Music recommendation app based on user affect using CNNs for emotion classification and valence/arousal regression. Song recommendations made using the Spotify Web API. Intended as a proof-of-concept that emotionally intelligent user interfaces are now feasbile on todays high-spec mobile phones.

CNNs are trained on AffectNet using Keras and deployed to iOS by conversion to coreML using coremltools. Networks are designed to minimise space consumption, but retain near state-of-the-art performance on AffectNet.