Dictate

Linux voice dictation system using faster-whisper (CTranslate2) for local speech-to-text conversion and automatic text input.

Overview

This system provides real-time voice dictation capabilities by:

• Recording audio from your system's microphone.
• Converting speech to text using highly optimized local models (OpenAI Whisper & Distil-Whisper).
• Automatically typing the recognized text into any application using pynput (with ydotool fallback).
• Supporting multilingual dictation and translation (e.g., speak French -> type English).

Features

• Fast: Uses faster-whisper backend with 8-bit quantization for <1s latency on modern CPUs.
• SOTA Models: Supports large-v3-turbo, distil-whisper, and standard OpenAI models.
• System Tray Control: Switch models, languages, and tasks (transcribe/translate) instantly from the tray.
• Smart Typing: Robust text injection handling Unicode characters (accents, emojis).
• Audio Management: "Discard Recording" feature to cancel bad takes.

Prerequisites

• Linux system with systemd
• Python 3.x
• Root access for installation
• Microphone
• ydotool (installed automatically by script)

Installation

1. Clone this repository:

git clone https://github.com/nilock/dictate
cd dictate

2. Run the installation script as root:

sudo ./installation.sh

This will:

• Install system dependencies (ydotool, portaudio, etc.)
• Create a Python virtual environment in /opt/dictation_venv
• Set up the background daemon (dictation.service) and tray app (dictation_tray.service)
• Configure permissions for ydotool (socket access)

Usage

Dictation

1. Trigger: Run dictation (or bind it to a hotkey like Ctrl+Alt+D).
2. Speak: The tray icon turns Red.
3. Stop: Run dictation again. The tray icon turns Grey (processing), then types the text.

Control

• Tray Menu: Right-click the system tray icon to:
  • Discard Recording: Cancel the current audio without typing.
  • Model: Select between speed (tiny, distil-small.en) and accuracy (large-v3-turbo).
  • Language: Force English (en), French (fr), or Auto-Detect.
  • Task: Choose Transcribe (Input Language -> Input Language) or Translate (Input Language -> English).
• CLI: You can also control it via terminal:

  dictation config --model large-v3-turbo
  dictation config --language fr --task translate
  dictation discard

Hotkey Setup

Bind the command /usr/local/bin/dictation to a custom keyboard shortcut in your desktop environment (GNOME, KDE, i3, etc.).

Troubleshooting

1. Check Status:

   systemctl status dictation
   systemctl --user status dictation_tray

2. View Logs:

   • Daemon (Recording/Transcribing): tail -f /tmp/dictation_daemon.log
   • Tray (Typing/UI): journalctl --user -u dictation_tray -f

3. Typing Issues: If text appears as numbers or gibberish (e.g. fran242...), ensure ydotool is working or try restarting the tray service to re-attempt pynput connection.

Configuration

Configuration is stored in ~/.config/dictation/config.json. You can edit this manually or use the CLI/Tray to update it.

Dependencies

• faster-whisper
• sounddevice
• numpy
• pynput
• pystray
• ydotool (System package)

License

GPL-3.0