Local Whisper Transcriber

Record audio in Obsidian, transcribe it locally with whisper.cpp, and optionally generate a structured AI summary.

Why this plugin exists

Most voice-note workflows are fragmented or cloud-only. This plugin is built for users who want:

• fast capture directly inside Obsidian
• local-first transcription on their own machine
• reliable meeting notes with minimal manual steps

It is designed to make spoken notes as easy as typed notes while keeping core transcription local.

What it does

• Records microphone audio from inside Obsidian.
• Transcribes with local whisper.cpp (ffmpeg + whisper-cli).
• Shows clear live status:
  • red pulsing mic while recording
  • recording status indicator
  • conversion/transcription/summarization progress bars
• Inserts output into the current note, or creates a note when needed.
• Optionally summarizes transcripts via Azure OpenAI Responses API.
• Optionally auto-records Teams calls (macOS), opens a dedicated meeting note at start, and can auto-stop at call end.

Requirements

• Obsidian desktop app (plugin is desktop-only).
• macOS.
• Local tools:
  • brew install ffmpeg
  • brew install whisper-cpp
• A local Whisper model file, for example:
  • ggml-model-whisper-small.bin

Quick Start

1. Open plugin settings.
2. Click Apply whisper.cpp preset.
3. Update model path in the command template if needed.
4. Click Check current command.
5. Start recording and stop to transcribe.

Default command template

ffmpeg -y -i {input} -ar 16000 -ac 1 {outputBase}.wav && whisper-cli -m '/path/to/ggml-model-whisper-small.bin' -f {outputBase}.wav -otxt -of {outputBase}

Placeholders

• {input}: absolute path to recorded audio file
• {outputDir}: output directory
• {outputBase}: output path without extension
• {outputTxt}: expected transcript .txt path

Teams Auto-Record (optional, macOS)

Enable Auto-record when Teams call starts to detect call windows and begin recording automatically.

• Creates and opens a new meeting note immediately when a call starts.
• Automatically starts recording when a call is detected.
• Automatically stops recording and transcribes when the call ends (if Auto-stop when Teams call ends is enabled).
• Meeting note title format:
  • YYYY-MM-DD <Teams call title>
• Supports:
  • Auto-stop when Teams call ends
  • configurable detection interval
  • configurable call keywords

If detection is blocked, grant Obsidian Accessibility access: System Settings -> Privacy & Security -> Accessibility.

OpenAI Summary (optional)

Enable Summarize with OpenAI and configure:

• Azure OpenAI URL (full Responses API URL, including api-version)
• Azure API key (Bearer token)
• OpenAI model (for example gpt-5.1-codex-mini)
• Summary prompt
• Max output tokens

The summary is inserted above the transcript, with specific insights called out.

Privacy and data flow

• Audio recording and transcription run locally on your machine.
• Audio/transcript are sent to OpenAI only if Summarize with OpenAI is enabled.
• Teams detection reads Teams window titles on macOS for automation; no call audio is captured unless recording is started.

Troubleshooting

• command not found:
  • install dependencies (ffmpeg, whisper-cpp)
  • use Check current command in settings
• No transcript output:
  • verify model path in command template
• Repeated macOS permission prompt for Teams:
  • allow Obsidian in Accessibility settings, then toggle Teams auto-record off/on

Build

npm install
npm run build

Manual install (development)

1. Copy to vault plugin folder:
   • <vault>/.obsidian/plugins/local-whisper-transcriber/
2. Ensure these files exist:
   • manifest.json
   • main.js
   • styles.css
3. Enable Local Whisper Transcriber in Obsidian Community Plugins.