Joplin Dictate

Record your voice, transcribe it locally with whisper.cpp, and turn it into a Joplin note or to-do. Transcription runs entirely on your machine — no audio leaves your computer. An optional step can polish the transcript with a local, OpenAI-compatible LLM.

Features

• One-click dictation from a dedicated Joplin panel (and the note toolbar / Tools menu).
• Pause / Resume and Cancel while recording.
• Transcribe an existing WAV file instead of recording.
• Local transcription via whisper.cpp (whisper-cli).
• Optional transcript polishing (grammar/punctuation/filler removal) via any OpenAI-compatible chat endpoint.
• Creates the note in the currently selected notebook, or a configurable fallback notebook.

Requirements

Platform note: recording uses pw-record from PipeWire, so this plugin currently targets Linux. File transcription works anywhere whisper-cli runs.

• PipeWire (pw-record) for audio capture — on Fedora: sudo dnf install pipewire-utils.
• whisper.cpp, built locally, with a ggml model (e.g. ggml-small.en.bin).
• Joplin 3.6 or newer.
• (Optional) A local OpenAI-compatible LLM server for transcript polishing.

Installation

From the Joplin plugin store (recommended)

Tools → Options → Plugins, search for Dictate, install, and restart Joplin.

Manual

Download the .jpl from the releases and install it via Tools → Options → Plugins → Install from file.

Configuration

Tools → Options → Dictate:

• Whisper.cpp directory — root of your whisper.cpp checkout (default: ~/whisper.cpp).
• Whisper model file — path to the ggml model (default: ~/whisper.cpp/models/ggml-small.en.bin).
• whisper-cli binary — path to the built executable (default: ~/whisper.cpp/build/bin/whisper-cli).
• Polish transcripts with LLM — off by default.
• LLM server URL — OpenAI-compatible base URL (e.g. http://localhost:1234).
• LLM model name — model identifier sent to that server.
• LLM request timeout (seconds) — advanced; default 60. Increase if your LLM server has a slow cold start.
• Use selected notebook — create the note in the notebook currently selected in Joplin.
• Default notebook ID — fallback notebook when the above is disabled.

Privacy: transcription is always local. If you enable Polish transcripts with LLM, the transcript text is sent to the LLM server URL you configure. Point it at a local endpoint (e.g. http://localhost:1234) to keep everything on your machine; a remote or plain-http URL sends note text over the network in cleartext.

Usage

Open the Dictate panel (toolbar mic button or Tools → Dictate):

1. Dictate — start recording. The button turns into Stop.
2. Pause / Resume — pause capture; you must Resume before you can Stop.
3. Cancel — discard the current recording.
4. Stop — finish, transcribe, and create the note.
5. Transcribe file… — pick a WAV file to transcribe instead of recording.

Building from source

npm install
npm run dist   # produces publish/dev.normg.joplin-dictate.jpl

License

MIT © Norm Green