Neumann Control

A portable app for Windows and Linux to find, read and control Neumann KH DSP speakers over the network, using the Sennheiser Sound Control protocol (SSC). It shows your speakers side by side like channel strips on a desk, with level faders, meters, groups and a master.

Alpha. This is an early version: expect rough edges and changes between versions, and keep an eye on your speakers' settings while you use it.

This is an independent project. It is not affiliated with or endorsed by Neumann or Sennheiser.

The desk with four speakers (two KH 120 II, two KH 750) and the master strip.

• Highlights
• Status
• Requirements
• Installation
• Getting started
• Using the app
• How writes stay safe
• Supported models
• Files and privacy
• Known limits
• Building from source
• Licence

• Find speakers on the local network (mDNS) and choose which to add, or add one by its IPv6 address. Added speakers are remembered.
• A desk of channel strips: status LEDs, the input and the standby countdown, a level fader, input meters (on the KH 750 also its output meters), mute and identify; what a strip shows can be changed per model or per speaker.
• A device panel per speaker with every setting the catalog knows for its model.
• Groups and a master: move or mute several speakers at once, and link settings within a group.
• EQ in a window of its own: a computed response graph and a table of the bands.
• History and undo: the writes of the session, each one revertible, Ctrl+Z for the newest.
• Keyboard shortcuts for the master, mute all and dim, also from other programs if you like.
• Presets of chosen settings and of EQs, and a comparison of EQ curves between speakers.
• Windows volume of a playback device as a strip on the desk (Windows only).
• Compact mode (Ctrl+K): one row per speaker or group in a small window.
• German and English interface, light and dark theme.
• Portable: one executable, settings and log next to it, no installer.

The interface follows the system's display language: German on a German Windows, English on any other (on Linux: German when the locale is German, from LANGUAGE, LC_ALL, LC_MESSAGES or LANG, whichever is set first). Settings → Language (in German: Einstellungen → Sprache) can fix one of them. This README uses the English names.

Alpha, below version 1.0 (? > About Neumann Control shows which one). The protocol layer, the polling and the interface are covered by automated tests against a built-in speaker simulator for the KH 120 II and the KH 750. On real KH 120 II and KH 750 speakers the app was checked for scanning, reading, metering and writing level, mute, identify and standby time, and for groups, the compact mode and the English interface. The Linux version was checked for scanning and reading real speakers, for the update check and on a KDE Plasma desktop under Wayland (Debian 13, in a virtual machine without graphics passthrough); not yet under X11 or GNOME.

• Windows 10 or 11 with a DirectX 12 capable graphics adapter, or
• Linux on x86_64 with glibc 2.39 or newer (Ubuntu 24.04, Debian 13 or later), a Wayland or X11 desktop and Vulkan or OpenGL graphics drivers. Under X11 the library libxkbcommon-x11 must be installed. The file dialog of the EQ import and export needs the desktop's XDG portal (or zenity), Check for updates needs curl, and the manual and links open with xdg-open.
• The computer and the speakers on the same layer 2 network segment (same switch or VLAN). The speakers are reached over IPv6 link-local addresses, which are not routed.
• IPv6 enabled on the network adapter in use.
• To scan, the firewall must allow mDNS (UDP 5353) for the app. Adding a speaker by address needs only the speaker's TCP port 45.

The app needs no installer. A release is a ZIP for Windows and a .tar.gz for Linux; each holds one folder with:

Put the folder anywhere you can write to, for example in your documents, and start neumann-control.exe. On Linux, unpack it with tar -xzf neumann-control-<version>-linux-x86_64.tar.gz and start ./neumann-control in the folder. The app keeps its settings and log in that folder (see Files and privacy). To update, replace the five files above with those of the new release; settings and log stay. To remove the app, delete the folder.

SHA256SUMS on the release page lists the checksums of both downloads; on Linux sha256sum -c --ignore-missing SHA256SUMS checks the one next to it.

One app runs per folder: starting it again while it runs brings the running app's window to the front. It is found by its title, so with copies from other folders running too, one of theirs may come forward instead. On Linux the second start just ends, as Wayland lets no program bring another's window forward.

1. Start the app (neumann-control.exe, on Linux ./neumann-control) and press Scan. The scan takes 4 seconds. The dialog "Speakers found" lists every speaker found with its name, model and masked address. Check the ones you want and press Add; each gets a strip. Speakers that already have one are marked "already added". Scan again looks once more, for a speaker that was still starting up; the list and your checks stay, new speakers are added below. Cancel adds nothing.

2. If the scan finds nothing (for example because mDNS is blocked), press + Address and enter the speaker's IPv6 address. A link-local address (fe80::...) needs the index of the network adapter after a %, for example fe80::1%7. Find the index in PowerShell:

   Get-NetAdapter | Select-Object Name, ifIndex

   or with netsh interface ipv6 show interfaces (column Idx). On Linux, ip link shows the index before each adapter's name.

Added speakers are remembered and reconnected at every start. When a link-local address cannot be reached through its stored adapter index (Windows numbers its adapters anew after a driver or adapter change), the app tries the same address through the computer's other adapters and keeps the index the speaker answers on; this needs no scan. When a scan finds a remembered speaker at a different address (link-local addresses and interface indexes can change), the stored address is updated. The app scans for that by itself, without a dialog: once at the start, and when a remembered speaker stays unreachable for 30 seconds, then at most every 5 minutes while one is missing. Speakers are matched by their serial; one added by address that never connected has none yet and no scan can follow it. IPv4 addresses are not accepted: the speakers are reachable only over IPv6.

Each speaker has a channel strip:

• its name (click it to open the device panel) and a grip to drag the strip elsewhere,
• four LEDs: connection, clip, device warning, and LOCAL (the rear CONTROL switch locks writing),
• by default the Input interface (a choice), the Current input and the standby Countdown; set-up values such as the delay, the standby time and the logo brightness are in the device panel,
• the output level fader (0–120 dB SPL in 0.5 dB steps) with its value below,
• input meters for channels A and B with clip lamps; the KH 750 also shows its output meters A, B and Sub, captioned "Out". A thin line marks each meter's peak of the last 1.5 seconds, and the number under the channel shows it in whole dB ("–" below the meter range),
• the response time, and by default Mute, Ident. (identify) and EQ, which opens the speaker's EQ window once the speaker has reported its model; on the KH 120 II also Auto standby, on the KH 750 Mute A and Mute B for its outputs A and B and its Temperature. Above the fader every model shows the same, so the faders of a mixed desk stand level,
• Details at the foot, which opens the device panel.

What a strip shows besides its name, LEDs, fader, meters and message is set per model, for every speaker of that model at once. With Same layout for every speaker of a model off in Settings, each speaker keeps its own strip and compact buttons instead: they start as its model's, and Default takes the speaker back to its model's layout. Switched on again, every speaker shows its model's layout; the own ones are kept for the next time.

• Right-click a knob or button on a strip to Hide it, Swap for… another setting, or move it up or down (past the fader too). Right-click a row in the device panel to put it on the strip (Add to strip) or take it off.
• The pen in the toolbar (Customize strips) switches every strip into edit mode: + Add adds an element, dragging one with the left button moves it (also across the fader; while you drag, the strip already shows it where it would land, faded, and the others make room; it goes to strips of the same model only, with own layouts to its own strip only; the note under the strip says which), right-click hides it, and Default brings back the model's default strip. The controls write nothing while you edit; the pen again (Done) ends it.
• Any number, switch, choice or read-only value of the device panel can go on a strip; the device name and lists such as the input mix stay in the panel.

On the KH 120 II the input meters are captioned with the rear input switch (Analog, AES3 L, AES3 R, Mono), and a dot marks the channels the speaker plays. That channel A is the left AES3 channel is assumed, not measured.

At the first start, and whenever no window size is saved for the mode (desk or compact), the window is made as large as the desk or the compact rows need and grows with them (a speaker added, a strip that widens once its model is known) until you size it yourself; from then on the app keeps the size you chose.

The toolbar shows the number of speakers (not in the narrowest window) and the average response time of the connected ones.

• Fader: grab it anywhere and drag; it moves with the pointer and never jumps to it. Shift makes it ten times finer, and the mouse wheel moves 0.5 dB per notch. A double-click on the value below it lets you type one. While you drag, the value stands beside the fader.
• Knobs: turn them by moving the pointer around the knob (clockwise is more; one and a half turns cover the whole range), with Shift for fine steps, or with the mouse wheel. A double-click lets you type a value (the delay in ms, or in meters, see below). While a knob moves, its value stands beside it.
• Values are written once: when you let go (only if the value changed), when the wheel stops, or when you press Enter. Escape cancels a drag or a typed value.

Click a speaker's name or Details under its strip. The panel opens in a window of its own, one per speaker, where the last panel was left (the first time over the middle of the main window, above the taskbar; a further open panel a little offset), which you can move anywhere, also beside the main window or onto another screen; opening it again brings it to the front. Its header shows name, model, firmware and the masked serial number; below it, tabs sort every setting the catalog knows for that model:

• Overview: strip colour, the device name to change, presets, EQ and Remove,
• Group: which group the speaker is in; while it is in one, that group's settings (as in the groups window, without deleting), Apply to group and Manage groups…,
• Outputs: one column per output, side by side (on the KH 750 its own woofer, then A and B),
• Inputs: the input settings and what the rear panel reports,
• Standby: standby time, level and the countdown,
• Device: the name (changed on Overview), identify, the device's own facts, the temperature trend, All values… and Full scan.

A tab without anything for that model is left out. The panel opens on Overview; another tab stays while it is open. In each topic what can be set comes before what is only shown. Settings the model may write get a control; the rest are shown greyed. Escape or the window's close button closes it.

Each output also has Input mix, one field per input (Input A, Input B, −100 to 0 dB): how much of each input that output plays. By default the KH 750 plays input A on output A, input B on output B and both on its own woofer. Mute persist is the speaker's own setting of that name; what it does has not been measured.

All values… opens a window of its own (Escape closes it) that lists every value the speaker reports in its own schema, with a path filter and copy buttons, including paths the catalog does not know yet. Full scan reads the same and writes it next to the executable as scan-<model>-<firmware>-<date>-<time>.json and .txt, with what the app knows about each path; serial number, MAC address, the speaker's name and any network address are left out. It is meant for collecting what other models report.

Save as preset… keeps chosen settings of the speaker under a name (all writable settings are offered, the name unchecked). Load preset… compares each setting the preset would change, as it is Now and as it would be written, and writes it as one history entry; settings this model may not write are listed as skipped. Both dialogs open in a window of their own in front of the panel. A preset belongs to its model: saving under a name this model already has replaces that preset, and the dialog says so (Replace). Delete in the load dialog asks before the preset goes.

Trend (2 h) on the Device tab shows what the app read in this session, at most the last two hours, on models with a temperature (KH 750): the temperature as a line, with its highest and lowest value at the left and the current one at its end in °C, and the warnings as a bar, red while a warning was active. Hover over them for the value and how long ago it was read. Gaps mark the times the speaker could not be read.

To remove a speaker, press Remove in its panel and confirm. The speaker is no longer remembered, and it also leaves its group and its place in the order; its strip colour and its own layouts are forgotten.

• Groups in the toolbar opens the groups window: the groups on the left, + Group below them, and the chosen group's settings on the right: name, strip colour, Show group strip, what it links (Linked), the speakers it has, and Delete group, which asks first. A change counts at once; an emptied name keeps the one the group had. A speaker also joins a group on the Group tab of its device panel. A speaker is in one group at most: checking one of another group moves it. Members are framed in the group's colour and followed by a group strip.
• The group fader moves every member by the same amount, each within its own range. With the loudest member at its maximum, the others still follow. Mute on the group strip mutes them all.
• Linked (on the group strip, in the groups window or on the panel's Group tab) links level, mute, auto standby or input interface: a change on one member goes to the others whose model may write it. Members that are not connected are left out. Apply to group in the panel copies a linked value to the other members.
• The group's name or Details at the foot of the group strip open the groups window on that group; right-click the group strip for Delete group…, which asks there first.
• Show group strip (on by default) takes the group strip off the desk; the members stay grouped and linked, and in the compact mode each gets its own row instead of the group's. Right-click a member's name for Edit group "…"…, which opens the groups window on its group, also while the group strip is off.
• Strip colour in the group's settings keeps the group's colour (Default), takes it off (None) or sets one of your own (Own). In a speaker's panel it follows the group (As the group), takes it off or sets the speaker's own; a strip without colour has no coloured frame.
• The master strip at the end (shown with Show master strip in Settings) moves and mutes all connected speakers the same way.

Drag a strip by the dots before its name to reorder: the other strips of its group (or the ungrouped ones) make room where it would land, and the order is remembered when you let go. Escape cancels.

Compact in the toolbar (or Ctrl+K) switches to a compact window and back. It has one row per ungrouped speaker and per group, each with LEDs (a group: its colour), the name, a lying level fader with its value and the loudest input as a small meter, and a master row at the bottom (with Show master strip). A speaker's row is framed in its own strip colour, if it has one, and has, by default, Mute, EQ and Details (the device panel); a group or the master row has Mute. A group row moves its members like the group strip; right-click it for Delete group…. A ! stands for a speaker's message (on a group or the master row, a member's), which shows on hover.

Right-click a button of a speaker's row to hide it, move it left or right, add another (a switch such as Ident., Mute A/Mute B or auto standby, or EQ and Details again) or go back to Default; right-click elsewhere in the row to add one when none is left. The buttons are chosen per model (or per speaker), like the strips.

The compact window keeps its own size and position and scrolls sideways when it is narrower than its rows. Scanning and adding speakers or groups stay on the desk.

EQ on a speaker's strip, its compact row or in its device panel opens its EQ in a window of its own (one per speaker; it can be moved to another screen). On the KH 750 choose the output first (Subwoofer for its own woofer, Output A and Output B for its two output sockets); then each EQ of that output has a tab: "EQ 2", "EQ 3", and on the KH 750 the crossovers of input A and input B ("Crossover A", "Crossover B").

• The graph shows every EQ of the output as its own curve and their sum as a thicker line, from 10 Hz to 24 kHz. The KH 750's crossovers lie one per input, so it shows a sum per input (input B dashed). The curves are computed from the settings, not measured, as the speakers' digital filters at 48 kHz (checked against measurements of every filter type). Each band has two levels: Level lifts or lowers every frequency while the band is on, Boost is the height of its filter (the peak or shelf); at the band's frequency both add up. For a shelving filter the speakers' Q is its slope (1 is the steepest without overshoot), as measured.
• Each band of the chosen EQ that is on is a point in its band's colour with its number inside; bands that are off have none and are switched on in the table. Drag a point to change frequency and boost, turn the mouse wheel over it to change Q. Level is changed in the table only, so a drag never moves the whole spectrum.
• A point stands at its band's own height; the curves add every band up, so a point can lie above or below them. Each band's own curve is drawn thin in its colour, the chosen band's stronger with its area to 0 dB.
• The table below lists every band with on/off, type, frequency, Q, level and boost; bands that do not fit below each other continue in a column to the right while the window is wide enough, else the table scrolls. Numbers take a comma or a point, and a value outside the allowed range is refused with the reason.
• A change is written when the gesture ends, and only for that one band, so the other bands stay as they are. Every change appears in the history and can be undone.
• Make flat sets level and boost of every band of the chosen EQ to 0 dB, after a question; frequency, Q, type and on/off stay. It is one history entry. On the crossovers it is offered only with Protect crossovers off (see below): flattening them removes the subwoofer's low pass.
• Reset switches every band of the chosen EQ off and sets it back to PARAMETRIC, 100 Hz, Q 0.7, level and boost 0 dB, after a question. The bands are switched off first; it is one history entry.
• Undo in the window takes back the speaker's newest EQ change, also with Ctrl+Z while the window is in front; the line beside it says which change that is. When the values changed since, the window asks first.
• Compare with lists the other speakers with the same output; a checked one shows the sum of its EQs there as a dashed line.
• Save as EQ preset… keeps all bands of the chosen EQ under a name; Load EQ preset… offers the presets with the same number of bands, compares each band's value that changes, now and as written, and writes them as one history entry. Writing several settings of a band one after another can pass a short level jump. A name already used by a preset of the same band count is replaced after the dialog says so; deleting asks first.
• Import… reads the filter lines of an Equalizer APO config.txt or of a filter file Room EQ Wizard exported for its "Generic" equaliser (Filter 1: ON PK Fc 100 Hz Gain -3.0 dB Q 1.41). A filter with a number goes to that band, one without to the next band; bands the file does not name are switched off; empty slots (None) are passed over. The file's Gain becomes the band's boost. Taken over: peaking (PK, PEQ, Modal), low and high pass with Q (LP/LPQ/LP Q, HP/HPQ/HP Q), first-order low and high pass (LP1, HP1), notch (NO), all pass (AP, REW's All pass) and shelving filters (LS, LSC, LS Q, and the same with HS); BW Oct instead of Q is converted. A shelf's Q in these files is the usual (cookbook) Q, the speakers' shelf Q is its slope, so it is converted; a plain LS/HS is taken as REW defines it (slope 1), and so is a plain LP or HP without Q (Q 0.707, both read from what REW stores). Refused: shelves by slope in dB or by corner frequency (LSC 10.8 dB, LS 6dB), band pass (BP: the speakers' band pass peaks at its Q, APO's at 0 dB), other types, any other missing Q, a Q or bandwidth that gives no Q above 0, filters for more than one channel, output device or processing stage or under a condition (Channel:, Device: or Stage: twice or after filters, If:), more filters than the EQ has bands and values the speaker would refuse: then nothing is written. Preamp and other APO commands are not taken over; the preview names them. The preview shows which bands change before Write sends them as one history entry.
• Export… saves the chosen EQ as Equalizer APO filter lines, numbered by band; shelves as LSC/HSC with their Q converted back. Bands that are off, a band's flat level, the 6 dB filters, band passes and inversion have no APO line; they are written as comments and listed after saving. Room EQ Wizard cannot load text filter files.
• With Protect crossovers on in Settings (off by default), the crossovers get no Make flat, no Reset, no EQ presets and no Import…. Without it they do, with a warning that the output then plays unfiltered.

In the main window:

While a text or number field is being edited, it keeps its keys. On Windows, with Shortcuts from other programs too in Settings (off by default), Ctrl+Shift+Alt with the same keys (↑, ↓, M, D) works while another program is in front; a key another program already holds is left out (the log says which). Shift is part of it because Windows takes AltGr for Ctrl+Alt: AltGr+M (µ) would otherwise mute every speaker. Linux has no shortcuts from other programs: Wayland gives no program the keys pressed in another one.

For screen readers each strip and each compact row is a group named after its speaker or group ("Strip: KH750 Links"), so its controls are told from the same ones on the other strips; the field a knob opens and the type choice of each EQ band carry names too.

History in the toolbar lists the settings the app wrote in this session (the newest 500), newest first, with the time, the speaker, the setting and the value before and after. A group or master move is one entry. Undo writes the values before again; Ctrl+Z does the same for the newest entry (not while you type in a value field; in an EQ window, for that speaker's newest EQ change). An undo is itself an entry, so it can be undone too.

When a value changed since it was written (in another app or on the speaker), the app lists the changes and asks before writing the old value back. A speaker that is not connected is left out, and its change can be undone later. The history starts empty at every start; identify is not recorded.

Settings opens a window of its own (Escape or the window's close button close it). It sets:

• the theme: like Windows (on Linux: like the system), light or dark,
• the language: like Windows (on Linux: like the system), German or English. It changes at once. German shows a decimal comma; typed values take a comma or a point.
• the lowest level the meters show: −120 to −30 dB in 10 dB steps,
• whether delays are shown and typed in milliseconds or in meters, as the distance sound travels in that time (343 m/s, at 20 °C); the speakers still get samples,
• whether response times are shown,
• whether the master strip (and its row in the compact mode) is shown (off by default),
• whether every speaker of a model has the same strip and compact layout (on by default; see The desk),
• on Windows, whether the keyboard shortcuts work from other programs too, with Ctrl+Shift+Alt (off by default; see Keyboard shortcuts),
• whether the log is detailed (Detailed log, off by default): neumann-control.log then also holds every write with its path and value (texts such as names left out), every failed request, where windows open and every ten seconds how many frames the window drew, how long they took, how much CPU the app used and what asked for the frames most, for troubleshooting,
• whether the EQ crossovers are protected (off by default),
• the EQ display: own curves of all bands, or only of the chosen band and the one under the pointer; the colour of each band number and of each EQ curve (Default colours resets them),
• on Windows, whether the desk shows a strip with the Windows volume, and of which playback device (Windows' default device or a chosen one): a fader in percent, mute and a peak meter per channel, named after its speaker position (L, R, C, LFE, BL, BR, SL, SR …) as the device's format gives it. It sets Windows' own volume, like the volume slider in the taskbar.

? in the toolbar opens the Manual (the file next to the executable, in your browser), the Wiki with the SSC protocol reference and the speakers' measured behaviour, and About Neumann Control: version, licence and a link to the project page. Check for updates there asks GitHub once for the newest release, pre-releases included, and, if it is newer, links to its page; the app does not download or install anything itself.

The app shows the control mode (NETWORK or LOCAL), which the CONTROL switch on the back of the speaker sets; it cannot be changed over the network. In LOCAL the speaker refuses every write, and the strip says so.

A speaker that stops answering is shown as unreachable and tried again 10 seconds after each attempt; an attempt can take a few seconds more for each network adapter of the computer.

When the app writes to a speaker:

• A value is written at the end of a gesture, never while you drag.
• A click on a fader without moving it writes nothing.
• A typed value outside the allowed range is refused with the reason; it is never cut to the range's end.
• The mouse wheel belongs to the control under the pointer where you started turning it. Turning it over a gap scrolls the panel; a knob that scrolls under the pointer is not changed.
• A wheel step on a control that has meanwhile disappeared (panel closed, mode switched) is dropped.
• A value the speaker reports outside its range is shown as it is and never written back.
• Models whose behaviour is unknown are read-only (see Supported models).

The catalog is crates/core/src/paths.toml. A product that is not listed there for a path is read-only, so models whose behaviour is unknown never receive a write. Settings that only exist on some models (subwoofer outputs, logo brightness) appear only for those models.

To bring another model to writing, its reported values are needed: on the Device tab of its device panel press Full scan and attach both files to a new-model issue. They hold no serial number, MAC address, name or network address.

The app talks to the speakers on the local network. It goes to the internet only when you click Check for updates: then it asks api.github.com for the list of this project's releases, with the app's name and version as its user agent.

The app is portable and writes only next to the executable, apart from an EQ Export… to the place you choose:

Speaker addresses and serial numbers are masked in the log and in the interface. An unreadable or malformed settings.json is renamed to settings.json.bad (.bad.1, .bad.2 … when that name is taken) and replaced by defaults. If only some values in it are not understood, or it holds values of a newer version (for example after going back to an older one), those values fall back to their defaults, the rest is kept, and a copy of the file is kept as settings.json.bad. A settings file that can be neither read nor moved aside (held open by another program) is not written over: the app says so and saves nothing in that run. The toolbar tells you whenever one of these happened at the start.

• SSC has no authentication. Anyone on the network segment can read and change the settings of every speaker; use the app only on a network you trust.
• A speaker in standby does not answer on the network, so it looks the same as a speaker that is switched off ("unreachable (off or standby)").
• A speaker accepts only a few TCP connections at a time (about four). With more clients connected (vendor app, home automation), it can look unreachable as well.
• On the KH 750, auto standby is shown but not written: it follows the AUTO STANDBY switch on the back, and the speaker refuses writes to it.
• What Mute persist does is not known yet; the app only shows and writes the speaker's own setting.
• Models other than the KH 120 II and the KH 750 are read-only. If you have one, please open an issue with its Full scan (see Supported models).
• Only IPv6 link-local addressing is supported; IPv4 is not.
• On Linux there is no strip with the system volume and no shortcuts from other programs, and only x86_64 is built. Under Wayland the app learns no window positions: it keeps the windows' sizes, and the desktop decides where they open. macOS is not supported.

Needs Rust 1.95 or newer, on Windows also the Windows SDK (its resource compiler embeds the app icon).

cargo build --release --locked -p neumann-control-app

The executable is target\release\neumann-control.exe (on Linux target/release/neumann-control).

check.ps1 runs the quality gates: guards against format widths taken at run time and stray carriage returns, a test of release-notes.ps1, format, licence check, cargo audit, clippy and the tests. It needs cargo-audit and cargo-about. build-release.ps1 runs them on a clean working tree, raises the patch version in Cargo.toml and commits that alone (-KeepVersion builds the version as it is), and writes release\neumann-control-<version>\ (on Linux release/neumann-control-<version>-linux-x86_64/; or -OutputRoot <folder>) with the executable, the manual, LICENSE, THIRD-PARTY-LICENSES.html and SHA256SUMS. On Linux it stops when the executable needs a glibc newer than 2.39, as one built on a newer system does. The manual is this README as one HTML file with its pictures embedded, made by crates/manual:

cargo run -p neumann-control-manual -- README.md neumann-control-manual.html

A release is a tag v<version> on main whose version matches Cargo.toml and has its own section in CHANGELOG.md. Pushing the tag starts .github/workflows/release.yml: on clean Windows and Linux runners it runs build-release.ps1 -KeepVersion (the scripts run in PowerShell 7 on Linux too) and drafts a GitHub release with the Windows folder as a ZIP, the Linux folder as a .tar.gz, a SHA256SUMS of both and that CHANGELOG section as its text, marked as a prerelease below 1.0 or with a version suffix (-alpha.1); the draft is published by hand. CI runs check.ps1 on Windows and on Linux.

Mutation testing uses cargo-mutants and cargo-nextest with the settings in .cargo/mutants.toml (simulator, drawing code, the manual's command line and the calls into Windows' browser and network left out) and the mutants profile in .config/nextest.toml, which ends a run at the first failing test:

$env:NEXTEST_PROFILE = 'mutants'
cargo mutants --workspace -j 4                    # everything, about four hours
cargo mutants --workspace -j 4 --in-diff <patch>  # the lines a change touched

More than four jobs did not make it faster, even on a machine with many more cores.

The workspace has three crates: crates/core (the SSC client, discovery, the path catalog, settings and a speaker simulator for the tests), crates/app (the window) and crates/manual (the manual generator, not part of the app).

MIT, see LICENSE. The licences of the bundled dependencies are in THIRD-PARTY-LICENSES.html; the Inter font is under the SIL Open Font License.