Skip to content

Sonero 0.1.4 — The microphone

Choose a tag to compare

@Tatsu21 Tatsu21 released this 26 Sep 21:41
· 21 commits to main since this release

One feature, and it is the one that was missing: the microphone now processes your
voice instead of passing it through. Everything else here is repair work on what
0.1.3 shipped.

The microphone is new, and new in the way that matters — this is the first
release where it does anything to your audio.
It is tested: 21 automated tests
over the chain and the noise model, and the whole path measured through real
recordings. It is also a few days old, half its stages have no controls yet, and
it has run on one person's hardware. Treat it as the good beta it is; the section
below says exactly what is finished and what is not.

For the Arch package and the update check, see the
0.1.3 notes — both are
unchanged here.

What's new

  • A microphone that does something. The Sonero Microphone source your
    applications select now runs a broadcast chain instead of forwarding audio: a
    high-pass, learned noise suppression, a gate that fades between sentences rather
    than chopping them, a de-esser, a ten-band voice equalizer, a compressor and a
    brick-wall limiter. Presets for podcasting, streaming and meetings, and a Raw
    preset that switches every stage off so you can hear what it is doing.
  • Noise suppression is RNNoise, and it runs ahead of everything else so the
    gate and the compressor measure speech against silence rather than against your
    room. On a quiet room here it takes about 9 dB out of the noise floor.
  • It works on every distribution, not just Arch. RNNoise is packaged by Arch
    and by nobody else — not Debian trixie, not Ubuntu 22.04 or 24.04, not Fedora in
    any release — so the .deb and the AppImage now carry it built and linked in.
    Building from source yourself: packaging/build-rnnoise.sh.
  • A voice equalizer of its own, ten bands, with presets. Every application
    recording from Sonero hears that curve, not just the one you set it up in.

Fixed

  • The microphone is no longer listed on the Channels page. It is an input: it
    has no output device to pick and no place in the stream mix. Worse, both pages
    owned an equalizer for it, and the Channels page overwrote the microphone's copy
    on every launch — so a curve set there did not survive a restart.
  • The combo box arrows were a small dash rather than an arrow, in every
    dropdown in the app.
  • The mixer left a third of the window empty beside the channel strips, and
    Master and Aux sat a few pixels lower than the other four because their header
    glyphs are a different kind of character.
  • The udev command on the Devices page pointed at a relative path, so copying
    it worked only for someone sitting in the source tree. It now shows the rule's
    real location and points at the one-click fix in Settings.

The microphone: what is finished and what is not

Working, and what the chain runs today:

  • The chain preset, input gain, output level, mute, noise suppression and the
    ten-band voice equalizer — all live, all remembered across restarts.
  • The gate, de-esser, compressor and limiter run. They are part of every preset
    and they are why the microphone already sounds processed.

Not finished:

  • The gate, de-esser and compressor have no controls. They run at their
    preset's values and their cards are marked PREPARED. Choosing a different chain
    preset is the only way to change them for now.
  • The input device cannot be chosen. The chain follows the system default
    microphone.
  • Monitoring — hearing yourself — is not implemented.
  • Noise suppression needs RNNoise. It is in the .deb and the AppImage; if you
    build from source without it, that one stage reports itself unavailable and your
    audio passes through untouched rather than silently going missing.

What "tested" means here: 15 automated tests over the processing chain, 6 over the
noise model, and the whole path measured through real recordings — mute produces
digital silence, a +10 dB equalizer band arrives as +10 dB, and the limiter never
passes its ceiling whatever the input gain is set to. What has not happened is
other people using it, on other microphones, in other rooms. If it does something
strange, that is worth an issue.

Requirements

  • PipeWire 0.3+ with WirePlumber — already running on every current desktop.
  • Qt 6.2+ (the AppImage carries its own).
  • RNNoise 0.2, for the microphone's noise suppression only. In the .deb and
    the AppImage already; rnnoise on Arch; packaging/build-rnnoise.sh elsewhere.
  • x86-64.

Which file to download

File For
none — build the package Arch, Manjaro, EndeavourOS (see the command below)
Sonero-0.1.4-x86_64.AppImage Anything. Portable, built on the oldest glibc, carries its own Qt.
sonero_0.1.4-ubuntu2204_amd64.deb Ubuntu 22.04, Linux Mint 21.x
sonero_0.1.4-ubuntu2404_amd64.deb Ubuntu 24.04, Linux Mint 22.x
sonero_0.1.4-debian13_amd64.deb Debian 13 (trixie)
Sonero-0.1.4-arch-x86_64.AppImage Arch, if you would rather not build
Sonero-0.1.4-fedora-x86_64.AppImage Fedora, or a distribution at least as new

The last two exist because CI builds on those distributions to catch bundling
regressions. They link a newer glibc and offer nothing the portable one does not.

# Arch, Manjaro, EndeavourOS — builds a real package and installs it with pacman
git clone https://github.com/Tatsu21/sonero.git
yay -B sonero/packaging/aur/sonero
# without an AUR helper: makepkg -si -D sonero/packaging/aur/sonero
# and packaging/aur/sonero-git builds main as it stands

# AppImage — needs FUSE 2, or run with --appimage-extract-and-run
chmod +x Sonero-0.1.4-x86_64.AppImage
./Sonero-0.1.4-x86_64.AppImage

# Debian / Ubuntu / Mint
sudo apt install ./sonero_0.1.4-ubuntu2204_amd64.deb

yay -S sonero still does not work: the package is not on the AUR, because
registrations there remain closed. The PKGBUILDs are finished and CI builds them
on every push; the day registrations reopen, the command above becomes
yay -S sonero.

Worth knowing before you install

  • Your microphone will sound different. The chain starts on the Podcast preset,
    which gates, compresses and de-esses. If you preferred it raw, the chain preset
    at the top of the Microphone page has a Raw setting that switches every stage
    off.
  • The update check is off until you switch it on, in Settings → Updates.
    Nothing contacts GitHub until you ask.
  • The AppImage deletes the bundle it replaces. Updating means downloading a new
    file, so without that every version you ever ran would pile up in ~/Downloads.
    It only removes a bundle a previous run recorded — never a file it went looking
    for, and never the one you just started.
  • On GNOME the tray icon needs the AppIndicator extension. GNOME ships no tray
    of its own; without the extension the window still works, but there is no tray
    icon and no background mode to return from.
  • Two features ask for permission, from Settings → System integration, which
    shows the exact commands before your desktop asks for a password: a udev rule for
    SteelSeries battery (the .deb and the pacman package install it for you), and a
    systemd drop-in that turns on BlueZ's experimental interfaces for Bluetooth
    battery. Both are reversible by deleting one file.
  • Changing a device's sample rate or bit depth lands on the next WirePlumber
    load.
    ALSA fixes the format when the PCM is opened, and Sonero will not restart
    your audio session behind your back — that would drop every Bluetooth link.

Not finished yet

These are visible in the app but greyed out and labelled, rather than left looking
clickable:

  • Profiles — saving whole configurations as named profiles is not implemented.
  • The microphone's gate, de-esser and compressor — see the section above.
  • SteelSeries hardware EQ — the dock accepts the HID writes and nothing audible
    changes; unsolved, so the card is disabled. The per-channel equalizer runs in
    software and works with any headset.

Tested hardware

Detection is generic, so any output PipeWire exposes should work. These are the
ones this release was actually tested against:

Device Connection
SteelSeries Arctis Nova Pro Wireless USB base station
Sennheiser MOMENTUM 4 Wireless Bluetooth, and wired over USB
Sony WH-CH720N Bluetooth

LDAC is untested. The codec picker ranks whatever A2DP profiles BlueZ reports
and puts LDAC on top, but none of the headsets above offers it, so that path has
never run on real hardware.

Battery over USB HID is model-specific, so only the Arctis Nova Pro Wireless is
implemented. Everything else still plays audio normally. If you want your headset
supported and you do not write C++,
hid/SKILL.md is a
written procedure meant to be handed to an AI assistant so it can do the work.

How this was built

Sonero was written with AI assistance — Anthropic's Claude, used as a pair
programmer. Every feature started as something I asked for, every change was read
before it was kept, and the parts that matter — gain staging, EQ headroom, the HID
protocol, Bluetooth codec behaviour — were checked against real hardware and real
recordings. Said openly because the project is open; the code is the same either
way.

Feedback

Bug reports and pull requests are welcome, and for the microphone especially: it is
the newest thing here and the least exercised. It is built in spare time, by one
person, and will move at that pace. If something breaks, the two most useful things
to include are the version from Settings → About — it names the exact commit —
and the log, which Sonero writes to the terminal:

./Sonero-0.1.4-x86_64.AppImage    # or just: Sonero

MIT licensed.

Full Changelog: 0.1.3...0.1.4