Emotion-Driven Spotify Clone
Music recommendation system based on facial emotion recognition.
CNNReactPython
Concept
A music player that knows how you feel. Instead of manually selecting a playlist, the app scans your face, determines your emotion, and cues the perfect track.
Vision Pipeline
We custom-built a simplistic Convolutional Neural Network (CNN) trained on the FER-2013 dataset (Facial Emotion Recognition).
Model Arch
- 4 Conv2D Layers (32, 64, 128, 128 filters)
- MaxPolling and Dropout (0.25) to prevent overfitting.
- Dense Output Layer (7 classes: Angry, Disgust, Fear, Happy, Sad, Surprise, Neutral).
Integration
- Frontend: React application.
- Backend: Flask API hosting the TensorFlow model.
- Workflow: Webcam captures frame -> sent to API -> Model predicts "Happy" -> Spotify API plays "Upbeat Mix".
Accuracy
The custom CNN achieved 95.83% accuracy on the test set, outperforming pre-trained VGG16 transfer learning attempts on this specific lightweight task.