push-to-whisper 0.1.1

Yet another voice memo tool

push-to-whisper

A smart voice memo tool aka push-to-stt-to-md-to-llm-to-clipboard-or-whatever.

What you can do with push-to-whisper:

• Record audio while holding a global key combination.
• Save the recording as a .wav file (e.g., directly into your Obsidian vault).
• Transcode it into .ogg or other formats for efficiency (via ffmpeg).
• Transcribe it into Markdown using Whisper (Currently supports whisper.cpp server).
• Refine the text using LLM APIs like OpenAI, Gemini, or Ollama (via LiteLLM).
  • Auto tagging, auto summarization, etc.
• Copy the result to your clipboard automatically.
• Notify success or send results to notification services like Slack, Discord, or Ntfy (via Apprise).

Every step above is modular. You can combine them to build your own custom workflow in a simple YAML configuration file.

Installation

1. Install system dependencies:

   # Debian/Ubuntu
   sudo apt install libgirepository1.0-dev libcairo2-dev python3-dev ffmpeg
   

2. Install the package using uv:

   uv tool install push-to-whisper
   

3. Install the systemd user service and generate a default config:

   push-to-whisper install-daemon
   

Configuration

The configuration file is located at ~/.config/push-to-whisper/config.yaml. You can customize the Whisper endpoint, LLM API keys (LiteLLM), and processing pipelines.

To re-initialize or export the default configuration:

push-to-whisper init --bare -o ~/.config/push-to-whisper/config.yaml

Usage

Once the daemon is installed via install-daemon, it will start automatically on login.

Default Shortcuts

On Linux (KDE/GNOME), shortcuts are managed by the system. After running install-daemon, you can assign keys to the following actions in your system settings:

• Transcription to Markdown: (Recommended: ALT+SHIFT+x) - Transcribe -> Transcode -> Save Audio -> Save Markdown -> Notify.
• Transcription to Clipboard: (Recommended: ALT+SHIFT+c) - Transcribe -> Transcode -> Copy to Clipboard -> Notify.

Note: Currently tested and supported only on Linux (Fedora) with KDE Plasma (Wayland). Native support for Windows and macOS is planned for future releases.

Development

• Formatting: uv run ruff format .
• Linting: uv run ruff check . --fix
• Testing: uv run pytest

License

MIT