Dansbart.se

Swedish folk music discovery platform powered by AI-driven dance style classification.

Overview

Dansbart.se is a platform for finding Swedish folk music to dance to. It uses machine learning to analyse and classify tracks by dance style — Polska, Hambo, Vals, and more — based on tempo, rhythm, and audio characteristics.

Architecture

frontend/               # React 19 + TypeScript + Tailwind CSS v4
backend/
  api/                  # Java Spring Boot REST API
  workers/
    feature/            # Python Celery worker (ML feature extraction, light tasks)
    audio/              # Python Celery worker (audio analysis via Essentia)

Service communication:

React frontend --> Java API (REST, :8000) --> PostgreSQL (jOOQ + pgvector)
                         |
                    Redis / Celery
                         |
             feature worker + audio worker

Technology Stack

Frontend: React 19, TypeScript, Tailwind CSS v4, Vitest

Backend API: Java 21, Spring Boot 3.2, jOOQ, Flyway, PostgreSQL + pgvector

Workers: Python, Celery, Redis

• Feature worker: ML feature extraction and light background tasks
• Audio worker: audio download and analysis via NeckenML Analyzer + Essentia (requires linux/amd64)

Getting Started

Prerequisites

• Docker and Docker Compose
• Git
• 4GB+ RAM recommended

Installation

1. Clone the repository:

   git clone https://github.com/svnoak/dansbart.se.git
   cd dansbart.se

2. Set up environment variables:

   cp .env.example .env
   # Edit .env with your configuration

3. Download MusiCNN models (required for audio analysis):

   cd backend/workers/audio && ./scripts/download_models.sh

4. Start the full stack:

   docker compose up

5. Access the application:

   • Frontend: http://localhost:5173 (Vite dev server)
   • API: http://localhost:8000
   • Swagger UI: http://localhost:8000/swagger-ui.html

Running Services Individually

# Frontend dev server only
docker compose up frontend

# API + dependencies
docker compose up backend db redis

# Audio worker only
docker compose up worker-audio db redis

# All workers
docker compose up worker-audio worker-feature

Development

Java API

cd backend/api
./mvnw clean install          # build
./mvnw spring-boot:run         # run locally
./mvnw test                    # tests
./mvnw generate-sources -Pgenerate-jooq  # regenerate jOOQ after schema changes

Database migrations are managed by Flyway and run automatically on startup.

React Frontend

cd frontend
npm install
npm run dev           # Vite dev server (port 5173)
npm run test:run      # Vitest
npm run build         # production build
npm run api:update    # re-export OpenAPI spec and regenerate TypeScript client

Python Workers

# Feature worker
cd backend/workers/feature
pip install -r requirements.txt
celery -A app.core.celery_app worker --loglevel=info --pool=solo -Q feature,light
pytest

# Audio worker (requires linux/amd64)
cd backend/workers/audio
pip install -r requirements.txt
celery -A app.core.celery_app worker --loglevel=info --pool=solo -Q audio
pytest

Deployment

Production images are built and pushed to GitHub Container Registry via CI on every merge to main:

• ghcr.io/svnoak/dansbart-frontend:production
• ghcr.io/svnoak/dansbart-api:production
• ghcr.io/svnoak/dansbart-feature-worker:production
• ghcr.io/svnoak/dansbart-audio-worker:production

Open Dataset

Analysis data and human feedback are publicly accessible:

• API: GET /api/export/dataset
• License: CC BY 4.0
• Example scripts: examples/

License

AGPL-3.0 — see LICENSE.

The core analysis engine, NeckenML Analyzer, is a separate MIT-licensed library.

Contributing

Issues and pull requests are welcome. See CONTRIBUTING.md for guidelines — all PRs must be linked to an issue.

---

• Website: https://dansbart.se
• NeckenML Analyzer: https://github.com/svnoak/neckenml-analyzer