Fono

Added

• Empty-transcript microphone recovery. When a recording lasts at
  least 3 seconds but produces no transcribed text — the typical
  symptom of an external dock advertising a passive capture endpoint
  the OS elected as the default source — Fono now pops a critical
  desktop notification naming the silent device, the recording
  duration, and the recourse: switch via the tray "Microphone"
  submenu, pavucontrol, or your OS sound settings. Auto-suggested
  alternatives are filtered to exclude HDMI / monitor / loopback /
  S/PDIF decoys.
• Tray "Microphone" submenu (Linux desktops with PulseAudio /
  PipeWire). One row per source the audio server reports, marked
  with the system default. Clicking a row runs
  pactl set-default-source so the change applies system-wide and
  is reflected in pavucontrol / GNOME / KDE settings, then
  hot-reloads the daemon so the next capture opens the new
  endpoint. Hidden on hosts where AudioStack::detect() returns
  Unknown (macOS, Windows, pure-ALSA Linux) — the OS owns
  microphone selection there.

Changed

• Microphone enumeration is now PulseAudio-first on Linux. When the
  audio stack is PulseAudio or PipeWire (Pulse compat layer),
  Fono lists sources via pactl list sources instead of cpal's
  ALSA host. Submenu rows show the source's friendly description
  ("Built-in Audio Analog Stereo", "Logitech BRIO") instead of
  cpal's raw plughw:CARD=… PCM names; the chronic
  snd_pcm_dsnoop_open: unable to open slave errors and the
  ALSA plugin pseudo-device clutter (pulse, oss, speex,
  default, surround51, …) that previously appeared in the
  submenu are gone. macOS, Windows, and pure-ALSA Linux fall back
  to cpal enumeration — unchanged.
• Microphone selection is fully delegated to the OS layer. Fono
  follows the PulseAudio / PipeWire default-source on Linux, the
  macOS Sound input device, and the Windows recording default.
  pavucontrol, GNOME / KDE settings, System Preferences and the
  Sound control panel are the canonical places to choose a
  microphone.
• fono doctor "Audio inputs:" section is now informational only.
  Lists every device the active stack reports with one row marked
  as the OS default; advice points at the tray submenu, pavucontrol,
  or OS sound settings.

Removed

• Tray "Languages" submenu removed. The Languages submenu that
  previously listed the configured BCP-47 peer set and offered
  a "Clear language memory" action has been removed from the tray.
  The language cache is cleared automatically and language preference
  is managed via config.toml or fono use language.
• [audio].input_device config field. Fono no longer keeps a
  capture-device override; the OS default is always used.
• fono use input <name> CLI subcommand. Use the tray "Microphone"
  submenu, pavucontrol, or your OS audio settings instead.
• First-run wizard's microphone picker. New users get the OS
  default; switching afterwards is a tray-submenu click on Linux
  desktops or an OS-settings change elsewhere.
• [general].language (deprecated scalar — use [general].languages).
• [stt.local].language (deprecated scalar — use
  [stt.local].languages or [general].languages).
• [general].cloud_force_primary_language (superseded by the
  in-memory language cache shipped in v0.3.x).
• cloud_force_primary builders, struct fields, and dead first-pass
  branches on GroqStt, GroqStreaming, and OpenAiStt.
• TrayAction::ClearInputDevice variant (no override to clear).

Full Changelog: https://github.com/bogdanr/fono/compare/v0.3.5...v0.3.6