Guided tour

01 First, the object

This is a studio microphone — the kind of mic you find in front of a singer in a vocal booth, not the one on a gaming headset.

This one has a particularity: it was not bought, it was built. The plans, the parts, the tests — everything is public and can be repeated.

02 What's inside

A studio mic is four things:

• the grille, on the left — it protects the capsule and filters out gusts of air;
• the metal tube, at the bottom — the body;
• the rings and seals — the hardware that holds everything together;
• the long black board — the electronics, slid inside the tube.

This cylinder is not just a decorative shell. As long as the electronics sat in the open, a 50 Hz hum — the hum of household mains power — polluted every recording. It disappeared the day everything was enclosed in metal. The body acts as a Faraday cage: it shields the inside from the electrical fields present in any room.

03 The part that hears

At the top of the mic sits the capsule. Inside it is an extremely thin membrane, stretched like a drum skin, and coated in gold.

1. You speak; your voice makes the air vibrate.
2. The air makes the membrane vibrate, by a few millionths of a millimetre.
3. That tiny movement becomes an electrical signal.

For the third step to work, the membrane must be electrically charged at all times — a bit like a small built-in battery. That is the job of one of the two boards below.

The problem: the signal that comes out is tiny and very fragile. Plugged straight into a speaker, you would hear nothing. And it is so weak that the slightest interference drowns it. Hence everything that follows.

04 The two circuit boards

The first one amplifies. It takes the tiny signal from the membrane and makes it strong enough to travel down a cable to a computer without degrading. This is the delicate part of the project: it handles a signal so faint that the components must let almost no current through — hence resistors of one billion ohms, an unusual value even in electronics.

The second one produces a high voltage. The membrane needs to be charged: this board produces that charge from the power supplied by the audio interface, raising the voltage step by step.

Why two boards and not one?

• The second one generates a very fast buzz — 76,000 oscillations per second, inaudible but real. Keeping it physically away from the fragile circuit stops it from polluting the sound.
• If one board fails, or if you want to swap one, you only replace that one.

05 How it is all drawn

Nothing is made by chance. Before a single board is ordered, everything is drawn in design software, in two steps:

1. The schematic — which component connects to which. The logical plan, like the wiring plan of a house: you know the switch controls the lamp, without yet knowing where the wire runs.
2. The layout — where each part sits on the board, and which path each copper track takes. The architect's plan.

The two screenshots above show the second step: each coloured line is a copper track, each round spot a pad where a component will be soldered.

The software then checks a long list of errors on its own: two tracks touching, a track too thin for the current it carries, a forgotten component. Only once these checks pass do the files go to the manufacturer, who sends back the bare boards one to two weeks later.

06 Assembly

The boards arrive bare. The components then have to be soldered on one by one — some are less than two millimetres across — before connecting the two boards to each other, to the capsule, and to the socket at the back of the mic.

Allow four to six hours for a complete build, and a good magnifier.

To find your way, the repository provides an interactive bill of materials: preamp · hex. The list of components on one side, the board on the other — click a component to see where it goes. It is the tool that makes this assembly practical without printing a plan.

07 The test bench

This is what the work really looks like: boards lying flat, crocodile clips, an oscilloscope probe, a soldering iron, and the capsule lying in the middle of the mess.

This photo is here on purpose. An electronics project is not a clean model on a white background: it is two days of debugging to understand why a channel stays silent, a component burnt along the way, and wires added by hand to make up for a layout error.

08 Does it really work?

Yes, on one of its two channels. The first recording dates from 12 April 2026, after two days of debugging. The other channel has never produced sound: a component on it was destroyed during debugging. The design fault behind it is fixed in the current revision of the boards (P3), which only a P3 board can confirm — and none has been made yet.

But "it works" is not enough: you also need to know how well. So the mic was set side by side with a very common commercial mic, the Rode NT1-A (about 200 €), both listening to the same source at the same moment. Comparing the two cancels out the effect of the room and the loudspeaker: what remains is the difference between the mics.

• ✓Its output level is comparable. In the calibrated June 2026 measurement, its sensitivity came within 0.1 dB of the commercial mic — a negligible gap.
• ✓It sounds different. Warmer and fuller in the low mids, softer in the highs. The NT1-A is bright, slightly biting; this one is its opposite. On a sibilant voice, it is more forgiving.
• ✓It is clean. No spurious resonance, no smearing: when a sound stops, the mic stops with it. No clipping up to −1 dBFS.
• !It still picks up some mains hum. Its background noise sits in the bass, with the lines of household mains power (50, 100, 150 Hz). Its absolute noise level has not been measured reliably yet.
• !It distorts a little in the mids. Around 500 Hz, 1–2 % distortion, more than the reference mic. The cause is still to be found.

The two weak points, explained

They are listed here because they are real and identified — and because a project that only showed its successes would not deserve to be believed on the rest.

• The hum comes from grounding: how well the electronics are bonded to the metal body. A solid bond reduces the hum and also steadies the voltage that charges the membrane.
• The distortion. At first the bass looked badly distorted (several percent at 50 Hz). Most of that turned out to be mains hum caught by the measurement; with the hum excluded, it drops to about 0.24 %. What remains is a smaller residual in the mids, around 500 Hz, still to be located.

09 Hear the difference

Everything so far can be seen. Here is what matters most: what it sounds like.

The two excerpts below date from June 2026, that is, the mic in its current state: mounted in its metal body, and after the voltage that charges the membrane was stabilised. Each time, the Talas and a Rode NT1-A — a commercial studio mic — listened to the same source at the same moment, on two inputs of the same audio interface.

Music

The most telling test: real music, played in the room, picked up by both mics at once.

What to listen for: the NT1-A is sharper in the treble — cymbals and consonants further forward, more "present", sometimes harsher. The Talas is rounder, calmer, with fuller low mids. Two different characters, neither better in absolute terms.

Pink noise

Pink noise contains every frequency in equal measure: it is the most revealing signal for comparing two mics, because any difference you hear has to come from the mic and not from the source.

What to listen for: the NT1-A hisses more in the treble; it is "brighter", a little more aggressive. The Talas is duller, rounder, more muted. It is exactly the "warm" signature the curves show — except that here you can hear it.

10 The measurements, as pictures

Unfold the nine graphs — optional; this is where people get lost

How to read these curves, in two sentences. From left to right: the low sounds, then the mids, then the highs. When a curve goes up, that range is more present; when it goes down, it is more discreet.

The sound signature

Timing accuracy

These two check that the mic does not shift sounds in time and does not linger after a sharp click. Both are good.

Distortion

Distortion is what the mic adds to the original sound. On the left, the rise at the low end is mostly mains hum caught by the measurement (see stop 8); the real residual is around 500 Hz. On the right, we check that the mic does not saturate as the volume goes up — it holds.

Noise and dynamics

On the left, where the noise comes from: the lines of household mains power (50, 100, 150 Hz) and a noise that sits in the bass; the NT1-A, for its part, hisses more in the highs. On the right: the mic does not squash the gaps between loud and quiet sounds.

Everything on one page

11 The project's history

A year of work, from the first sketch to today.

• 21 March 2026First sources published: the project goes public.
• 1 April 2026First complete release — schematics, fabrication files, parts list, photos and assembly guide.
• 12 April 2026First successful recording, after two days of debugging. Only one of the two channels works: the other was damaged during testing. First comparison with an entry-level mic the same day.
• 29 April 2026The prototype is documented inside out — failures included — and a second version of the boards is drawn.
• 11 May 2026Second version of the power-supply board drawn. The mic is mounted in its metal body: the 50 Hz hum disappears.
• 31 May 2026Reference recording session: background noise, frequency sweeps, channel matching.
• 3–4 June 2026Measurement campaign against the Rode NT1-A. Key finding: the voltage charging the membrane was unstable, which skewed all earlier measurements. Once stabilised, the mic's sensitivity comes within 0.1 dB of the NT1-A.
• 28 June 2026Third version of the boards (P3): layout reworked to remove the wires added by hand.
• August 2026The documentation is checked line by line against the board files; the check is now automated.
• 2 September 2026Current P3 fabrication files published: two-layer boards, every automated check passes. Not fabricated yet.

The second version of the boards was drawn but never made; its fixes were carried into the third.

12 Why everything is public

The plans, the measurements, the parts list and the mistakes are published under an open licence. Anyone can build this mic, modify it, or draw on it, including to sell it — provided they publish their own modifications in return.

That point deserves emphasis: the repository also documents the failures — a diode burnt during debugging, two layout errors on the first version, a hum that only went away once the electronics were enclosed in metal, an unstable voltage that skewed weeks of measurements.

That is unusual. Most projects only show what worked. Here, someone rebuilding the mic knows in advance where they will run into trouble — and that is exactly what makes the project useful to others.

13 Where it stands, and what comes next

The third version of both boards carries the hand-added wires into the design, and every automated check passes on it. No P3 board has been made yet.

What remains, in order:

1. Fabricate and measure the P3 — both boards, both channels.
2. Measure several capsules — to know how much one capsule differs from the next, and publish the raw data.
3. Reference measurements under controlled conditions — frequency response, noise, distortion, maximum loudness, directivity.
4. Improve the measurement toolkit so that anyone can use it with a low-cost audio interface.
5. Complete the hardware release — open-hardware registration (OSHWA), a parts list sourced in Europe with alternatives, frozen fabrication files.
6. Rebuild from the documentation alone — two complete units built only from the published files, then measured.
7. Validation by others — one or two volunteers build it from the documentation; every point where they get stuck is written down and fixed.

This work is self-funded, which is how everything so far was done. An application to the NGI Zero Commons Fund (NLnet Foundation) was not selected in September 2026 — 44 of about 1 000 applications were funded that round. The list above is still the work that funding would pay for.

An idea for later, beyond this list: a mic whose way of hearing could be changed — picking up only from the front, or all around — by turning a knob.

Today this is a research project run by one person. Nothing is for sale.

14 Frequently asked questions

"Why not just buy a mic?"

Because the goal is not to own a mic, it is to understand how it works and be able to modify it. A commercial mic is a closed box: if it doesn't sound the way you want, there is nothing you can do about it.

"Does it sound better than a commercial mic?"

No — and that is not the point. It sounds different: warmer, softer. On some voices that is better, on others not. It also still picks up some mains hum, which comes from grounding and is being worked on.

"How long did it take?"

About a year, from the first schematic to today, working on it in stretches.

"Can I buy one?"

Not today: no series exists and nothing is for sale. talas.fr has a waiting list, with no commitment and no payment. In the meantime everything is published so you can build one yourself — allow about 78 € of parts, plus an adapter cable, and a day's work if you already know how to solder.

"What is the hardest part?"

Handling such a weak signal without polluting it. Everything else — soldering, ordering the boards, assembling — is within reach of a patient person.

15 Going further

And the workshop that made it

The Talas One is the first object of a workshop, not of a product range. The rule there is the same everywhere: publish the plans, publish the costs, and build nothing that people cannot take back control of.

Applied to a service rather than an object, it gives Veza — a platform where musicians share their tracks, work on them together and sell their work. Self-hosted, no resale of data, the commission stated up front.

talas.fr — the mic and the workshop · veza.fr — the platform

Short glossary