Three centimetres

There is a version of this product where the microphone lives on the temple. It is a much easier product to build. The temple is where the battery is, where the board is, where the button is; putting the microphone there means no wire crosses the hinge and no channel has to be moulded through the nose bridge. Every engineer who looks at the problem suggests it in the first ten minutes.

We put it under the left lens rim instead, and the entire reason is three centimetres.

The arithmetic

The bottom of the left rim sits roughly 6.5 centimetres from an average mouth. The front of the temple sits roughly 9.5. Sound falls off with distance, and that ratio works out to about 3.3 dB more signal at the rim.

Three decibels does not sound like an argument. Here is why it is one.

Ambient noise — a bus, wind, a café, a road — is far-field. It arrives at both candidate positions at essentially the same strength, because three centimetres is nothing compared to the distance to a bus. So moving the microphone closer to your mouth raises the signal without raising the noise. The 3.3 dB is not a louder recording. It is 3.3 dB of pure signal-to-noise ratio, which is the only number that matters when the recording is going to be handed to a speech-to-text model on a windy street.

Everyone spent the last three years improving the model. Nobody moved the microphone.

What it costs to do it this way

Honestly: a moulded channel through the nose bridge, two microphone wires joining there and entering the temple through one hinge, and a front-frame tool that has to hold an acoustic cavity and a sub-millimetre port to a tolerance TR90 nylon does not particularly want to hold. It is why the front frame gets a steel tool rather than an aluminium one — aluminium blurs exactly the detail we are paying for.

It also means the microphone faces straight into the wind when you run, so it needs a metal mesh and acoustic foam over the port and wind suppression in firmware, and it means sweat runs down your forehead and lands precisely on it, so the port geometry and the internal drainage exist because of where we chose to put it.

All of that is downstream of three centimetres. We took the harder frame because the alternative was a slightly worse transcript forever.

Then take them off

The geometry has one more move in it. Worn, the main mic is 4 to 7 centimetres from your lips — the range covers the difference between a small face and a large one. Take the glasses off, hold them up to your mouth, and the same microphone arrives at 2 to 3 centimetres.

Nothing switches. No mode, no button, no setting. You have simply made the distance smaller with your hand, which is what people already do with a phone, except this time the microphone was designed for the distance you moved it to. At 2 to 3 centimetres a whisper arrives loud and clean enough for a model to transcribe it, which is the whole of what we mean by a quiet-voice engine.

It is an unusual thing to design for — a gesture where the user takes the product off to use it better. But it is honest about what is happening. The distance is the feature. Anything that shortens the distance is an upgrade.