wispr-fox v3.1.0-nightly.1

Pre-release / nightly build. Not the latest stable.
Stable users: see Releases page and pick the most recent "Latest" tag.

wispr-fox v3.1.0-nightly.1 — Noise reduction for fan-heavy laptop mics

Built-in laptop mics pick up the machine's own fans — a low whirring bed under
everything you say that makes the transcriber mishear words. This nightly adds
an opt-in noise-reduction stage that scrubs that noise out of the audio sent
for transcription, tuned on real recordings from the affected machines.

Noise reduction — Off / On / Aggressive

New in Settings → Dictation → Noise reduction:

• Off (default) — nothing changes; audio goes out exactly as recorded.
• On — a rumble filter that removes the deep fan hum below ~90 Hz. It cuts
  the loudest part of the fan signature and by design cannot touch speech —
  safe to leave on permanently.
• Aggressive — the rumble filter plus an AI speech denoiser (RNNoise, the
  same family of tech Discord and OBS use). On the test recordings this took
  the noise floor down by 20–45 dB and roughly doubled-to-tripled the
  signal-to-noise ratio. If a transcript ever comes out worse, just step back
  down to On.

Built so it can't hurt you

• Your saved recordings are untouched. The cleanup happens on a temporary
  copy that only the transcription provider sees. History always keeps and
  plays the original audio, so you can compare and diagnose after the fact.
• No noticeable wait. Processing runs locally at roughly 130–600× realtime:
  a 10-second dictation costs ~15–80 ms, a full minute well under half a
  second — background noise compared to the transcription request itself.
• The avatar tells you what's happening. The floater shows a quick
  "clearing noise…" beat for exactly as long as the denoiser runs, and each
  recording's flight recorder logs the precise time it took (visible in the
  per-recording timeline).
• Never costs a dictation. If the denoiser hits any problem, the pipeline
  quietly falls back to the raw recording and carries on.

Notes

• Applies to hotkey dictations. Uploaded audio files are transcribed as-is.
• Desktop only for now (Windows + macOS + Linux builds all include it).

---

Built from commit 742e116.
Provenance: built by GitHub Actions from the tagged commit. See the release notes header for which agent authored the change set.

Pick your installer:

• Windows: wispr-fox_*_x64-setup.exe (recommended) or *_x64_en-US.msi
• macOS Apple Silicon (M1/M2/M3/M4): wispr-fox_*_aarch64.dmg
• Linux: *.AppImage (portable) or *.deb (Ubuntu/Debian)

First-time install — please read

Windows. Run the .exe, click through SmartScreen ("More info" → "Run anyway") on first launch since the binary isn't code-signed yet.

macOS. The build isn't notarized. One of two things will happen on first launch:

• "developer cannot be verified" dialog: Cancel → right-click the app in Applications → Open → click Open in the new dialog.
• "App is damaged" dialog: open Terminal and run xattr -dr com.apple.quarantine /Applications/wispr-fox.app, then double-click normally.

Then grant two permissions when prompted:

1. Microphone (first time you press F8) — required
2. Accessibility (first time it tries to paste) — required for fast text injection; if you skip it, the app falls back to clipboard+Cmd+V which still works but is slower

Full install walkthrough in the README.

Linux. AppImage is portable — chmod +x and run. Untested by maintainer; report issues.