Here's how we're recording mic samples for earbuds with bone conduction pickups

In the coming year, many of the earbuds you’ll see hit the shelves will come with a feature that promises to make your phone calls far clearer than they were before. While it’s true that the reason we offer microphone samples taken in controlled environments is to show how microphones in existing products fail, we found ourselves in a little bit of hot water because our old setup made it impossible to show how this new feature, bone conduction, could make microphones perform well.

So we upgraded our equipment.

The problem

In wireless earbuds and headphones, the microphones used for noise cancelation often have to double as an array to allow you to speak to your friends, family, and colleagues. Such tiny mics are an engineering triumph in their own right, and while microphones in earbuds have come a long way — you can’t squeeze blood from a stone.

Because of this, some companies like Bose, Sony, and Nothing have elected to augment new earbuds with either bone-conduction sensors or pickups to help earbuds achieve better-sounding results in challenging conditions like wind or a noisy room. Traditional mic arrays can struggle with this, and there’s only so much you can do to get rid of unwanted noise at the source. On paper, this is somewhat of a no-brainer; use someone’s actual voice itself to help the earbuds determine what’s noise and what’s signal? Easy decision, right?

Unfortunately for us at SoundGuys, this meant that our old setup of a calibrated speaker affixed to a stand was no longer sufficient. Consequently, the samples we took for models like the Nothing Ear (3) and Bose QuietComfort Ultra Earbuds (2nd Generation) weren’t as good as they might have been on a human. Because the earbuds couldn’t use vibrations sent through a meaty, bony head to function the way they’re supposed to, they would instead use their onboard mics alone to mixed results. Some mics even refused to pick up sound at all in the presence of noise. That’s a problem.

It didn’t take long for some emails to come in from frustrated manufacturers to with the contention that our mic samples couldn’t be representative of what most people would hear over a call — and we agree. But what I took issue with was the idea that it’s impossible to use a mouth simulator to record somewhat realistic samples with earbuds using bone conduction pickups.

Our solution

This puzzle was clear: how could we make something that transmits vibration approximately like a human head, so earbuds with bone conduction-augmented microphones could record our standardized samples? That’s tough. Head and torso simulators don’t really attempt to do this, and we also can’t use a real head for obvious reasons. As metal as it would be to have an actual skull as a test fixture, I don’t think everyone would appreciate me using one for… this.

But we had two advantages:

1. The ears for the test head we use are embedded with a bone-like material and a good-enough approximation of a cartilaginous ear.
2. We remember our Beethoven.

If we could vibrate those ears from the top, bottom, and back (think: the top of your mouth/bottom of your skull), we stood a good chance of getting enough force from the simulated mouth to the bone-conduction pickups, which could then be turned into a signal. As these ears are meant to be very close to a human’s in anatomy, we had the most complicated part of the problem dealt with already — as we had a used and worn-out spare set just collecting dust, so we recycled them for this project.

The only remaining challenge is to somehow get a rigid medium for the speaker to transmit vibrations to these ears. For that, we grabbed a BambuLabs P1S with AMS, specialty nozzles, and a whole boatload of filament. Through research, we discovered that nylon with embedded glass strands with 50% gyroid infill has similar hardness to human bone when printed, so we got to work fabricating candidate fixtures out of polylactic acid (PLA) to make sure we weren’t creating any additional mechanical problems. After the print, the fixture requires annealing and post-processing in order to make sure that warping and rigitidy loss doesn’t occur over time.

We then used the following earbuds to test the results of these test fixtures to see if we could get our recordings somewhere in the ballpark of “correct.”

1. Bose QuietComfort Ultra Earbuds (2nd Gen)
2. Nothing Ear (3)
3. Sony LinkBuds Clip

One of my favorite things in life is making friends with extremely talented people. And these people, like me, love puzzles. I’ve been getting advice from doctors, prosthetics experts (like my sister), and others over the past year for related projects. But it was current SoundGuy and skilled maker, Shiun Okada who was extremely generous with his time and expertise, that took the lead on modeling, printing, and assembling the fixtures.

This process took us two weeks, significantly reduced by the printer’s utterly insane speed. But after 39 design iterations, 14 test fixtures, and 140-someodd samples taken on both the fixtures and from real people, we achieved a working fixture. Though the PLA prototype actually worked well, we were ready to proceed with fabricating our “final” fixture.

Results

After making it to our final prototype, we had decent enough results to kick the tires on the above earbuds one last time before making the materials switch to glass-reinforced nylon (PA6-GF). Given that there’s a roughly 100% chance that we’ll be updating with new iterations to solve minor problems over the next month or so, we’ll revsit the design if our final gut-checks with in-studio recordings show that the fixture doesn’t transmit sound as well as our staff’s heads do. For now, we’ll roll with what we have.

Below is a quick before-and-after for two of the earbuds listed above. Take special note of how hard the Nothing Ear (3) struggles on the old fixture, versus how it sounds when we recorded the exact same samples using the new fixture.

Bose QuietComfort Ultra Earbuds (2nd Gen) wind noise

Old fixture:

New fixture:

Nothing Ear (3) wind noise

Old fixture:

New fixture:

Nothing Ear (3) street noise

Old fixture

New fixture

In the old Nothing Ear (3) samples, you could hear that once outside noise was introduced, the earbuds went searching for some sort of signal from the bone conduction pickup and got none, so it recorded a bunch of silence instead. Now that the fixture can carry the vibration from speaker to earbud, that’s no longer an issue.

One thing’s for sure: the new samples we take from here on out will be closer to what you experience in your calls. That said, this is only an approximation, not a completely objective representation of what a product will sound like when you use it because of other factors like connection quality, contact with your ear, and network tomfoolery. But for now, we can at least make sure we’re not being unfair to products that use the new bone conduction mics in our samples. We’ll go ahead and update the affected reviews with new samples, and a link to this story explaining the situation.