LionScribe is a web application built with Flask, SQLite, and styled using the Tailwind CSS CDN. It records live microphone audio during meetings, transcribes speech continuously with OpenAI Whisper (base model), exports .mp3 audio and .txt transcripts, and uses Llama 3 to automatically generate structured meeting notes and a formatted PDF summary report.
- New Meeting Creation on Launch:
- The home page opens directly with a clean "Create a New Meeting" form.
- Big Start Meeting Button:
- Interactive live room with a large, accessible Start button to trigger microphone recording and live speech capture.
- Live OpenAI Whisper Transcription:
- Continuous audio stream transcription using OpenAI Whisper (
basemodel — lightweight, fast, and balanced accuracy).
- Continuous audio stream transcription using OpenAI Whisper (
- Persistent Storage:
- Audio: Finalized
.mp3audio file saved instorage/audio/. - Transcripts: Formatted
.txttranscript file saved instorage/transcripts/. - Database: SQLite database (
instance/lionscribe.db) storing meeting metadata, timestamps, durations, and segments.
- Audio: Finalized
- Llama 3 Meeting Notes & PDF Summary:
- Triggered immediately on clicking "End Meeting".
- Summarizes the transcript into Executive Summary, Key Discussion Points, Decisions Made, and Action Items with checkboxes.
- Generates a styled, downloadable Meeting Summary PDF report via ReportLab in
storage/summaries/.
- Tailwind CSS CDN UI:
- Modern, responsive dark-mode interface with live audio wave animations, timer, audio player, and instant download buttons.
lionscribe/
├── app.py # Flask routes, WebSocket/API endpoints, file delivery
├── config.py # Configuration for directories, Whisper & Llama 3
├── models.py # SQLite models (Meeting, TranscriptSegment)
├── requirements.txt # Python dependencies
├── .env.example # Example environment variables
├── .gitignore # Git ignore rules
├── services/
│ ├── __init__.py
│ ├── audio_service.py # Audio chunk processing & MP3 encoding via ffmpeg
│ ├── whisper_service.py # OpenAI Whisper model loader & .txt export
│ ├── llama_service.py # Llama 3 notes summarizer (Ollama/Groq/OpenAI/Fallback)
│ └── pdf_service.py # ReportLab PDF generator for meeting summaries
├── storage/ # Output directory for saved artifacts
│ ├── audio/ # Saved .mp3 audio recordings
│ ├── transcripts/ # Saved .txt meeting transcripts
│ └── summaries/ # Saved .pdf meeting summary reports
├── static/
│ └── js/
│ └── meeting.js # Web audio recording, live streaming & UI updates
└── templates/
├── base.html # Tailwind CSS CDN layout & navigation
├── index.html # "Create a New Meeting" form & past meeting archive
├── meeting.html # Live room with Big Start Button & live Whisper stream
└── summary.html # Completed meeting notes, audio player & download links
- Python 3.10+ (Virtual environment available at
~/scribe/bin) ffmpeg(installed on system)
~/scribe/bin/pip install -r requirements.txt~/scribe/bin/python app.pyOpen your browser and navigate to: http://localhost:5000
LionScribe uses balanced, accessible models:
| Component | Default Model | Customization |
|---|---|---|
| Speech Transcription | OpenAI Whisper base |
Set WHISPER_MODEL=small or tiny in .env |
| Meeting Notes Summary | Llama 3 (8B) | Supports Ollama (llama3:8b), Groq (GROQ_API_KEY), or built-in fallback |
To configure remote or local Llama 3 endpoints, copy .env.example to .env:
cp .env.example .envAnd adjust LLAMA_MODEL, LLAMA_API_BASE, or GROQ_API_KEY.