TwinGear Cloud · Partner program

Your gear. Your capture. A twin you keep.

Anyone can apply · Free if released · No delivery dates

TwinGear Cloud is a partner program anyone can apply to. You record the hardware you own by following our capture guide, we train a parametric model of it, and you get the twin back as an encrypted .napx. It loads in Twin-Loader, our dedicated loader plugin (free for partners, in development), so the twin runs inside your DAW like any other plugin. If you allow us to sell the twin as a plugin, training is free. If you keep it private, training is paid, based on the amount of data. Our resources are limited, so we may choose among applicants.

How it works

Four steps. Two are yours, two are ours.

  1. 01You

    Apply

    Tell us about the gear, your recording setup, and why you want in. About ten minutes.

  2. 02noshigeSound

    We get in touch

    Souta reads every application and writes back personally when your gear and setup fit the program. We cannot reply to everyone.

  3. 03You

    Capture, sample first

    You record input/output pairs with our source audio, following the guide below: a short sample for us to check, then the full data, following the plan for your unit.

  4. 04noshigeSound

    Train, then download

    We train the twin on our side. Progress shows in your account, and the finished .napx with its license key lands there too.

Two ways to take part

Release it, or keep it private.

Training data grows fast: every extra knob multiplies the recordings. Four knobs at 11 steps each already make 14,641 combinations. So the first choice is what happens to the twin, and the capture follows from it.

Up to 11 steps per knob

Release it as a plugin

You allow us to sell the twin once it is ready. A twin for sale is a public product, so we cover the knobs and switches it needs, in up to 11 steps per knob.

Training is free. There is no revenue share at this stage. You keep the twin, with a free license key.

There is no fixed limit on knobs or switches. We do not capture every combination: for each unit, we work out which controls really affect each other and plan the capture with you.

Keep it private

Only you use the twin, and you can keep the capture small: for example one knob in up to 11 steps (the model fills in the positions between them), or a few settings you actually use (no in-between positions).

Training is paid, based on the amount of data. We quote the fee before you start capturing and charge it only when the twin is finished. If training does not reach a good enough result, there is no charge.

There is no fixed limit here either. Tell us in the form what you have in mind and we will quote it.

What a capture looks like.

One recording is one setting of your unit, played through with our source audio. How many recordings you need depends on what the twin should do. These examples give a sense of scale.

  • A few settings you actually use

    6 settings = 6 recordings

    Set the unit, record, then move to the next setting. The twin reproduces those settings and switches between them.

  • One knob

    11 steps = 11 recordings

    Move the knob through 0 to 10, one recording per step, with everything else fixed. The model fills in the positions between the steps.

  • Knobs that work on their own

    3 knobs × 11 steps = 33 recordings

    Sweep one knob at a time while the others stay fixed. This suits knobs that barely affect each other.

  • Knobs that affect each other

    2 knobs combined: 11 × 11 steps = 121 recordings

    Every pair of positions is recorded, so the count grows fast. Which knobs to combine, and how finely, is what we plan with you.

  • Switches

    1 knob (11 steps) × a Low / High switch = 22 recordings

    A two-position switch doubles the recordings it affects. How to handle switches depends on the unit.

With the full mono source set, one recording takes about an hour, so 11 recordings are about 11 hours of recording. The source length is part of the plan we make with you.

A full capture is a serious commitment. It can take many days and a lot of data, and a small mistake can cost all of that time. We tell you what your unit needs before you start. If you enjoy that process and look forward to the result, we would love to hear from you.

Who we are looking for

A small group of partners.

This program is niche by nature. You take on the data capture, the hardest part of AI training, so it takes deep knowledge of both analog audio and digital data, and a high-level setup. Even a tiny difference can make training fail. Anyone can apply, but our resources are limited, so we may choose among applicants. What tips the balance:

  • You share the idea: real hardware reproduced at the waveform level and run on a GPU inside the DAW — and, in time, large-scale twins built in the digital domain.
  • If you release the twin, you are willing to help spread the word once it exists — a video, a post, a word to the people you work with.
  • You own the gear, plus a high-quality audio interface and cables. The interface needs to record a loopback at the same time as the unit.

What we do not do

Promise a delivery date, or guarantee a reply. How long training takes to reach a good result depends on the gear. If you do not hear back, you are welcome to apply again later with different gear.

What you get

The twin, the gallery, the tools.

The twin itself

An encrypted .napx of your gear, bound to your account, with a free license key. Yours to keep.

Your gear in the gallery

If you want it, a photo of your gear and your name are credited in the gallery alongside the finished twin.

The tools, free

Twin-Capture (recording) and Twin-Loader (a plugin dedicated to loading trained twins) are in development. Partners receive both at no cost when they ship.

Capture guide

How to record a twin-ready capture.

Read this before applying — it is what the capture session will look like. Each rule comes from training real units; skip one and the quality of the trained twin drops.

Wiring diagram: the DAW sends one source signal to an audio interface; line out 1 loops straight back into line in 1 (input.wav), line out 2 passes through the hardware into line in 2 (target.wav); both files are recorded in one take.
  1. 1Source audio plays from the DAW (our file, about an hour).
  2. 2Line out 1 and 2 carry the same signal.
  3. 3Loopback: out 1 goes straight back into in 1. That recording is input.wav.
  4. 4Line in 1 and 2 are recorded at the same time.
  5. 5Your gear sits between out 2 and in 2. That recording is target.wav.
  6. 6Both files come from one take, per setting.
The model compares target.wav against the loopback, not against the original file. The loopback carries your converters and cables, so the twin learns only the gear.

The DAW session

Playback tracks send the source out; record tracks capture what comes back. Arm the record tracks together and run one take per setting. For stereo gear, every track becomes an L / R pair, fed from the stereo source set.

Mono gear
12345678910111213141516
Playback · plays the source audio· Track 1–2
1OUTPUT TO LOOP
source audio
2OUTPUT TO HARD
source audio
Record · what you send us· Track 3–4
3INPUT FROM LOOP
input_00.wav
4INPUT FROM HARD
target_00.wav
Stereo gear
12345678910111213141516
Playback · plays the source audio· Track 1–4
1OUTPUT TO LOOP L
source audio
2OUTPUT TO LOOP R
source audio
3OUTPUT TO HARD L
source audio
4OUTPUT TO HARD R
source audio
Record · what you send us· Track 5–8
5INPUT FROM LOOP L
input_00.wav · L
6INPUT FROM LOOP R
input_00.wav · R
7INPUT FROM HARD L
target_00.wav · L
8INPUT FROM HARD R
target_00.wav · R

Four stereo tracks work just as well — one stereo track per pair (OUTPUT TO LOOP, OUTPUT TO HARD, INPUT FROM LOOP, INPUT FROM HARD), each holding L and R as stereo data. Either way, use the stereo source set, not the mono one.

Naming the tracks like this keeps the files easy to tell apart later. Playback tracks carry the same source file; the record tracks become input_NN.wav and target_NN.wav, numbered per setting.

  1. 1

    What the recording should contain

    Everything the twin will meet in real use: vocals, acoustic and electronic sources, mix-bus material, noise — and a wide range of levels, from quiet to loud. A model trained at one level drifts from the real unit at every other level.

  2. 2

    Use our source audio

    We share a folder of source audio — all our own material, cleared for training. It holds two separate sets: a mono set (about an hour, 48 kHz) for mono captures, and a stereo set with different material on the left and right channels for stereo captures, so the twin can learn how the two channels interact. Use the set that matches the capture; never mix them. Your own material works too, if it is as varied.

  3. 3

    Record input and output as a pair, in one take

    Send the source out of two line outputs. One goes straight back into a line input of the same interface — that loopback is input.wav. The other goes through your gear into another line input — that is target.wav. Record both in a single take, as in the diagram above. Use the same cable model — ideally two identical cables — for the loopback and the gear path, so the character of a cable does not end up in the data.

  4. 4

    Fix the levels, then never touch them

    Decide the loudest input you will use, set it just below clipping, and leave every level alone for the whole session. If a level moves, the model cannot tell why the output changed. Set the level so that the range you would actually use through the plugin is captured in balance.

  5. 5

    Capture to your plan

    For a twin that will be released: the knobs and switches it needs, in up to 11 steps per knob, following the plan we make with you. For a private twin: as little as one knob in up to 11 steps (0 to 10; the model interpolates between them) with everything else fixed, or a few settings recorded as they are. A single setting is also fine. The twin then reproduces that setting only.

  6. 6

    Send a short sample first

    Before the full capture, record a few minutes of one setting and send it to us. We check the recording itself — levels, loopback alignment, format — and reply. Only then record the full data. We do not train on the sample; it is there to catch a setup problem before it costs you a whole session.

  7. 7

    Mono or stereo

    Mono and stereo are two separate captures and are never mixed. If the left and right channels interact (linked compressors, stereo-width circuits), record in stereo with the stereo source set, whose left and right carry different material — that is how the twin learns the link between the channels. If the channels are independent, a mono capture of the better channel with the mono set is enough; the twin is then identical on both sides.

  8. 8

    Format and files

    Record at one of 44.1, 48, 88.2 or 96 kHz, and send the WAV files at that same rate: 24-bit, no resampling. The twin is trained at that rate and works only at that rate. Number the files by setting so we can tell which is which: input_00.wav and target_00.wav for the first setting (knob at 0, or snapshot 1), input_01.wav and target_01.wav for the next, and so on up to _10. Add a short note saying what each number is. If you record more settings than this, we agree the file names in your plan. You send the files directly to us for now; we agree the route when we are in touch.

FAQ

Before you ask.

Application

Tell us about the gear.

No account needed to apply. If we go ahead, you will use a noshigeSound account to follow progress and download the twin.

The gear

Up to five units. One is plenty to start.

What should happen to the twin?
Recording setup
About you
Spreading the word

For a twin to be released, partners who can help people hear about it get priority. Be honest — “none” is a valid answer.

Where you are active
What you could do once the twin exists
Before you send

All three are required. They are the terms of the program.