Resonance

Record → Transcribe → Summarize → Create note.
A local‑first recording & meeting notes workflow for Obsidian.

Installation · Configuration · Usage · Settings · Troubleshooting

What it does

Resonance captures audio with FFmpeg, transcribes it locally using whisper.cpp, summarizes the transcript with the LLM of your choice, and creates a Markdown note in your vault, all from Obsidian.

Features

• 🎙️ Record microphone and optionally system audio
• 🧠 Local transcription via whisper.cpp
• ✨ AI summary (Gemini, OpenAI, Claude, Ollama)
• 📚 Library to review, play, download or delete recordings/transcripts

Requirements

• Obsidian Desktop ≥ 1.5
• FFmpeg installed locally
• whisper.cpp built locally (binary and model .bin)
• LLM API Key or local Ollama (for summaries)

Manual Installation

User install from release:

1. Download the zip containing manifest.json, main.js, styles.css.
2. Create <YourVault>/.obsidian/plugins/resonance/.
3. Copy the three files there and enable the plugin in Obsidian.

Configuration

Follow these steps once to fully set up recording, local transcription and summarization.

1) FFmpeg

• macOS:
  • Install with Homebrew: brew install ffmpeg
  • Typical path: /opt/homebrew/bin/ffmpeg (Apple Silicon) or /usr/local/bin/ffmpeg (Intel)
• Windows:
  • Download a static build (e.g. from the BtbN or Gyan packages)
  • Unzip to C:/ffmpeg/ so the executable is at C:/ffmpeg/bin/ffmpeg.exe
  • Optionally add C:/ffmpeg/bin to PATH, or set the full path in settings
• Linux:
  • Install via your package manager, e.g. Debian/Ubuntu: sudo apt install ffmpeg, Fedora: sudo dnf install ffmpeg

In Obsidian → Resonance → FFmpeg:

• Set “FFmpeg path” or click “Detect”. On macOS you may need to grant microphone permissions to Obsidian.

2) whisper.cpp (local transcription)

Clone and build the project, then select the whisper-cli binary.

• macOS/Linux (generic):

git clone https://github.com/ggerganov/whisper.cpp
cd whisper.cpp
cmake -S . -B build
cmake --build build -j

• The executable is typically at build/bin/whisper-cli
• Windows (CMake + MSVC):
  • Install CMake and Visual Studio Build Tools
  • From a Developer PowerShell:

git clone https://github.com/ggerganov/whisper.cpp
cd whisper.cpp
cmake -S . -B build -A x64
cmake --build build --config Release

• The executable is typically at build/bin/Release/whisper-cli.exe

In Obsidian → Resonance → Whisper:

• Set “whisper.cpp repo path”, then click “Detect” to auto‑find whisper-cli, or set it manually.
• Pick a model preset and click “Download”, or set a model .bin path manually.
• Choose the “Transcription language” (or leave Automatic).

Models (manual download): see the official ggml models, e.g. small/medium/large. Place the .bin in <repo>/models/ and select it in settings.

3) LLM

• Create an API key from Google AI Studio or the LLM of your choice.
• In Obsidian → Resonance → LLM: paste the key and pick the model.

If you want everything to run locally (without sending any data to external services), you can use Ollama as your LLM provider:

• Install Ollama following the official instructions for your operating system.
• Start Ollama (ollama serve).
• In Obsidian → Resonance → LLM:
  • Set the provider to "Ollama"
  • Enter the endpoint (default: http://localhost:11434)
  • Choose a supported model (e.g. llama3.1 or others available via ollama pull <model>)

Note: With this setup, both transcription (whisper.cpp) and note generation (LLM) are performed entirely on your computer, with no data sent externally.

4) Audio devices (mic and system audio)

Select the proper backend for your OS and pick devices from the scanned list.

• Backend:

  • macOS: avfoundation
  • Windows: dshow
  • Linux: pulse (or alsa)
  • “Automatic” chooses based on OS

• macOS system audio:

  • Install a virtual loopback driver (e.g. BlackHole 2ch)
  • Route system output to that device (or create a Multi‑Output/aggregate device if needed)
  • Click “Refresh devices”, then select your mic and the virtual device as “System audio”

• Windows system audio:

  • Install VB‑Audio Cable or VoiceMeeter
  • Set Windows output to the virtual device (or use “Stereo Mix” if available)
  • Click “Refresh devices”, pick mic and the virtual device for “System audio”

• Linux system audio:

  • With PulseAudio, choose the monitor of your output sink (e.g. alsa_output.*.monitor)
  • Click “Refresh devices”, then select mic and the monitor device

Use “Test audio config” to record a 1‑second MP3. If it fails, verify permissions, backend, and selected devices.

5) Obsidian output

• Set the “Notes folder” where Resonance will create the generated Markdown notes. If empty, the vault root is used.

6) Recording quality and limits

• Adjust sample rate, channels (mono/stereo), MP3 bitrate, and “Max recordings kept”. Older items beyond the limit are auto‑deleted (0 = infinite).

7) Done!

• You may need to restart Obsidian after changing settings.

First run checklist

• FFmpeg path set and working (test passes)
• Whisper repo path + whisper-cli set
• Model .bin selected (or downloaded)
• API key and model set
• Mic and (optional) system audio selected
• Notes folder set

Usage

1. Click the microphone icon in the ribbon to start/stop. Pick a scenario when prompted.
2. Watch the timer in the status bar.
3. When finished, a note named <Scenario> YYYY-MM-DD HH-mm.md is created in your selected folder.
4. Open the Library (audio file icon) to browse, listen, download or delete recordings and transcripts.

How it works (overview)

1. FFmpeg writes an .mp3 to <vault>/.obsidian/plugins/resonance/recordings/
2. whisper.cpp transcribes locally and writes a .txt transcript
3. The LLM of your choice summarizes the transcript
4. Resonance creates a Markdown note in your chosen folder

Privacy

• Audio never leaves your machine.
• Only the text transcript is sent for summarization—unless you use Ollama, in which case everything stays local.
• The API Key is stored locally in your vault.

Troubleshooting

• Incomplete configuration: set FFmpeg path, whisper main, model, and API key.
• No audio: verify backend/device; use Scan and the 3‑second Test.
• Noise in recordings: match the sample rate in settings with your mic.
• Empty transcription: check the .mp3 file and the model path.
• LLM error: verify API key and selected model.

Contributing

Issues and PRs are welcome. If you’d like to help with docs or UX, please open an issue to coordinate.

---

Built with Vite. This plugin is desktop‑only.