Nuntius: Live Transcription for Obsidian

Live speech-to-text and translation inside Obsidian.

Transcribe your microphone, system audio, or an audio file using a cloud provider (Soniox) or a local engine (whisper.cpp). You see partial text instantly and can copy it with one command. Clean final text is written into your note.

Desktop only. macOS, Windows, Linux.

What makes it different

• Copy text before it is final. Partial (non-final) words show up the moment they are recognized, and you can select and copy them right away. The copy box is built to behave like normal text, so grabbing a half-finished sentence just works.
• Keep editing the note while it writes. Final text streams straight into your .md file, but the note stays fully editable. Fix a typo or jump the cursor mid-session and the live writing keeps going without breaking your edit or the document.

Features

• Live transcription. A side panel shows partial and final text as you speak, with speaker labels, while clean final text goes into the active note.
• Multiple audio sources. Microphone, system audio (via a loopback device like BlackHole), or both.
• File and embed transcription. Import an audio file, or transcribe audio embedded in a note (![[recording.m4a]]) with a per-embed Transcript button.
• Translation. One-way or two-way, alongside the transcript (depends on the provider).
• Pluggable providers. Soniox (realtime cloud) and local whisper.cpp out of the box. An OpenAI provider is stubbed for a future release.
• Quality modes for the local engine: live (fast, streaming), full (re-transcribe on Stop for best accuracy), or both.

Requirements

• Obsidian 1.5.0 or newer (desktop).
• Soniox provider: a Soniox API key.
• Local provider: a built whisper.cpp binary (whisper-cli).
• System audio capture: a loopback audio device (e.g. BlackHole on macOS).

Installation

Manual

1. Download main.js, manifest.json, and styles.css from the latest release.
2. Copy them into your vault under <vault>/.obsidian/plugins/obsidian-nuntius/.
3. Reload Obsidian and enable Nuntius in Settings, Community plugins.

From source

npm install
npm run build        # type-check + bundle to main.js

Usage

1. Open Settings, Nuntius and set your provider (API key for Soniox, or the binary path for local whisper.cpp), audio source, and language.
2. Open the note you want to write into.
3. Run Nuntius: Start/stop live transcription from the command palette, or click the microphone ribbon icon to open the panel.

Capturing system audio

macOS (and Windows) cannot record system output directly, so you need a virtual loopback device. On macOS install BlackHole (brew install blackhole-2ch); on Windows use an equivalent such as VB-CABLE.

1. In Audio MIDI Setup, create a Multi-Output Device that includes both your real output (speakers or headphones) and BlackHole 2ch. Set them all to the same sample rate (48000 Hz) and enable Drift Correction on the non-primary devices.
2. Set that Multi-Output Device as your system output, so you still hear the audio while a copy is routed into BlackHole.
3. In Nuntius settings, set Input to System audio and System device to BlackHole 2ch. Use the Both source if you also want your microphone.

Commands

Command	What it does
Start/stop live transcription	Start or stop a live session into the active note.
Copy live transcript to clipboard	Copy the current live transcript text.
Set transcript insertion point at cursor	Pin where final text is written.
Open live transcript panel	Reveal the side panel.
Transcribe audio file	Pick an audio file from disk and transcribe it.
Transcribe audio embedded in current note	Transcribe ![[audio]] embeds in the note.

License

Copyright © 2026 denysk0. All rights reserved.

This project is not open source. No license is granted to use, copy, modify, or distribute the code without prior written permission.