Moodsic
- STACK: NextJS, TensorFlow
- YEAR: 2024
- Next JS
- TensorFlow.js
- Spotify Web API
- Facial Emotion Detection
- Custom Mapping Algorithm
- 8,000+ Labeled Songs
Moodsic is a web-based application that generates personalized Spotify playlists based on your current emotional state. The app uses a TensorFlow-powered facial emotion detection model to classify your mood from a selfie, then feeds those emotion values into a custom mapping algorithm that defines the parameters for Spotify's recommendation engine. This makes sure that every playlist not only matches how you feel, but stays as musically coherent as possible from start to finish.
To support meaningful recommendations, a dataset of over 8,000 manually labeled songs was compiled and used as seeds—uniting emotion confidence values with quantifiable song attributes like energy, tempo, and valence. The result is a simple, interactive experience that converts your raw emotional input into dynamic, tailored music discovery.
Under the hood
Moodsic uses TensorFlow.js running in-browser to perform real-time facial emotion detection from a user-submitted webcam image. The model outputs a confidence-weighted distribution across predefined emotion classes (e.g., happy, sad, angry, neutral), which are then passed into a custom-built mapping function written in JavaScript. This function translates emotional weights into specific Spotify audio feature parameters—such as valence, energy, danceability, and tempo—used by the Spotify Web Recommendations API to fetch track suggestions.
To guide the API and maintain playlist coherence, Moodsic pulls from a local seed bank of 8,000+ tracks, preprocessed and labeled with emotion-related metadata using the Spotify Track Audio Features endpoint and manually curated associations. The result is a playlist uniquely tuned to the user's emotional state.
