Skip to content

Co-vengers/SoundNote

BranchesTags

Repository files navigation

SoundNote

A production-ready Django application for transcribing video and audio files using OpenAI's Whisper model, with containerization, async task processing, and comprehensive security features.

Overview

Video Transcriber is a modern, full-featured web application for transcribing video and audio content. It leverages OpenAI's Whisper, a state-of-the-art speech recognition model, combined with Docker containerization, PostgreSQL, Celery async tasks, and strict security controls to provide a reliable, scalable platform for media transcription.

Key Features

Core Transcription

  • Multiple File Formats: MP4, MPEG, MOV, AVI, WebM, OGG, MP3, WAV, FLAC
  • Long Video Support: Automatic chunking for videos over 15 minutes
  • Configurable Models: Tiny, Base, Small (default), Medium, Large (accuracy vs. speed tradeoff)
  • Timestamp Segments: Per-utterance transcripts with precise timing
  • Multi-format Export: Download as TXT (plaintext) or SRT (subtitles for video players)

Architecture & Reliability

  • Docker Containerization: Reproducible deployments with docker-compose
  • PostgreSQL Database: Production-grade data persistence and concurrent access
  • Celery Task Queue: Asynchronous transcription processing with Redis broker
  • Graceful Error Handling: Automatic recovery from worker crashes and deleted records
  • Persistent Model Cache: Pre-downloaded Whisper models survive container restarts

Security

  • User Authentication: Registration, login, password reset with email
  • Authorization: Users can only access their own videos (prevents IDOR attacks)
  • Brute-force Protection: Lock accounts after 5 failed login attempts (24-hour cooldown)
  • CSRF Protection: All forms include CSRF tokens
  • Secure Cookies: HTTPS-only cookies in production
  • Input Validation: File type, size, and content-type validation

User Experience

  • Real-time Status Updates: AJAX polling shows transcription progress
  • Pagination: Video list with 6 items per page
  • Bootstrap UI: Responsive, mobile-friendly interface
  • User-friendly Titles: Video filenames automatically used as titles
  • Progress Tracking: Visual status indicators (Pending, Processing, Completed, Failed)

Installation

Prerequisites

  • Docker & Docker Compose (recommended for production)
  • Python 3.12+ (for local development)
  • FFmpeg (for audio extraction from video files)

Quick Start with Docker (Recommended)

  1. Clone the repository:

    git clone https://github.com/Co-vengers/video_transcriber.git
cd video_transcriber

  2. Create environment configuration:

    cp video_transcriber/.env.example video_transcriber/.env
# Edit video_transcriber/.env with your settings

  3. Build and start services:

    docker-compose up --build

  4. Access the application:

  5. Create admin user (optional, in another terminal):

    docker-compose exec -T web python manage.py createsuperuser

Local Development Setup

If you prefer running locally without Docker:

  1. Clone the repository:

    git clone https://github.com/Co-vengers/video_transcriber.git
cd video_transcriber

  2. Create a virtual environment:

    python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

  3. Install dependencies:

    pip install -r requirements.txt

  4. Set up environment variables:

    cp video_transcriber/.env.example video_transcriber/.env
# Edit .env with development settings (use SQLite for local dev)

  5. Run migrations:

    python manage.py migrate

  6. Create a superuser:

    python manage.py createsuperuser

  7. Start Redis (in separate terminal):

    # Using Homebrew on macOS:
brew services start redis
# Or run directly:
redis-server

  8. Start Celery worker (in separate terminal):

    cd video_transcriber
celery -A video_transcriber worker --loglevel=info

  9. Start Django development server:

    cd video_transcriber
python manage.py runserver

  10. Access at http://localhost:8000

Usage

Web Interface Workflow

1. Register / Login

  • Click "Register" to create a new account or "Login" with existing credentials
  • Password reset available via email link
  • Brute-force protection: Account locks after 5 failed attempts (24-hour cooldown)

2. Upload Video

  • Click "Upload Video" from main menu
  • Select a video or audio file (max 500 MB)
  • Supported formats: MP4, MPEG, MOV, AVI, WebM, OGG, MP3, WAV, FLAC
  • Choose Whisper model size:
    • Tiny: Fastest, ~39M parameters, basic accuracy
    • Base: Balanced, ~74M parameters
    • Small: Default, ~244M parameters, good accuracy/speed tradeoff
    • Medium: Slower, ~769M parameters, better accuracy
    • Large: Slowest, ~1.5B parameters, highest accuracy
  • Click "Upload" to start transcription

3. Monitor Progress

  • Videos appear in "Videos" list with status badge
  • Status updates in real-time (Pending → Processing → Completed/Failed)
  • Progress visible without page refresh

4. Download Transcripts

Once transcription completes, download in multiple formats:

  • TXT: Plain text transcript (copy/paste friendly)
  • SRT: SubRip format with timestamps (import into video players)
    • Format: HH:MM:SS,mmm --> HH:MM:SS,mmm
    • Compatible with VLC, YouTube, browser video players

5. Manage Videos

  • View all your transcribed videos with status
  • Click video title to see full transcript and segments
  • Delete videos to free up storage
  • Videos are private (only you can see your transcripts)

Admin Interface

Access Django admin at /admin:

  • Manage user accounts
  • View/filter transcription jobs by status and date
  • Search videos by title or username
  • Monitor transcription history

API Endpoints (for developers)

Endpoint Method Purpose
/ GET Upload form
/videos/ GET List user's videos (paginated)
/videos/<id>/ GET View transcript and segments
/videos/<id>/status/ GET JSON status (for AJAX polling)
/videos/<id>/download/<fmt>/ GET Download transcript (fmt: txt or srt)
/videos/<id>/delete/ POST Delete video
/register/ GET/POST User registration
/login/ GET/POST User login
/logout/ POST User logout
/password-reset/ GET/POST Password reset flow

Architecture

Technology Stack

  • Web Framework: Django 5.1.7 (Python web framework)
  • Application Server: Gunicorn 23.0.0 (production WSGI server)
  • Database: PostgreSQL 16 (production relational database)
  • Message Broker: Redis 7 (in-memory message queue)
  • Task Queue: Celery 5.4.0 (asynchronous job processing)
  • ML/AI Engine: OpenAI Whisper 20240930 (speech-to-text)
  • Containerization: Docker & docker-compose (reproducible deployments)
  • Frontend: Bootstrap 5 (responsive CSS framework)

System Architecture

┌─────────────────────────────────────────────────────────────┐
│                      Docker Compose                          │
├─────────────────────────────────────────────────────────────┤
│                                                               │
│  ┌─────────────────┐  ┌──────────────┐  ┌──────────────┐   │
│  │   Web Service   │  │   Worker     │  │   Database   │   │
│  │   (Gunicorn)    │  │  (Celery)    │  │(PostgreSQL)  │   │
│  │   2 workers     │  │ concurrency=4│  │              │   │
│  └────────┬────────┘  └──────┬───────┘  └──────────────┘   │
│           │                   │                              │
│           └──────────┬────────┘                              │
│                      │                                       │
│              ┌───────▼────────┐                              │
│              │  Redis Broker  │                              │
│              │  (Task Queue)  │                              │
│              └────────────────┘                              │
│                                                               │
│            Persistent Volumes:                               │
│            - whisper_cache: /cache/whisper (461MB model)     │
│            - postgres_data: PostgreSQL data                  │
│                                                               │
└─────────────────────────────────────────────────────────────┘

Data Flow

  1. User uploads video → Web service saves to /media/videos/
  2. Task is queued → Gunicorn sends process_transcription task to Redis
  3. Worker picks up task → Celery worker loads Whisper model (cached)
  4. Chunked transcription → Long videos split into 10-minute chunks
  5. Results saved → Transcripts and segments stored in PostgreSQL
  6. Frontend updates → AJAX polling shows real-time status
  7. User downloads → Download as TXT or SRT subtitle format

Project Structure

video_transcriber/
├── Dockerfile              # Container image definition
├── docker-compose.yml      # Multi-service orchestration
├── requirements.txt        # Python dependencies
├── CHANGES_DOCUMENTATION.md # Full changelog and rationale
│
├── video_transcriber/      # Django project config
│   ├── settings.py         # Django configuration
│   ├── urls.py             # Main URL routing
│   ├── wsgi.py             # WSGI application
│   ├── celery.py           # Celery configuration
│   └── __init__.py         # Celery import
│
├── transcription/          # Django app (main logic)
│   ├── models.py           # Video model with ownership
│   ├── views.py            # All HTTP view handlers
│   ├── urls.py             # App URL patterns
│   ├── forms.py            # VideoUploadForm with validation
│   ├── tasks.py            # Celery transcription task
│   ├── admin.py            # Django admin configuration
│   ├── utils.py            # Whisper transcription utilities
│   ├── exports.py          # TXT/SRT export functions
│   │
│   ├── management/
│   │   └── commands/
│   │       └── requeue_stale_transcriptions.py  # Stale task recovery
│   │
│   ├── migrations/         # Database schema versions
│   │   ├── 0001_initial.py
│   │   ├── 0002_video_user.py
│   │   ├── 0003_alter_video_file_alter_video_user.py
│   │   ├── 0004_video_status.py
│   │   ├── 0005_video_segments.py
│   │   └── 0006_fix_upload_to_path.py
│   │
│   ├── templates/          # HTML templates
│   │   ├── base.html       # Navigation, Bootstrap layout
│   │   ├── upload.html     # Video upload form
│   │   ├── video_list.html # Paginated video gallery
│   │   ├── video_detail.html # Transcript viewer, download
│   │   └── auth/           # Authentication templates
│   │       ├── login.html
│   │       ├── register.html
│   │       ├── password_reset.html
│   │       └── lockout.html
│   │
│   ├── static/             # CSS, JS, fonts
│   │   └── transcription/
│   │       └── favicon.svg
│   │
│   └── tests.py            # Unit tests
│
├── media/                  # User uploads (not in repo)
│   └── videos/
│
├── .env.example            # Environment template
├── .python-version         # Python 3.12
├── .gitignore              # Excluded files
└── README.md               # This file

Security Features

Authentication & Authorization

  • ✅ User registration with password validation
  • ✅ Secure password reset via email
  • ✅ CSRF tokens on all forms
  • ✅ Permission checks (users can only access their own videos)
  • ✅ IDOR prevention (returns 404 if accessing others' content)

Attack Prevention

  • ✅ Brute-force protection (5 failed attempts → 24-hour lockout)
  • ✅ HTTPS-only cookies in production
  • ✅ Secure cookie flags (HttpOnly, SameSite)
  • ✅ Security headers (X-Frame-Options=DENY, X-Content-Type-Options=nosniff)
  • ✅ SQL injection prevention (parameterized queries via ORM)

Data Protection

  • ✅ Environment-based secrets (not in code)
  • ✅ File type validation (extension + MIME type)
  • ✅ File size limits (500 MB max)
  • ✅ Input sanitization on all forms
  • ✅ Secure file storage outside web root

Worker Reliability

  • ✅ Graceful error handling (no crashes on deleted records)
  • ✅ Task timeouts (1 hour hard, 55-minute soft)
  • ✅ Auto-requeue on worker loss
  • ✅ Automatic stale task recovery on startup
  • ✅ Non-root worker process (nobody:nogroup)

Environment Configuration

Docker Setup (.env file)

# Security
SECRET_KEY=your-secret-key-here

# Debug Mode (False in production)
DEBUG=True

# Allowed Hosts
ALLOWED_HOSTS=localhost,127.0.0.1,0.0.0.0

# Database
DB_ENGINE=postgres           # or 'sqlite' for development
DB_NAME=video_transcriber
DB_USER=postgres
DB_PASSWORD=postgres
DB_HOST=db                   # Docker service name
DB_PORT=5432

# Message Broker & Results
CELERY_BROKER_URL=redis://redis:6379/0
CELERY_RESULT_BACKEND=redis://redis:6379/0

# Stale Task Recovery
STALE_PROCESSING_MINUTES=45   # How old before marking as stale
RECOVERY_MODEL_SIZE=small      # Model to use for requeue

Email Configuration (Optional)

For password reset emails:

EMAIL_BACKEND=django.core.mail.backends.smtp.EmailBackend
EMAIL_HOST=smtp.gmail.com
EMAIL_PORT=587
EMAIL_USE_TLS=True
EMAIL_HOST_USER=your-email@gmail.com
EMAIL_HOST_PASSWORD=your-app-password
DEFAULT_FROM_EMAIL=noreply@example.com

Performance & Deployment

Performance Tips

  • GPU Support: Using CUDA GPU significantly improves transcription speed
  • Model Selection:
    • tiny (39M params): 5-10x faster, lower accuracy
    • base (74M params): 2-5x faster, decent accuracy
    • small (244M params): Default, good balance
    • medium (769M params): Slower, better accuracy
    • large (1.5B params): Very slow, best accuracy
  • Long Videos: Automatically chunked (no manual splitting needed)
  • Batch Processing: Queue multiple uploads for parallel processing

Scaling

  • Horizontal Scale: Add more worker containers for higher throughput
  • Concurrent Limit: Currently --pool=solo --concurrency=4 (adjust as needed)
  • Database: PostgreSQL handles concurrent access safely
  • Cache: Model stays in memory, re-downloads on restart (persists across container restarts via volume)

Production Checklist

  • Set DEBUG=False in .env
  • Generate strong SECRET_KEY (use python -c "from django.core.management.utils import get_random_secret_key; print(get_random_secret_key())")
  • Configure ALLOWED_HOSTS with your domain
  • Set up HTTPS (nginx reverse proxy or cloud provider)
  • Configure email backend for password resets
  • Use strong database password
  • Enable PostgreSQL backups
  • Monitor worker health and logs
  • Set up log aggregation (e.g., CloudWatch, ELK)
  • Configure resource limits in docker-compose

Troubleshooting

Common Issues

Videos stuck in "Processing" status:

# Manually requeue stale videos
docker-compose exec -T web python manage.py requeue_stale_transcriptions --minutes=0

Worker not picking up tasks:

# Check Celery worker logs
docker-compose logs -f worker

# Restart worker
docker-compose restart worker

Database connection errors:

# Check PostgreSQL is healthy
docker-compose exec db pg_isready -U postgres

# View database service logs
docker-compose logs db

Redis connection issues:

# Verify Redis is accessible
docker-compose exec redis redis-cli ping
# Should return: PONG

Model download stuck:

  • First run downloads 461MB model (~70 seconds)
  • Model is cached in persistent volume whisper_cache:/cache/whisper
  • Subsequent runs load from cache (~5 seconds)

Testing

Run unit tests:

# With Docker
docker-compose exec -T web python manage.py test

# Locally
python manage.py test transcription

Test coverage includes:

  • ✅ Authorization (IDOR prevention)
  • ✅ AJAX status endpoint
  • ✅ Chunked transcription merging
  • ✅ Task deletion safety
  • ✅ Form validation

Contributing

Contributions are welcome! Please follow these steps:

  1. Fork the repository
  2. Create a feature branch: git checkout -b feat/amazing-feature
  3. Make your changes and add tests if applicable
  4. Commit with conventional messages: git commit -m "feat: description"
  5. Push to your fork: git push origin feat/amazing-feature
  6. Open a Pull Request against main branch

Development Workflow

# Create local development environment
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

# Set up local .env with SQLite
cp video_transcriber/.env.example video_transcriber/.env
# Edit to use: DB_ENGINE=sqlite, remove CELERY_* vars for testing

# Run migrations
python manage.py migrate

# Run tests
python manage.py test

# Start development servers (in separate terminals)
redis-server
celery -A video_transcriber worker --loglevel=info
python manage.py runserver

License

This project is licensed under the MIT License - see the LICENSE file for details.

Changelog

See CHANGES_DOCUMENTATION.md for detailed documentation of all changes, including:

  • Infrastructure & containerization improvements
  • Security hardening details
  • Task reliability enhancements
  • Feature additions and rationale
  • Migration guide from previous version

See IMPROVEMENTS.md for planned future enhancements.

Acknowledgments

  • OpenAI Whisper — State-of-the-art speech recognition
  • Django — Web framework
  • Celery — Async task queue
  • PostgreSQL — Reliable database
  • Bootstrap — Responsive CSS framework
  • All contributors who have helped build and improve this tool

Contact & Support

Version

  • Current Version: 2.0.0 (Production-ready)
  • Python: 3.12+
  • Django: 5.1.7
  • Release Date: March 2026

About

Video to Text converter

Topics

django openai-whisper

Resources

Readme

License

MIT license
Activity

Stars

0 stars

Watchers

1 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages