MusicAI - Intelligent Music Recommendation System

A Django-based music recommendation web application that finds similar songs using advanced audio feature analysis and machine learning.

Features

• Smart Song Search: Search for tracks by song name, artist, or album
• Similar Song Discovery: Find songs with similar audio characteristics using hybrid similarity (cosine + euclidean)
• Mood-based Recommendations: Discover songs with similar emotional qualities
• Artist Deep Dive: Explore more tracks from the same artist
• Audio Feature Analysis: Leverages 10 audio features including danceability, energy, valence, tempo, and more
• Machine Learning Models: KNN + K-Means clustering for intelligent recommendations

How It Works

Audio Features Used

• Danceability: How suitable the track is for dancing
• Energy: Intensity and activity measure
• Valence: Musical positiveness (happy/sad)
• Tempo: Speed in BPM
• Acousticness: Confidence of acoustic sound
• Instrumentalness: Predicts vocal content
• Speechiness: Presence of spoken words
• Liveness: Audience presence probability
• Loudness: Overall volume in dB
• Duration: Track length

Recommendation Algorithm

1. Feature Weighting: Prioritizes danceability, energy, and valence (1.4x weight)
2. Scaling: StandardScaler for audio features, MinMaxScaler for tempo/duration
3. Hybrid Similarity: 75% cosine similarity + 25% euclidean distance
4. Artist Boost: 5% bonus for same-artist tracks
5. Popularity Bias: 10% influence from track popularity
6. Match Score: 0-100% similarity percentage

Expected Accuracy

• Subjective Match Quality: 60-75% (typical for feature-based recommenders)
• Same-mood Clustering: ~6 mood clusters (Happy/High Energy, Warm/Relaxed, etc.)
• No deep learning required - fast and efficient

Quick Start

Prerequisites

• Python 3.10 or higher
• pip (Python package manager)
• Git

Installation

1. Clone the repository

   git clone https://github.com/dhruvindave007/MusicApp.git
   cd MusicApp

2. Create a virtual environment

   # Windows
   python -m venv venv
   venv\Scripts\activate
   
   # macOS/Linux
   python3 -m venv venv
   source venv/bin/activate

3. Install dependencies

   pip install -r requirements.txt

4. Set up the database

   python manage.py migrate

5. Train the recommendation models

   python manage.py train_recommender

   This will:

   • Load music/data/spotify_tracks.csv (114,000+ tracks)
   • Extract and scale audio features
   • Train KNN model (15 neighbors)
   • Train K-Means clustering (6 mood clusters)
   • Save models to music/artifacts/

6. Run the development server

   python manage.py runserver

7. Open your browser Navigate to http://127.0.0.1:8000/

📁 Project Structure

MusicApp/
│
├── manage.py                          # Django management script
├── requirements.txt                   # Python dependencies
├── db.sqlite3                         # SQLite database
│
├── MusicAI/                          # Django project settings
│   ├── settings.py
│   ├── urls.py
│   └── wsgi.py
│
├── music/                             # Main application
│   ├── views.py                       # Request handlers
│   ├── urls.py                        # URL routing
│   ├── models.py                      # Database models
│   ├── utils.py                       # Similarity calculations
│   ├── recommender.py                 # ML recommendation engine
│   │
│   ├── data/
│   │   └── spotify_tracks.csv         # Dataset (114k+ tracks)
│   │
│   ├── artifacts/                     # Trained models
│   │   ├── knn.joblib                 # KNN model
│   │   ├── kmeans.joblib              # K-Means clustering
│   │   ├── scalers.joblib             # Feature scalers
│   │   ├── cluster_names.joblib       # Mood cluster labels
│   │   └── tracks.parquet             # Processed data
│   │
│   ├── management/commands/
│   │   └── train_recommender.py       # Model training command
│   │
│   └── templates/music/               # HTML templates
│       ├── search.html
│       ├── song_detail.html
│       └── results.html
│
└── static/                            # CSS, JS, images
    └── style.css

Usage

Search for Songs

1. Go to the home page
2. Enter a song name, artist, or album
3. Click search or press Enter

View Similar Songs

1. Click on any song from search results
2. See the top 10 most similar songs with match percentages
3. Explore "More from this Artist" section

Understanding Match Scores

• 90-100%: Extremely similar (same mood, tempo, energy)
• 75-89%: Very similar (comparable vibe)
• 60-74%: Moderately similar (some shared characteristics)
• Below 60%: Loosely related

🛠️ Advanced Usage

Custom Training Parameters

# Train with custom settings
python manage.py train_recommender \
  --csv music/data/spotify_tracks.csv \
  --out music/artifacts \
  --neighbors 20 \
  --clusters 8

Using Your Own Dataset

1. Prepare a CSV with these columns:

   • Required: track_name, artist_name, id (Spotify track ID)
   • Audio features: danceability, energy, loudness, speechiness, acousticness, instrumentalness, liveness, valence, tempo, duration_ms
   • Optional: album_name, popularity, year

2. Place it in music/data/

3. Train the model:

   python manage.py train_recommender --csv music/data/your_dataset.csv

Testing

Run accuracy evaluation tests:

python manage.py test music

This validates:

• CSV loading and cleaning
• Feature extraction and scaling
• Model training and prediction
• Similarity calculation accuracy
• Recommendation diversity

🔧 Configuration

Edit MusicAI/settings.py to customize:

• Database settings
• Static files location
• Allowed hosts
• Debug mode

Dependencies

Core packages:

• Django 5.2.4: Web framework
• pandas 2.3.1: Data manipulation
• scikit-learn 1.7.1: Machine learning models
• numpy 2.3.2: Numerical computing
• joblib 1.5.1: Model persistence

See requirements.txt for complete list.

Troubleshooting

"ModuleNotFoundError: No module named 'pandas'"

• Ensure virtual environment is activated
• Run pip install -r requirements.txt

"No such file or directory: spotify_tracks.csv"

• Verify music/data/spotify_tracks.csv exists
• Check file path in training command

"No artifacts found" error on startup

• Run python manage.py train_recommender first
• Models must be trained before running the server

Search returns no results

• Dataset might not contain the artist/song
• Try partial names or alternative spellings
• Check if CSV loaded correctly

Deployment

For production deployment:

1. Set DEBUG = False in settings.py
2. Configure ALLOWED_HOSTS
3. Use proper database (PostgreSQL/MySQL)
4. Set up static file serving
5. Use production WSGI server (Gunicorn/uWSGI)
6. Configure SECRET_KEY via environment variable

License

This project is for educational purposes.

Author

Dhruvin Dave

• GitHub: @dhruvindave007

🙏 Acknowledgments

• Spotify for audio feature data
• scikit-learn community
• Django framework

📈 Future Enhancements

• Add user accounts and favorite tracking
• Playlist generation
• Real-time Spotify API integration
• Deep learning models (neural collaborative filtering)
• Genre-based filtering
• Export recommendations to Spotify
• Audio preview integration
• A/B testing framework for algorithm improvements

---

Note: This system uses audio feature similarity, not collaborative filtering. It finds songs that sound similar, not songs that similar users liked.