Audio modules

Self-contained flake with a PipeWire audio stack and a CLI backend — UIs (quickshell, waybar, whatever) are expected to be thin frontends over the audio-* commands, so you can build your own.

• pipewire (also default) — PipeWire + WirePlumber with:

  • RNNoise denoised virtual source (rnnoise_source): the whole audio stack lives behind one virtual mic. Denoising is bypassed in-graph via a dry/wet mixer (flip Gain 1/Gain 2 on the filter-chain node with pw-cli set-param <id> Props), so toggling the filter never switches devices out from under apps. VAD tuned to gate hallucinated noise while a 700ms grace period keeps speech from clipping.
  • Hot-plug mic combiner (combined_mics): every physical mic (ALSA + bluetooth) mixed into one mono virtual source via combine-stream rules; per-mic levels adjustable as recording streams. Latency compensation is deliberately off (wireless dongles report bogus latencies).
  • Firefox with pipewire-screenaudio native messaging host, for sharing system audio in screen captures (note: this overrides the firefox package — omit or override if you build firefox differently).
  • rtkit, ALSA (+32-bit), PulseAudio emulation, noisetorch.

• tools — the backend CLI: ~30 audio-* commands on PATH plus an audioctl dispatcher (audioctl help lists them). Toggles/status for the RNNoise bypass, mic blend, output duplication (MIX), USB headroom, the xrun guard and auto-mic daemons, plus enumerators (sinks, sources, app streams, blend mics, dup sinks) and level/VAD meters that stream values for UI meters. Also runs the two daemons as user services:

  • audio-xrun-guard — passive USB crackle guard; raises the USB sinks' ALSA headroom on real underruns, decays it when quiet. Opt-in via flag file (audio-xrun-guard-toggle).
  • auto-mic — VAD-driven automatic mic switcher, idle until configured (~/.config/auto-mic/config.json).

  Declares programs.audioctl.enable (default true) — frontends can gate their audio controls on it via home-manager's osConfig.

Usage

{
  inputs.audio.url = "github:flashgnash/audio.nix";

  outputs = { nixpkgs, audio, ... }: {
    nixosConfigurations.myhost = nixpkgs.lib.nixosSystem {
      modules = [
        audio.nixosModules.default
        # ... your other modules
      ];
    };
  };
}

Notes

• Do NOT add a static combine-stream output duplicator to this config: with passive device streams the sinks never wake (apps playing into it hang); with active ones every sink runs from boot, which can wedge a USB sink in a permanent XRUN (no audio at all). Load module-combine-sink via pipewire-pulse (pactl load-module) on demand instead, and unload it when done.
• The combiner's match regexes must only ever hit hardware inputs — matching another Audio/Source like rnnoise_source creates a feedback loop.