A New EQ Strategy (Theory) for good personalized sound quality

[nerdemoji]

Skip to the bolded text if you don’t care about the story.

Before I learned about audio science, I got the AirPod Max and thought that they were the pinnacle of sound quality (actually there was a recent study that showed they had preference very close to Harman 2018). When I learned some Audio Science, I listened to them with Oratory1990’s EQ settings and they sounded amazing. I learned about speakers, too and thought they would be a good choice to further improve my sound.

When I got speakers (KEF Q150 + SMSL AL200), I was excited and they sounded good but there was some mid bass note that sounded really loud and long. I thought “well I know that it is a room resonance, but I don’t know what frequency it is at and I don’t have a measurement mic”. I had EqualizerAPO installed and could go in and guess the frequency, but then a thought occurred, if it was only 1 note on music, then surely I could go into Tone Generator and find it. Sure enough, I found a huge spike at 130 hz and promptly removed it from using the equalizer and I went into tone generator until it sounded flat. I did the same thing with other smaller bumps in the region and the difference was remarkable.

Time passes, and I was reading a thread and someone mentioned that the LF of a speaker in a room is similar to the HF of a headphone because both are the right wavelength for reflections in their spaces. Because of the 90 degree HRTF being different between people, the same headphone will have peaks and dips that are ear-specific. So, I had an idea. Here was my method.

1: Start with the general EQ to Harman 2018 from Oratory1990, Resolve, Amir, etc. I used EqualizerAPO (Peace GUI).

2: Go into Tone generator
https://www.szynalski.com/tone-generator/

3: Listen to it (from 1khz it should sound like it is rising in volume.) For me peaks and dips started at 4khz. I heard a slight dip at 4khz and a slight bump up at 5khz. A huge dip at 6khz and a big spike at 7-8khz. Big dip at 10khz. Rise after that and roll off at 17khz.

4: Estimate EQ to flatten and test until it sounds smooth (dips might not be removed by EQ in my experience). I didn’t remove the 10khz dip and I didn’t touch anything after 10khz.

5: This part is important. I want to know if this works for more than just me. I had my brother generate a random number 1 or 2 and choose the EQ and play a song of my choice beforehand in the other room. I was able to guess it and found preference in the new one.

If you guys have the time and resources, I would like to know what you guys think of this. If not, is this bad or against science from a conceptual perspective? I heard EQ is “blunt” for headphones at High Frequency. Is this because of the nature of the treble in headphones or content in music or because reviewers didn’t want people to EQ out dips and peaks in their headphones from measurements because the on-head sound would be different in HF for different people?

Lots of questions…

 

[IAtaman]

I like experimentation. Always fun and you learn something new.

One question, would it be not more reliable to do this via something like Earful instead of a tone generator?

 

[abdo123]

I think pink noise is more effective than playing individual tones.

 

[Keith_W]

Your method has some validity. It seems conceptually similar to what David Griesinger recommends - take a look at his website and read about "DGSonicFocus". The idea is that headphones should be equalized to your HRTF. He initially measured his HRTF using invasive probe microphones shoved into the ear canal, before he realized that he could obtain the same information by using your own eardrums as a microphone by playing test tones. He has an app you can download and try.

Good on you for coming up with this idea

 

[Blumlein 88]

Doesn't this have an issue that our hearing is not linear with frequency? Fletcher-Munson curves of equal loudness issues. Earful mentioned above takes this into account.

 

[Scgorg]

Human ears are not spectrum analyzers, and our ability to interpret frequency response based on swept sine waves is poor. Wide-band signals like pink noise are significantly better for this purpose, as abdo123 mentioned above.

Additionally, hearing sharp dips or peaks in the frequency response when using swept sines does not necessarily mean anything is wrong. If you have a sharp dip at 8kHz due to your particular HRTF at some arbitrary angle, this is a feature of your hearing. You are used to it, and in daily life it helps you localize sound. When you're listening to anything, there will be a ton of dips and peaks due to your HRTF, yet we never see anyone trying to compensate for their HRTF with loudspeakers for example (e.g. using in-ear mics and determining that a sharp dip or peak at 8kHz needs to be equalized).

What is interesting, however, is the difference between your HRTF's interaction with a good loudspeaker in a room and a given headphone. This is what the above Griesinger "DGSonicFocus" is (partially) about. More advanced examples include the smyth realiser, which uses a convolution filter to match a headphone's impulse response to a speaker's binaural in-room impulse response (BRIR), as measured with in-ear microphones. With head-tracking, this in principle gives you a perfect replication of a a given loudspeaker setup in a room through headphones (barring the physicality that comes with sound from a loudspeaker). In reality it likely won't be quite perfect, but most people who have used the smyth confirm that it is very close indeed.

Your idea is not bad, I think it's a fairly common thing to think of (among headphone nerds), but the rationale behind it is unfortunately flawed. With that being said, I can very well believe that you did in fact improve your listening experience by doing this, it might just not be the optimal way.

 

[thewas]

Human ears are not spectrum analyzers, and our ability to interpret frequency response based on swept sine waves is poor. Wide-band signals like pink noise are significantly better for this purpose, as abdo123 mentioned above.

Additionally, hearing sharp dips or peaks in the frequency response when using swept sines does not necessarily mean anything is wrong. If you have a sharp dip at 8kHz due to your particular HRTF at some arbitrary angle, this is a feature of your hearing. You are used to it, and in daily life it helps you localize sound. When you're listening to anything, there will be a ton of dips and peaks due to your HRTF, yet we never see anyone trying to compensate for their HRTF with loudspeakers for example (e.g. using in-ear mics and determining that a sharp dip or peak at 8kHz needs to be equalized).

Mainly due to these 2 reasons this method doesn't work well with sweeps for headphones, I had tried it several times a decade ago and it never gave audibly good results, the method of Griesinger with wide band noise comparison works quite better in my experience, here is the old "DIY" version

and the newer one with an app