Voice Terminal

Voice Terminal is a side-by-side fork of Xfce Terminal with local push-to-talk speech recognition. It keeps the familiar Xfce/VTE terminal experience while adding an offline dictation bar that can paste recognized text into the active terminal tab.

This fork is based on xfce4-terminal 1.1.4 and installs separately as voiceterm, with its own desktop files, icon, helper scripts, and Xfconf channel.

Voice Features

• Push-to-talk dictation with F8.
• Press F8 again to discard the current dictation.
• Press Enter to finalize and insert the recognized text.
• Optional Auto Enter mode to submit the pasted command after dictation.
• Per-window GPU affinity for local Whisper inference.
• Unload Model button to stop the speech helper and release VRAM without closing terminal windows.
• Local-only helper based on Python, PyTorch, openai-whisper, and sounddevice.

The speech model files are not stored in this repository. Use the bootstrap helper to prepare the local Python/Conda environment and download the selected Whisper model:

helpers/bootstrap-voice-env.sh

The default model is turbo (large-v3-turbo). It can be changed from Preferences > Voice, or by setting VOICETERM_WHISPER_MODEL before running the bootstrap helper.

Build

From a checkout:

./autogen.sh
make
sudo make install

The installed binary is:

voiceterm

Debian Package

Build an installable .deb for the current machine architecture with:

packaging/build-deb.sh

The package is written to dist/. It includes the terminal binary, desktop files, icons, and voice helper scripts. Whisper model files and Python packages are intentionally not bundled; prepare the local ASR runtime after installing with:

/usr/share/voiceterm/helpers/bootstrap-voice-env.sh

Upstream

This project is a fork of Xfce Terminal:

https://gitlab.xfce.org/apps/xfce4-terminal

Xfce Terminal is based on the VTE terminal widget library and remains the foundation of the terminal emulator in this fork.