AI Autosound Tuning Assistant (Autosound Tuning Skill)

🇬🇧 English · 🇩🇪 Deutsch · 🇵🇱 Polski · 🇺🇦 Українська · ❓ FAQ · 📘 TCC quick guide (EN)

In simple terms: This is your personal AI car audio tuning master. You want a perfect soundstage and a smooth tonal balance, but graphs, phases, and delays seem too complicated? This assistant will take care of the hard parts. It reads your microphone measurements and guides you step-by-step to perfect sound.

• You measure — AI calculates: It works together with REW software, analyzes your cabin acoustics, and proposes exact settings for EQ, crossovers, and time alignment.
• Minimum time in the car: The main calculations are done at your desk at home. The first result takes two visits to the car: one to measure, and one to enter the ready-to-use numbers and check them. After that you improve the tune round by round, by ear and by measurement, as far as you want to take it.
• Writes nothing to your DSP — you load it: The assistant never touches your processor directly. It shows you numbers and graphs and prepares the EQ for import: on a Helix the whole Full EQ bank goes in through DSP PC-Tool in one step, and for processors without a file import the free REW-EQ-CopyPaste-Assistant pastes it. You decide what goes in.
• Not a regular chat: The project state and all settings are saved to files on your disk, so nothing is "forgotten" between sessions and you can always roll back a step.
• Two AIs, and your ear decides: one AI proposes settings, a second one checks them. The check is part of the method; only the automatic link between them is optional — without it, you paste the package into any AI chat by hand. The final judge is your ear: you listen and decide, not just approve.
• Works with facts: The AI doesn't guess settings. If the measurements are wrong or not enough, a check catches it and asks you to retake the measurement — or to go on knowingly, accepting the risk of error.

Proven in Competitions

With version 2.x of this method, the author's car took four awards in 2026 at EMMA and AYA championships (the first award was won before it was bundled into a skill, using AI hints from the same graphs, which inspired this project). Version 3.1, with a graphical interface, is the current release. The fifth award — 3rd place at the German EMMA Final 2026 — came with 3.x, by refining the existing tune rather than tuning from scratch. The 2.8.x line behind the first four is still available: see the FAQ, path 3.

Tuned With Us!

AYA Finale 2026 (17.10.2026), Einsteiger 5000: 1st place and Vereinsmeister.

Your system can sound like a champion too!

[!CAUTION] AI is an assistant, but the responsibility is yours. A manually entered number with a typo can burn a tweeter. Always check crossover frequencies before unmuting the sound, and always start at a low volume.

What You Need to Start

The app installs with a single command. For hardware and subscriptions, you'll need the following:

1. Measurement microphone (e.g., UMIK-1, or preferably an XLR microphone with a sound interface and physical loopback).
2. Processor (DSP) in your car.
3. REW (Room EQ Wizard) software — beta version is required (the current release build, V5.31.3 of July 2024, has no API at all — check Help → About before you start). Get the beta build from roomeqwizard.com/beta.html. After launching REW, go to Preferences → API, check Start the API when REW starts, and click Start server.
4. Paid Claude subscription (Pro or Max) — Claude does the heavy lifting; it is the supported path, and the app is built for it. Other AIs can drive the run through omp (installed only when you ask) — at your own risk.

(Recommended: a free GitHub account, to back your tuning history up to a private repository.)

How to Install and Start

By default the installer sets up: Claude Code, the tuning method with the Python libraries its tools need, the Autosound TCC desktop app, and the Gemini reviewer (agy). It takes 10–20 minutes the first time on a Mac without the developer tools, 5–15 on Windows without Git, a few minutes otherwise. Depending on your system, other installers may ask for permission along the way — that is normal (details in the FAQ).

macOS — open Terminal (⌘-Space, type "terminal", Enter) and paste:

curl -fsSL https://raw.githubusercontent.com/ayukhno/autosound-tuning-skill/v3.2.0/install.sh | bash

Windows — open PowerShell (Start, type "powershell", Enter) and paste:

irm https://raw.githubusercontent.com/ayukhno/autosound-tuning-skill/v3.2.0/install.ps1 | iex

(The version in the address pins the installer itself; it always installs the newest release.)

Options:

What it does	macOS	Windows
models other than Claude, through omp	--with-omp	-WithOmp
the project backup on GitHub (a private repository; the keys stay out of it)	--github	-GitHub
the method without the app (terminal only)	--terminal	-Terminal
show the plan and change nothing	--dry-run	-DryRun

With options — for example omp and the GitHub backup — on macOS:

curl -fsSL https://raw.githubusercontent.com/ayukhno/autosound-tuning-skill/v3.2.0/install.sh | bash -s -- --with-omp --github

and on Windows, as two lines:

$i = irm https://raw.githubusercontent.com/ayukhno/autosound-tuning-skill/v3.2.0/install.ps1
& ([scriptblock]::Create($i)) -WithOmp -GitHub

How it ended is in the installer's exit code: 0 ready · 1 stopped before the end — a refusal, an error or an interruption; what was done before it stays, and the last lines say what (a stop before the method's step leaves the method as it was) · 2 a wrong option to install.sh, or a bad -Channel or -Plugin to install.ps1 (PowerShell itself refuses an option it does not know, with 1) · 3 installed but not ready, its last lines naming what is missing and what to do.

Already in Claude Code? The method also installs as a plugin:

claude plugin marketplace add ayukhno/autosound-tuning-skill
claude plugin install autosound-tuning

A plugin brings the method's files only: in the first session type /autosound-tuning:setup — it checks the plugin against its signed release and installs the rest. The TCC desktop app is not included this way; add it with /autosound-tuning:setup app (or /autosound-tuning:install-tcc).

After installation:

1. The installer's last step signs you in: Claude, then the Gemini reviewer (agy), and GitHub if you asked for it.
2. Open the Autosound TCC app from your desktop.
3. Create an empty folder for your car (e.g., MyCarTuning) and select it in the app, with AI main: Claude Opus (SDK) and AI critic: Gemini Pro (High) — or Gemini Flash (High) if Pro is not offered to you.
4. Important: keep Claude Opus's effort at xhigh or higher (the default; max for hard steps).
5. Type in the app chat: "tune a new car from scratch". The AI will start asking questions and lead you by the hand.

Prefer to talk to the AI in a terminal? That works too, and many find it handier: keep TCC beside it in Control mode, where you see and control every parameter while the session works (TCC guide, FAQ).

▶ Target curves: the method comes with its own curve for competitions, SQ-Comp-Ref. The Target Curve Visualizer analyzes and compares curves side by side — SQ-Comp-Ref, your own from REW, or the standard ones from Nono Tuning Tool — and saves the one you choose. How to pick one: the target-curve guide.

What the Tuning Process Looks Like

1. Preparation at home: You tell the AI about your system (which speakers, which processor).
2. Measurements in the car (the first of two visits): with protective filters on the DSP, you record each driver on its own, in one session; the TCC app walks you through it step by step. The tune is then designed at the desk.
3. Math at the desk: You sit at your computer (without the car nearby). The AI analyzes measurements, joins the subwoofer to the midbass, evens out the soundstage, and calculates the EQ. The desk only predicts the results; the car then verifies them.
4. Back in the car (the second visit) — verify, correct, fine-tune: enter the numbers into the DSP, check them with a few measurements and by ear, and correct what is off; the second AI checks the result at the desk afterwards. Two visits give the first result; after it you improve the tune round by round. Fine tuning is best done in the car with the most capable model (today Claude Fable): you tell the AI what you hear, and the better you want it to sound, the more rounds and visits it takes. Working at the desk is possible too — the model and the method adapt.

Feedback, Support, and Privacy

Privacy: The skill learns from every tune and, only with your explicit consent, sends generalized lessons to a shared knowledge base. It never collects personal data and never sends full measurements.

Issues and bugs:

• If something is wrong with the tuning logic itself: Open an issue on GitHub (autosound-tuning-skill).
• If the issue is related to the GUI (Autosound TCC) — write to the TCC app repository.
• No GitHub account? Ask the session ("report a bug") or use TCC's report window: your report goes to the author through a Google Form instead — text only.

This tool is completely free. The code and scripts are licensed under MIT, and the documentation and method themselves under CC BY-SA 4.0.

Credits: part of the DSP math is ported from Resonalyze by DIMOSUS (MIT) — LICENSES/NOTICE.md.

If it saved you weeks of tuning time and you want to thank the author, you can do it here: 💜 GitHub Sponsors · ☕ Monobank Jar (UA)

Good sound!