STT & LLM

Local speech-to-text transcription and LLM-powered text operations for Obsidian.

Features

• Speech-to-Text Recording: Transcribe voice to text directly into your notes using a local Whisper-based server
• Summarize Selection: Generate summaries of selected text using your local LLM
• Auto-Tag Notes: Automatically generate relevant tags for your notes
• Custom Prompts: Send selected text to your LLM with custom instructions

All processing happens locally on your machine—no cloud services required.

Requirements

STT Server (Required for voice transcription)

This plugin requires a companion server for speech-to-text functionality:

stt-server - A local WebSocket server using OpenAI's Whisper model.

Quick Setup with Docker (Recommended)

The easiest way to run the STT server:

# CPU version
docker run -p 8765:8765 ghcr.io/kevinraymond/stt-server:cpu

# GPU version (NVIDIA)
docker run --gpus all -p 8765:8765 ghcr.io/kevinraymond/stt-server:gpu

The server will be available at ws://localhost:8765.

Manual Setup

If you prefer not to use Docker:

Prerequisites: Python 3.10+, uv, ffmpeg

git clone https://github.com/kevinraymond/stt-server.git
cd stt-server
uv sync
uv run obsidian-stt-server --auto

See the stt-server repository for advanced configuration.

LLM API (Optional, for text operations)

For summarization, auto-tagging, and custom prompts, you need a local LLM server with an OpenAI-compatible API:

• LM Studio
• Ollama
• llama.cpp server
• Any other OpenAI-compatible endpoint

Installation

From Community Plugins (Recommended)

1. Open Obsidian Settings
2. Go to Community Plugins and disable Safe Mode
3. Click Browse and search for "STT & LLM"
4. Install and enable the plugin

Manual Installation

1. Download main.js, manifest.json, and styles.css from the latest release
2. Create a folder obsidian-stt-llm in your vault's .obsidian/plugins/ directory
3. Copy the downloaded files into this folder
4. Reload Obsidian and enable the plugin in Settings → Community Plugins

Configuration

Open Settings → STT & LLM to configure:

Speech-to-Text

• Server URL: WebSocket URL for the STT server (default: ws://localhost:8765)

LLM Settings

• API URL: Your LLM server endpoint (default: http://localhost:11434)
• Model: The model to use (e.g., llama3.2, mistral)
• Summarization Prompt: Template for summarization requests
• Tagging Prompt: Template for auto-tag generation
• Custom Prompt Default: Default prompt for custom operations

Usage

Voice Recording

1. Click the microphone icon in the ribbon, or use the command "Toggle STT recording"
2. Speak into your microphone
3. Click again to stop—the transcription will be inserted at your cursor

LLM Operations

Once LLM is configured, additional controls appear:

• Summarize: Select text → right-click → "Summarize Selection" (or use command palette)
• Auto-Tag: Open a note → right-click → "Generate Tags for Note"
• Custom Prompt: Select text → right-click → "Send with Custom Prompt"

Commands

Command	Description
Toggle STT recording	Start/stop voice recording
Summarize selection	Summarize the selected text
Generate tags for current note	Auto-generate tags based on note content
Send selection with custom prompt	Process selection with a custom instruction

Network Usage Disclosure

This plugin makes local network connections only:

• STT Server: WebSocket connection to ws://localhost:8765 (configurable) for speech-to-text transcription
• LLM API: HTTP requests to http://localhost:11434 (configurable) for text operations

No data is sent to external cloud services. All processing occurs on your local machine.

License

MIT

Support

If you find this plugin useful, consider sponsoring the development.