Skip to content
LothnicPublic

About

Local, private voice dictation for Linux. Hold a key, speak, and clean text appears in the app you're using.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

188 Commits

Folders and files

Repository files navigation

Falco

Falco is a local voice dictation app for Linux. Hold a dictation key, speak, and the text is typed into the application you are using. Speech recognition runs on your machine, with optional local transcript cleanup; models are downloaded once.

It is an experimental community app built on TuxTalk, in a fork of the Phonon repository. It uses Phonon's separately distributed refiner for optional cleanup and adds a Rune dashboard and recording overlay. Falco is not an official Phonon or TuxTalk release.

Website · Full Linux guide · Upstream Phonon README

Features

  • Dictate anywhere. Hold a configurable key to talk, or use a toggle key. Falco types into the focused app on Wayland and X11.
  • On-device speech recognition. Parakeet Ultra v3 is the default and runs on the CPU or through Vulkan. Whisper models cover more languages and can use CUDA.
  • Optional local cleanup. The Phonon refiner fixes self-corrections, removes fillers and false starts, and tidies capitalization. The refiner is English-only, and dictation uses the raw transcript until it is installed.
  • Your vocabulary. Custom words steer recognition, and voice commands handle punctuation, formatting, undo, and "fix that word". Falco Labs adds optional voice editing, selection, and app, window, key, and mouse control.
  • Rune dashboard and overlay. Overview, Recent, Your words, and Settings pages, plus a sound-responsive recording overlay. The GTK tray and recording indicator remain the fallback.
  • Private by design. Audio and transcripts stay on the machine. No account is required, and dictation needs no network once the models are downloaded.

How it works

microphone → Parakeet Ultra or Whisper → cleanup rules → optional refiner → focused app

On first run, Falco downloads the speech model. To add cleanup, open Preferences → Advanced and enable Refine dictation with Phonon's local model; the app offers a one-time download of about 1 GB for the refiner and its llama.cpp engine.

Recent dictations in the Rune dashboard are session data, kept only in memory for the current login; they are not saved as a permanent history.

Requirements

  • Linux with Python 3.10–3.14 and GTK 3. Ubuntu is the primary tested distribution; GNOME and KDE Plasma on Wayland or X11 are the intended environments.
  • PyGObject, D-Bus, evdev, PortAudio, and wl-clipboard. The apt command below installs these.
  • Rust 1.99.1 or newer only for the optional Rune dashboard and overlay. Without it, Falco uses the GTK tray interface.
  • About 1.7 GB for the default speech model and refiner.

Falco needs access to keyboard events and virtual input to read the dictation key and type into other applications. The first-run screens guide you through the permissions it needs.

Install

The quick installer handles the system packages and the app in one command:

curl -fsSL https://falco.lothnic.dev/install.sh | bash

It supports apt, dnf, pacman, and zypper, installs the dependencies with sudo, and then installs Falco under your user account. Add --with-rune (with Rust 1.99.1 or newer already installed) to build the Rune dashboard and overlay first:

curl -fsSL https://falco.lothnic.dev/install.sh | bash -s -- --with-rune

Use --no-deps if you prefer to install the packages yourself, and --ref to install a tag or branch other than main. The installer is plain Bash; you can download it and read it before running it.

Manual install

Install the system packages (package names vary by distribution):

sudo apt install python3 python3-venv python3-pip python3-gi python3-gi-cairo \
  python3-dbus python3-evdev gir1.2-gtk-3.0 \
  gir1.2-ayatanaappindicator3-0.1 libportaudio2 wl-clipboard

Clone and install:

git clone https://github.com/Lothnic/falco.git
cd falco

# Optional: build the Rune dashboard and recording overlay.
cargo build --release --locked --manifest-path linux/rune-overlay/Cargo.toml

# Installs under ~/.local/share/falco-app and adds ~/.local/bin/falco.
./linux/install.sh

Start Falco from your application menu, or run falco (use ~/.local/bin/falco if that directory is not on your PATH). The first-run screens configure permissions, a dictation key, and a speech model. If both Rune binaries are present, the launcher opens the Rune dashboard; otherwise it opens the GTK tray interface.

The installer refuses to overwrite an existing installation. Remove ~/.local/bin/falco, ~/.local/share/applications/falco.desktop, and ~/.local/share/falco-app before reinstalling.

Development

Set up a development environment from a checkout:

cd linux
python3 -m venv --system-site-packages .venv
.venv/bin/pip install -e .
.venv/bin/python -m localdictation.tray

Run the Linux tests (from linux/):

.venv/bin/pip install pytest
PYTHONPATH=src .venv/bin/python -m pytest tests

Build and test the Rune UI (from the repository root):

cargo build --release --locked --manifest-path linux/rune-overlay/Cargo.toml
cargo test --manifest-path linux/rune-overlay/Cargo.toml

Build and test the website:

cd website
npm install
npm test

localdictation is Falco's internal Python package name, retained from TuxTalk.

Repository layout

Only linux/ is the Falco application. website/ is the Falco landing page and browser demo. The remaining code is the upstream Phonon source, kept for history and reference under its own licenses.

Path Contents
linux/ Falco: Python dictation service, tray and preferences, Rune dashboard and overlay
crates/ Phonon Rust workspace: audio, ASR, refiner, pipeline, CLI, and Windows app
bar/ Phonon macOS Swift app and capsule
sidecar/ Phonon speech and refiner sidecars
engine/ Phonon native refiner engine
ml/ Phonon refiner training and evaluation pipeline
android/ Phonon Android keyboard (development build)
ios/ Phonon iPhone keyboard (development build)
website/ Falco landing page, browser dictation demo, and hosted install.sh

UPSTREAM-README.md holds the Phonon README this fork started from. SPEC.md and DISTRIBUTION.md document the upstream macOS and Windows projects and do not describe Falco.

Status

Falco is an early source package. Transcription and cleanup have been tested on Ubuntu, but the packaged app still needs broader end-to-end testing across desktop environments. The Rune dashboard and overlay, type-at-pause, and voice editing are experimental. Falco is not a port of the full upstream Phonon pipeline; it uses the refiner as an optional cleanup stage.

License

The Falco code in linux/ is MIT-licensed, based on TuxTalk, copyright 2024–2026 Ron Brand. See linux/LICENSE. The upstream Phonon source elsewhere in the repository remains GPL-3.0 under the root LICENSE; linux/ does not relicense it.

Model weights are downloaded separately and keep their own licenses. The Phonon refiner is under the LFM Open License, which restricts commercial use by larger organizations; Parakeet is CC-BY-4.0; llama.cpp is MIT. See the Phonon refiner model card, linux/README.md, and THIRD_PARTY.md.

About

Local, private voice dictation for Linux. Hold a key, speak, and clean text appears in the app you're using.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages