ear thing.
Bayart 5y ago
The precompiled AutoEQ/results/ tree has been a godsend, and so is Wavelet. I honestly would think about buying a pair of headphones that's not in it, unless I'm actively looking for a specific sound signature (but it's really all about price and comfort as far as I'm concerned).
rweichler 5y ago
If you have an iPhone X or older, you can jailbreak it and use EQE (https://eqe.fm https://eqe.fm) which is system-wide and has AutoEQ integration built-in
scblock 5y ago
I would caution against applying an EQ curve to headphones without review beforehand. If EQ includes large boosts (which the example image and table shows) that can easily introduce clipping distortion if applied linearly without appropriate headroom adjustments. Additionally, depending on transducer performance large bass boosts may simply increase transducer distortion in the bass frequency. Some headphones respond well to EQ, and others not so much. I am not opposed to EQ, but it should be applied judiciously, and at least partially by ear. I don't believe trying to exactly match a target curve will necessarily provide good results compared to a more judicious approach.
- sanjiwatsuki 5y ago
  AFAIK, every tool that integrates with these things apply a negative boost equal to the peak added to avoid clipping.
  - dsr_ 5y ago
    And every recommendation on the site starts with the appropriate amount of negative gain. But if you miss it, somehow, then setting it to the next whole number of decibels larger than the largest gain will do.
- scblock 5y ago
  As an example of my point for using a judicious approach above, applying either of the sets of Sennheiser HD8XX measurements and resulting AutoEQ curves from the Crinacle and oratory1990 folders in the Roon DSP system collapses the headstage and tilts the perceived tone of the headphones from relatively full and engaging with a dip in the 2-3 kHz region to thin and hollow, with an overly bright, brittle top end. Also, these are two sets of measurements of theoretically the same headphones, each with 10 adjustment points, some quite broad, some very narrow. The overall shape of the resulting curve for the same target compensation is similar, but they have some significant differences in the specifics, and they sound different. We have to remember that measurement systems and individual measurement setup can vary quite a bit, so settings from this tool will bake all of that in as well. As a comparison, based on review of the measured curves in more of a a big picture way, applying a much simpler EQ with a broad 1.5 dB bass lift up to about 100 Hz, another broad lift of about 3 dB around 2 kHz, and a slight drop of about 1 dB centered around 8 kHz brings the bass and vocals up a little but keeps headstage and overall tone intact. Maybe it's this particular DSP implementation, but I would be wary about trying to use any of the precompiled results directly. EQ can have real benefits, but is going to be more personal than automated settings will capture.
natdempk 5y ago
It would be interesting to compare the suggested EQ curves here vs. commercial offerings that claim to introduce a neutral response like Sonarworks. In theory I think they should be doing quite similar things, but curious if there are any real differences or if they land in the same general area.
- dsr_ 5y ago
  This is exactly what Sonarworks is doing, but without a specific software package to implement it (and sell to you).
rokweom 5y ago
Autoeq is not just a collection of ready-made EQ settings for different headphones, it's also a database of measurements from few different review sites and a tool for creating custom EQ curves. Do you have a specific target that you like? You can create a custom curve tuned for that target. Do you want to know how a certain model of headphones sounds? You can make your current ones sound almost exactly like them. Really cool project.
londons_explore 5y ago
All this assumes the headphones are linear devices... which clearly isn't the case! None of these tools seem to do anything but rudimentary nonlinear correction... Even simple things like measuring and cancelling harmonics nobody seems to do...
khimaros 5y ago
would it make any sense to use this for tuning a car stereo? particularly, to compensate for engine or road noise?
- dsr_ 5y ago
  Not really. That's an immediate noise cancellation function, which only works (a) with headphones and (b) by recording the outside noise and adding its inverse to the sound being played.
  - khimaros 5y ago
    i don't mean necessarily doing active NC. i was more curious about doing an initial tune with the car at idle or while driving on the road as a way to compensate for the average. is there any value in doing something like that? maybe even just to tune the EQ profile of the car.
    - TylerE 5y ago
      No, as the frequencies involved will vary based on speed, engine RPM, and what seat you're sitting in.
    - jerrysievert 5y ago
      some cars/car stereos have had auto eq'ing with an active microphone in the past, I have to assume they still do. my 2007 has active eq'ing to counteract road noise, and a different eq setting when the top is up, based on a switch. it's a Bose stereo.
tanvach 5y ago
I created and opened up the design of a standalone DSP [1] that leverages AutoEQ. The DSP is optimized for our home brewed headphones for VR, but I’ve been using it for my desktop headphones too. AutoEQ is not a perfect tool, but an amazing (and free) starting point to get the most out of your audio systems. [1] https://github.com/tanvach/prettygood_dsp https://github.com/tanvach/prettygood_dsp
starky 5y ago
Unfortunately on Windows is uses EqualizerAPO which has to be re-enabled every time Windows is updated or it stops working. I've given up on my headphone EQs because of it.

184 points by ishitatsuyuki 5y ago

FYI, EQs will distort audio. Especially if the adjustments are large. Your average EQ will introduce phase shifts to the various frequencies that make up a sound. Linear-phase EQs don't phase shift but they also introduce pre-ringing and post-ringing effects.

bitbang 5y ago
Depends largely on how it's implemented. There is a big difference between the quality of a well implemented convolution engine and a crappy biquad filter.
- l33tbro 5y ago
  Where would something like Fabfilter fall in the scale? I imagine there's some pretty complex stuff going on under the hood.
  - serverholic 5y ago
    Fabfilter isn't really doing anything magical. Its default mode is a pretty normal EQ with phase shifting. Linear-phase mode is like any other linear-phase mode with the same drawbacks.
    - l33tbro 5y ago
      Probably the UI that made me think it was a Wonka creation. It's really not sonically much different to a stock EQ.
- nitrogen 5y ago
  Note that not all biquad filters are crappy.
ishitatsuyuki 5y ago
Analog distortion of signals also affects the phase. A digital filter might or might not compensate for it, but it's not always bad. Also, human ears tend to be more sensitive to amplitude than to the phase. Especially if the same filter is applied on both channels, which leaves the phase difference unchanged.
- serverholic 5y ago
  It's more than just human ear sensitivity. Phase shifting changes the relationship between frequencies in a sound. For example, phase shifting can change the amplitude of a sound. Analog hardware is especially sensitive to changes in amplitude so you might be introducing distortion just by shifting phase.
h2odragon 5y ago
At headphone sizes, does your DAC have a good enough clock to even speak about phasing?
praash 5y ago
Phase shifts make no practical difference when you're equalizing headphones. You really shouldn't need to use notch filters there. EQ phase is relevant if the dry signal has a chance to mix with the EQ output. Changes in phase is not "distortion". When you end up making large boosts (especially to make sub bass audible), that's when your headphone drivers might start to distort sound. This is fixed with less volume, obviously.
- serverholic 5y ago
  An EQ that cuts frequencies can actually increase the amplitude of the waveform due to phase shifts. It's not just boosting. What I'm talking about is more relevant on analog gear which is more sensitive to volume.
willis936 5y ago
The primary measurement is called group delay, which is the derivative of phase with respect to frequency (since a linear phase is just a uniform delay). Fortunately we have studies from the 70s publicly available that establish a group delay audibility threshold. FIRs add vanishingly small amounts of distortion. Their real drawback is in the added delay. For asynchronous music playback this is nothing to worry about, but for video or anything interactive (communications or gaming) it's going to be a real tough pill to swallow.

thih9 5y ago

It even has settings for Apple’s Ear Pods [1] (the wired earbuds). I remember being surprised how applying these values makes them resemble regular headphones. [1]: https://github.com/jaakkopasanen/AutoEq/tree/master/results/oratory1990/harman_in-ear_2019v2/Apple%20EarPods https://github.com/jaakkopasanen/AutoEq/tree/master/results/...

kohlerm 5y ago

it's great project. No idea why some people do not want to eq their headphones at least a little bit. It can make a huge difference. Also frequency response is IMHO at least partially a matter of taste.

h2odragon 5y ago
> a matter of taste Taste, yes; and individual circumstances. My hearing is so good it might count as a disability, but I know where I have peaks and valleys in my sensitivity and a good EQ can help with the spikes taken by tinnitus ringing from youthful big boom car stereo work and explosives. I checked their suggestions for my headphones against my EQ profile and theirs is pretty good. I like more lowfreq and much less high freq than the "flat" they're correcting for. They're offering a great resource for skipping the "what does this set of cans sound like" stage. I probably spent 60hr or more dialing these in when i got them.
- mckirk 5y ago
  In which sense is your hearing 'so good it might count as a disability'?
  - h2odragon 5y ago
    https://en.wikipedia.org/wiki/Hyperacusis https://en.wikipedia.org/wiki/Hyperacusis I don't hear "notes" and "chords", I hear frequencies. My range (at near 50) still goes up to where i can hear the bats talking. In a room with people, i cannot not hear their pulse, breathing, and other biological proceedings...
    - O_H_E 5y ago
      > i cannot not hear their pulse, breathing, and other biological proceedings Oh WOW, that is ...... wild and interesting
  - pvarangot 5y ago
    I had it as a kid, went away in my teen years and now I only have tinnitus. For me it was only on certain frequencies that were boosted and they are near the principal harmonic of my always-there ringing tinnitus or in the frequency of another noise I sometimes hear that's more like a pure sinewave. As far as I know it's more of a brain thing and not like a super-ear thing.

Bayart 5y ago

The precompiled AutoEQ/results/ tree has been a godsend, and so is Wavelet. I honestly would think about buying a pair of headphones that's not in it, unless I'm actively looking for a specific sound signature (but it's really all about price and comfort as far as I'm concerned).

rweichler 5y ago

If you have an iPhone X or older, you can jailbreak it and use EQE (https://eqe.fm https://eqe.fm) which is system-wide and has AutoEQ integration built-in

scblock 5y ago

I would caution against applying an EQ curve to headphones without review beforehand. If EQ includes large boosts (which the example image and table shows) that can easily introduce clipping distortion if applied linearly without appropriate headroom adjustments. Additionally, depending on transducer performance large bass boosts may simply increase transducer distortion in the bass frequency. Some headphones respond well to EQ, and others not so much. I am not opposed to EQ, but it should be applied judiciously, and at least partially by ear. I don't believe trying to exactly match a target curve will necessarily provide good results compared to a more judicious approach.

sanjiwatsuki 5y ago
AFAIK, every tool that integrates with these things apply a negative boost equal to the peak added to avoid clipping.
- dsr_ 5y ago
  And every recommendation on the site starts with the appropriate amount of negative gain. But if you miss it, somehow, then setting it to the next whole number of decibels larger than the largest gain will do.
scblock 5y ago
As an example of my point for using a judicious approach above, applying either of the sets of Sennheiser HD8XX measurements and resulting AutoEQ curves from the Crinacle and oratory1990 folders in the Roon DSP system collapses the headstage and tilts the perceived tone of the headphones from relatively full and engaging with a dip in the 2-3 kHz region to thin and hollow, with an overly bright, brittle top end. Also, these are two sets of measurements of theoretically the same headphones, each with 10 adjustment points, some quite broad, some very narrow. The overall shape of the resulting curve for the same target compensation is similar, but they have some significant differences in the specifics, and they sound different. We have to remember that measurement systems and individual measurement setup can vary quite a bit, so settings from this tool will bake all of that in as well. As a comparison, based on review of the measured curves in more of a a big picture way, applying a much simpler EQ with a broad 1.5 dB bass lift up to about 100 Hz, another broad lift of about 3 dB around 2 kHz, and a slight drop of about 1 dB centered around 8 kHz brings the bass and vocals up a little but keeps headstage and overall tone intact. Maybe it's this particular DSP implementation, but I would be wary about trying to use any of the precompiled results directly. EQ can have real benefits, but is going to be more personal than automated settings will capture.

natdempk 5y ago

It would be interesting to compare the suggested EQ curves here vs. commercial offerings that claim to introduce a neutral response like Sonarworks. In theory I think they should be doing quite similar things, but curious if there are any real differences or if they land in the same general area.

dsr_ 5y ago
This is exactly what Sonarworks is doing, but without a specific software package to implement it (and sell to you).

rokweom 5y ago

Autoeq is not just a collection of ready-made EQ settings for different headphones, it's also a database of measurements from few different review sites and a tool for creating custom EQ curves. Do you have a specific target that you like? You can create a custom curve tuned for that target. Do you want to know how a certain model of headphones sounds? You can make your current ones sound almost exactly like them. Really cool project.

londons_explore 5y ago

All this assumes the headphones are linear devices... which clearly isn't the case! None of these tools seem to do anything but rudimentary nonlinear correction... Even simple things like measuring and cancelling harmonics nobody seems to do...

khimaros 5y ago

would it make any sense to use this for tuning a car stereo? particularly, to compensate for engine or road noise?

dsr_ 5y ago
Not really. That's an immediate noise cancellation function, which only works (a) with headphones and (b) by recording the outside noise and adding its inverse to the sound being played.
- khimaros 5y ago
  i don't mean necessarily doing active NC. i was more curious about doing an initial tune with the car at idle or while driving on the road as a way to compensate for the average. is there any value in doing something like that? maybe even just to tune the EQ profile of the car.
  - TylerE 5y ago
    No, as the frequencies involved will vary based on speed, engine RPM, and what seat you're sitting in.
  - jerrysievert 5y ago
    some cars/car stereos have had auto eq'ing with an active microphone in the past, I have to assume they still do. my 2007 has active eq'ing to counteract road noise, and a different eq setting when the top is up, based on a switch. it's a Bose stereo.

tanvach 5y ago

I created and opened up the design of a standalone DSP [1] that leverages AutoEQ. The DSP is optimized for our home brewed headphones for VR, but I’ve been using it for my desktop headphones too. AutoEQ is not a perfect tool, but an amazing (and free) starting point to get the most out of your audio systems. [1] https://github.com/tanvach/prettygood_dsp https://github.com/tanvach/prettygood_dsp

starky 5y ago

Unfortunately on Windows is uses EqualizerAPO which has to be re-enabled every time Windows is updated or it stops working. I've given up on my headphone EQs because of it.