nerd-dictation

nerd-dictation turns speech into text for people who want voice input in their Linux applications. It runs recognition locally with VOSK and works offline, with command-line controls for dictation.

• Application input: It can type recognized speech into applications or send text to other programs through standard output.
• Text customization: Python scripts let you edit recognized text, including replacements and capitalization, before it reaches an application.
• Spoken numbers: Optional number conversion handles quantities, ordinals and sequences such as phone numbers.
• Recording controls: You can stop dictation after silence or pause recording while keeping the model in memory to avoid loading it again on resume.

The tool is open source under GPL-3.0. It supports X11, Wayland and Linux terminals through compatible input tools, and can capture audio through PulseAudio, PipeWire or SoX. Elograf is a separate tray-icon front end for people who prefer graphical controls.

Dictation requires manual activation rather than an always-running background service. Cold starts can take time on slower systems or with larger models. VOSK produces lowercase text, so capitalization depends on custom processing.