Live captions for Linux. NowCaption shows what is being said in videos, calls, streams and games, or into your microphone, in a small window that floats over your desktop. It is made for people who are deaf or hard of hearing.
- Private. Everything runs on your computer, offline. No account, and audio never leaves it.
- Accurate. Punctuation and capital letters included. 2.4% word errors on clear speech, a third of upstream Live Captions' 8.1% (benchmark).
- Light. Runs on the CPU as speech arrives. No GPU needed.
flatpak install --user --from https://techlm77.github.io/NowCaption/nowcaption.flatpakrefUpdates then arrive with flatpak update or your software centre. Works on x86_64
and ARM64. You can also download a .flatpak file from the
latest release and run
flatpak install --user NowCaption-*.flatpak.
To uninstall: flatpak uninstall io.github.Techlm77.NowCaption.
Captions start straight away from whatever your computer is playing. Drag them to move, drag an edge to resize. Right-click for the menu:
- Desktop audio or Microphone
- Background filter: picks voices out of music and noise. Off by default, as it uses a little more CPU and slightly lowers accuracy on plain speech.
- Text size, background, position
- Copy, save or clear the transcript
On KDE Plasma, Hyprland and Sway the captions float above all windows. On GNOME, press Alt+Space on the window and choose Always on Top.
No captions? Make sure sound plays through your default output device. After switching to headphones, choose Desktop audio again.
Needs GTK 4.12+ and libadwaita 1.5+ (Debian 13, Ubuntu 24.04, Fedora 40, Arch).
# Arch
sudo pacman -S python-gobject python-cairo gtk4 libadwaita gst-plugins-base gst-plugins-good gtk4-layer-shell
# Debian / Ubuntu
sudo apt install python3-gi python3-gi-cairo python3-venv gir1.2-gtk-4.0 gir1.2-adw-1 \
gir1.2-gstreamer-1.0 gir1.2-gst-plugins-base-1.0 gstreamer1.0-plugins-good gstreamer1.0-pulseaudio
# Fedora
sudo dnf install python3-gobject python3-cairo gtk4 libadwaita gstreamer1-plugins-base gstreamer1-plugins-good
./tools/run-dev.sh # downloads the models on first run
./tools/build-flatpak.sh # or build the FlatpakTests: PYTHONPATH=src .venv/bin/python tests/test_reliability.py (fast, no models);
the other files in tests/ say what they cover.
| Part | Model | License |
|---|---|---|
| Speech recognition | NVIDIA FastConformer streaming, int8 ONNX, 80 ms lookahead | CC-BY-4.0 |
| Punctuation and casing | Edge-Punct-Casing | Apache-2.0 |
| Background filter | DPDFNet2 by Ceva | Apache-2.0 |
Everything bundled is listed in MODEL-LICENSES.md. NowCaption's own code is MIT licensed.