AetherTune

Right Music. Every Moment.

---

Overview

AetherTune is a context-aware music recommendation system that analyzes your current mood, activity, and real-time listening context to decide whether what you're playing right now is the right fit — and if not, suggests better alternatives instantly.

Unlike traditional platforms that rely on listening history, AetherTune focuses on what you need right now: evaluating your currently playing Spotify track against your activity using a trained machine learning model, then recommending replacements when the vibe doesn't match.

---

The Problem

Modern music platforms are reactive, not intelligent:

• Recommendations are driven by past behaviour, not present context
• Users waste time scrolling to find the right music for their mood
• No platform truly understands what you're doing right now

Users often don't know what they want to listen to — they just know the current song isn't working.

---

The Solution

AetherTune introduces moment-based personalization through three pillars:

Pillar	Description
Mood-Aware	Evaluates audio features (valence, energy, acousticness) against your emotional state
Activity-Driven	Tailors recommendations to studying, driving, meditating, or exercising
Adaptive Learning	Adjusts its tolerance thresholds over time based on your feedback

---

How It Works

User selects Activity + Age
        ↓
Spotify API fetches currently playing track
        ↓
ML Model evaluates audio features against activity profile
        ↓
Dynamic Threshold check (sigmoid + feedback decay)
        ↓
  ┌─────────────────────────────────┐
  │  Match?  → "Song is suitable" │
  │  No match? → 5 Recommendations│
  └─────────────────────────────────┘
        ↓
User submits feedback → Threshold adapts over time

Adaptive Threshold System

AetherTune doesn't use a fixed confidence cutoff. Instead, it computes a sigmoid-based dynamic threshold by age group, then adjusts it through a time-decaying feedback loop:

• More feedback = more personalised threshold
• Inactivity causes gradual decay back to the base — keeping the model fresh
• Adjustments are capped at ±0.2 to prevent drift

---

Project Structure

AetherTune/
├── main.py                  # FastAPI app — /predict and /feedback endpoints
├── auth.py                  # Spotify OAuth2 login & callback
├── spotify_service.py       # Fetches currently playing track + audio features
├── recommendation.py        # Activity-based Spotify track search
├── ml_model.py              # Model training script (Random Forest)
├── database.py              # SQLite schema — feedback & tolerance tables
├── models/
│   ├── activity_model.pkl   # Trained MultiOutputClassifier
│   └── scaler.pkl           # StandardScaler for feature normalisation
├── data/
│   ├── spotify_dataset.csv          # Raw Spotify audio features dataset
│   └── spotify_dataset_labeled.csv  # Labelled dataset with activity targets
├── static/
│   ├── logo.png
│   └── style.css
└── templates/
    └── index.html           # Jinja2 web frontend

---

ML Model Details

Property	Value
Model	MultiOutputClassifier wrapping RandomForestClassifier
Labels	studying, driving, meditating, exercising
Features	danceability, energy, loudness, speechiness, acousticness, instrumentalness, liveness, valence, tempo, age
Scaler	StandardScaler
Train/Test Split	80 / 20

Label generation logic (rule-based for training):

studying   → energy < 0.6  AND speechiness < 0.4
driving    → energy > 0.5  AND tempo > 90
meditating → acousticness > 0.6 AND energy < 0.4
exercising → energy > 0.7  AND tempo > 120

---

Getting Started

Prerequisites

• Python 3.10+
• A Spotify Developer Account with a registered app

1. Clone the Repository

git clone https://github.com/your-username/AetherTune.git
cd AetherTune

2. Install Dependencies

pip install -r requirements.txt

3. Configure Environment Variables

Create a .env file in the project root:

SPOTIFY_CLIENT_ID=your_spotify_client_id
SPOTIFY_CLIENT_SECRET=your_spotify_client_secret

Never commit your .env file. Add it to .gitignore.

4. (Optional) Retrain the Model

The pre-trained model is already included in models/. To retrain:

python ml_model.py

5. Run the App

uvicorn main:app --reload

Visit http://127.0.0.1:8000 in your browser.

---

API Reference

GET /login

Redirects to Spotify's OAuth2 authorization page.

GET /callback?code=

Handles the OAuth2 callback and initialises the Spotify client.

GET /predict?activity={0-3}&age={int}

Analyses the currently playing track and returns a suitability verdict.

Activity codes:

Code	Activity
0	Studying
1	Driving
2	Meditating
3	Exercising

Example response (suitable):

{
  "Current_Track": {
    "name": "Lo-Fi Chill",
    "artist": "ChillHop Music",
    "cover": "https://...",
    "duration": 210000,
    "progress": 45000,
    "spotify_url": "https://open.spotify.com/track/..."
  },
  "message": "Song is suitable for studying ✅"
}

Example response (not suitable):

{
  "Current_Track": { ... },
  "ALERT": "Song is NOT suitable for studying ❌",
  "Recommended_Tracks": [
    { "name": "...", "artist": "...", "id": "...", "url": "..." }
  ]
}

POST /feedback?age={int}&activity={0-3}&liked={bool}

Submits user feedback to personalise the tolerance threshold.

---

🛠️ Tech Stack

Layer	Technology
Backend	FastAPI, Uvicorn
ML	Scikit-learn (Random Forest, MultiOutputClassifier)
Data	Pandas, NumPy
Spotify Integration	Spotipy
Database	SQLite
Frontend	Jinja2 Templates, HTML/CSS
Auth	Spotify OAuth2 (via Spotipy)

---

🗺️ Roadmap

• Spotify OAuth2 integration
• Real-time track evaluation via ML model
• Activity-based recommendation fallback
• Adaptive feedback loop with time decay
• Replace simulated audio features with Spotify's live Audio Features API
• User accounts and persistent profiles
• Browser extension for passive, always-on context detection
• Mood inference from time-of-day and calendar context
• Expand to Apple Music and YouTube Music

---

Authors

Name	Role
Riddhi Mishra	Co-creator
Ishan Gupta	Co-creator
Abhijeet Kumar	Co-creator
Yadvendra Tripathi	Co-creator
Avishi Sinha	Co-creator

---

Notes

• Audio features in spotify_service.py are currently simulated with random.uniform. Replace with a live call to Spotify's Audio Features endpoint for production use.
• The spotify_client in auth.py is stored in memory (global variable). For a production deployment, use a proper session or token store (e.g., Redis).

---

License

This project is licensed under the MIT License.

---

_{Built with 🎵 by the AetherTune Team}