OPINIONATED OMARCHY
TROUBLESHOOTING CORPUS
790
RECORDS
790
AUDITED
0
UNCHECKED
2388
SOURCES
LIVE
audio-input medium common DISPUTE

People on calls hear your keyboard, fan, and an echo of themselves

archendeavouroscachyosmanjaroomarchypipewirelaptopdesktopwayland
CORRECTED An auditor found the fix wrong or incomplete and rewrote it.
AUDIT NOTE

Checked on this Omarchy 4 workstation (omarchy 4.0.2-1, pipewire and pipewire-pulse 1:1.6.8-1, wireplumber 0.5.15-1, kernel 7.1.9-arch1-2) and against the upstream pages. Everything technical in the record still holds on 1.6.8 and I confirmed it locally rather than from memory: strings /usr/lib/spa-0.2/aec/libspa-aec-webrtc.so contains all four keys the record sets (webrtc.high_pass_filter, webrtc.noise_suppression, webrtc.gain_control, webrtc.voice_detection), strings /usr/lib/pipewire-0.3/libpipewire-module-echo-cancel.so contains library.name, node.latency, monitor.mode, aec.args and all four props blocks, and docs.pipewire.org/page_module_echo_cancel.html now renders as PipeWire 1.6.8 with an example matching the record arg for arg. The module ships in pipewire-audio 1:1.6.8-1 (pacman -Qo). noise-suppression-for-voice is extra 1.21-1, not flagged out of date, and its file list carries usr/lib/ladspa/librnnoise_ladspa.so. swh-plugins 0.4.17-7 and easyeffects 8.2.8-1 are also in extra. The label noise_suppressor_mono and all three control names match werman's README verbatim, and both version footnotes match the Arch PipeWire page verbatim. The record's own structure already separates the three symptoms correctly, so that check passed: part A is echo and says both the source and the sink must be routed through the module, part B is keyboard, fan and hiss and is source side only. Three Omarchy 4 facts are missing and they are why this is corrected rather than ok. First, the record hosts the RNNoise chain in ~/.config/pipewire/pipewire.conf.d/. That still loads on 1.6.8, but PipeWire 1.6 ships /usr/share/pipewire/filter-chain/source-rnnoise.conf whose header says to copy it into ~/.config/pipewire/filter-chain.conf.d/ and run pipewire -c filter-chain.conf, and Omarchy's own speaker tuning migrated off the daemon drop-in for a stated reason I read in /usr/share/omarchy/default/systemd/user/omarchy-speaker-tuning.service: a malformed tuning breaks only that client, where a bad drop-in in the daemon config stops PipeWire from starting at all. /usr/share/pipewire/filter-chain.conf confirms it merges ~/.config/pipewire/filter-chain.conf.d/. Second, /usr/share/omarchy/bin/omarchy-audio-tuning refuses to install while EasyEffects is running and sets omarchy_speaker_tuning as the default sink, so the record's EasyEffects recommendation and its instruction to make the AEC sink the default both silently bypass a shipped tuning (/usr/share/omarchy/default/audio/tunings/dell-xps-2026/ is the one currently shipped). Third, Omarchy ships /usr/share/omarchy/bin/omarchy-restart-audio, which restarts the same three units and additionally recovers a stuck USB audio device, so it is the supported restart here. I also corrected the symptom: pavucontrol is not installed on Omarchy 4 and is owned by no package. I checked the ALPM guard at /usr/share/libalpm/hooks/00-omarchy-update-guard.hook and it triggers only on Operation = Upgrade, so the record's pacman -S --needed is not blocked, and I left it with a note about syncing first. NOT exercised: I changed nothing on this machine. No config was written, no filter chain was loaded, no service restarted and no default device moved, so neither half was run end to end here. noise-suppression-for-voice, swh-plugins and easyeffects are not installed, so the LADSPA label was confirmed from upstream and the package file list rather than from a loaded plugin. Live state read only: PulseAudio on PipeWire 1.6.8, default sink alsa_output.pci-0000_00_1f.3.analog-stereo, default source the USB Blue Yeti, no filter nodes present, and ~/.config/pipewire does not exist.

The cause above was not rewritten and may still contain the error described. The fix below is the corrected version.
RISK

Setting a virtual sink as your system default means that if you later delete the config without resetting the default, apps chase a sink that no longer exists and you get silence. Reset with wpctl set-default <real sink id> before removing it. Echo cancellation also adds latency (default node.latency 1024/48000, about 21 ms), so do not leave it as the default output for music or gaming. A malformed drop-in under ~/.config/pipewire/pipewire.conf.d/ stops the PipeWire daemon from starting, which takes all audio with it. Recover by deleting the file and running omarchy restart audio. That is the reason part B is hosted as a pipewire -c filter-chain.conf client instead. On Omarchy, both an AEC sink set as default and EasyEffects will take playback off the shipped speaker tuning. Check omarchy audio tuning status first.

Symptom

On Zoom, Google Meet, Discord or Teams, everyone complains that they can hear your mechanical keyboard, your laptop fan, and room hiss whenever you are not speaking. Worse, when you use speakers instead of headphones, the other side hears their own voice bounced back at them half a second later. The same machine was fine under Windows or macOS. Nothing in the Omarchy audio menu, wpctl status or pactl list offers a noise suppression or echo cancellation switch, and Omarchy 4 does not install pavucontrol either.

Cause

PipeWire performs no DSP on capture streams by default: what the microphone hears is what the app gets. module-echo-cancel exists but is not loaded unless you write a drop-in for it. The browser-side WebRTC AEC that handles this on other platforms is frequently bypassed on Linux (the app has to own both the capture and playback path for it to work), and on Wayland screen/audio capture paths it often does not engage at all. Neither GNOME nor KDE ships a toggle for it either, so nothing on an Arch/Hyprland box will do it for you.

Fix

There are three symptoms in the title and **two** independent fixes. Noise suppression is a filter on the source only. Echo cancellation needs the source *and* the sink, because the module can only subtract what it can see. Do not expect one config to do both.

**A. Echo (they hear themselves): PipeWire's WebRTC AEC.** Create ~/.config/pipewire/pipewire.conf.d/20-echo-cancel.conf:


context.modules = [
  { name = libpipewire-module-echo-cancel
    args = {
      # library.name  = aec/libspa-aec-webrtc
      # node.latency  = 1024/48000
      # monitor.mode  = false
      aec.args = {
        webrtc.high_pass_filter  = true
        webrtc.noise_suppression = true
        webrtc.gain_control      = false
        webrtc.voice_detection   = true
      }
      capture.props = {
        node.name = "Echo Cancellation Capture"
      }
      source.props = {
        node.name        = "Echo Cancellation Source"
        node.description = "Echo Cancelled Mic"
      }
      sink.props = {
        node.name        = "Echo Cancellation Sink"
        node.description = "Echo Cancelled Output"
      }
      playback.props = {
        node.name = "Echo Cancellation Playback"
      }
    }
  }
]

This one has to go in pipewire.conf.d/, because it loads into the daemon. The daemon reads that directory only at startup, so it needs a restart. On Omarchy use the shipped wrapper, which also recovers a USB audio device that hangs on the way down:


omarchy restart audio      # or: omarchy-restart-audio

On plain Arch:


systemctl --user restart pipewire pipewire-pulse wireplumber
wpctl status

Both ends must go through the module, so set the new source as the default input **and** the new sink as the default output:


# take the IDs from `wpctl status`
wpctl set-default <id of Echo Cancellation Source>
wpctl set-default <id of Echo Cancellation Sink>

**On Omarchy, check omarchy audio tuning status before you touch the default sink.** If a speaker tuning is installed, omarchy_speaker_tuning is the default sink and every app is moved onto it. Making the Echo Cancellation Sink the default takes playback back off the tuned path. Turn the tuning off deliberately (omarchy audio tuning off) rather than leaving both half applied.

**B. Keyboard, fan and hiss: RNNoise filter chain.** Separate DSP, and it can be stacked with A.


sudo pacman -S --needed noise-suppression-for-voice swh-plugins

On Omarchy run omarchy update first if you have not synced recently. The ALPM guard only fires on an upgrade, so a plain -S of a new package is not blocked, but installing against a stale database is still how a partial upgrade starts. Never reach for pacman -Sy.

PipeWire 1.6 ships a worked example of this chain at /usr/share/pipewire/filter-chain/source-rnnoise.conf, and its own header tells you where it belongs. **Host it as a filter-chain client, not as a daemon drop-in.** A malformed fragment under filter-chain.conf.d/ breaks only that client. The same mistake under pipewire.conf.d/ stops the audio daemon from starting at all, which is why Omarchy's speaker tuning moved out of pipewire.conf.d/ and runs its own pipewire -c client instead.


mkdir -p ~/.config/pipewire/filter-chain.conf.d

Create ~/.config/pipewire/filter-chain.conf.d/99-input-denoising.conf:


context.modules = [
{   name = libpipewire-module-filter-chain
    flags = [ nofail ]
    args = {
        node.description = "Noise Canceling source"
        media.name       = "Noise Canceling source"
        filter.graph = {
            nodes = [
                {
                    type   = ladspa
                    name   = rnnoise
                    plugin = librnnoise_ladspa
                    label  = noise_suppressor_mono
                    control = {
                        "VAD Threshold (%)"          = 50.0
                        "VAD Grace Period (ms)"      = 200
                        "Retroactive VAD Grace (ms)" = 0
                    }
                }
            ]
        }
        capture.props = {
            node.name     = "capture.rnnoise_source"
            node.passive  = true
            audio.rate    = 48000
            audio.channels = 1
            audio.position = [ MONO ]
        }
        playback.props = {
            node.name   = "rnnoise_source"
            media.class = Audio/Source
            audio.rate  = 48000
            audio.channels = 1
            audio.position = [ MONO ]
        }
    }
}
]

Try it in a terminal first. Nothing is restarted and Ctrl+C takes it away again:


pipewire -c filter-chain.conf

filter-chain.conf comes from /usr/share/pipewire/ and merges every fragment in ~/.config/pipewire/filter-chain.conf.d/. Once it works, keep it with a user unit, ~/.config/systemd/user/rnnoise-source.service:


[Unit]
Description=RNNoise noise-canceling source
After=pipewire.service wireplumber.service
Requires=pipewire.service
Wants=wireplumber.service
PartOf=pipewire.service

[Service]
Type=simple
ExecStart=/usr/bin/pipewire -c filter-chain.conf
Restart=on-failure
RestartSec=2

[Install]
WantedBy=graphical-session.target

systemctl --user daemon-reload
systemctl --user enable --now rnnoise-source.service

Then pick **Noise Canceling source** as the input in the app, or wpctl set-default <its id>.

*Alternative, if you would rather keep it in the daemon:* the same context.modules block works in ~/.config/pipewire/pipewire.conf.d/99-input-denoising.conf, which is what the Arch wiki and upstream's README still show. It costs an audio restart on every edit and a typo costs you the daemon, so prefer the client above.

Two version-sensitive details in that file:

  • PipeWire 1.6.3 and newer resolve plugin as a *name* in the LADSPA search path, hence plugin = librnnoise_ladspa. On older PipeWire use the full path instead: plugin = /usr/lib/ladspa/librnnoise_ladspa.so.
  • The explicit audio.channels = 1 and audio.position = [ MONO ] lines in both prop blocks are needed since PipeWire 1.4.10 when the source reports stereo. Without them the chain can fail to instantiate.

**If you would rather click than edit config:** sudo pacman -S easyeffects, run it, go to the *Input* tab and add the *Noise Reduction* effect. It does the same RNNoise work with a GUI and per-app presets. **On Omarchy, EasyEffects and the shipped speaker tuning are mutually exclusive.** omarchy audio tuning on refuses to install while EasyEffects is running, because EasyEffects moves anything following the default sink onto its own sink and the tuning would be bypassed. Pick one:


systemctl --user disable --now easyeffects.service   # to keep the tuning

Verify

wpctl status lists the new virtual nodes under Sources (and Sinks for the AEC). For the filter-chain client, systemctl --user status rnnoise-source.service should be active and the node should appear without an audio restart. Record and listen back:


parecord --device=rnnoise_source /tmp/test.wav   # type loudly while recording
# Ctrl+C, then:
paplay /tmp/test.wav

For the AEC, play music out of the speakers, record from the Echo Cancellation Source, and confirm the music is largely absent from the recording. pw-top shows the filter nodes running while a call is active. parecord and paplay come from libpulse, which is already installed alongside pipewire-pulse.

Sources

mic-noise-suppression-echo-cancellation-pipewire