This master's thesis presents the development of an emotion-conditioned symbolic music generation system based on Russell's circumplex model. The research follows a machine learning pipeline methodology, which includes:

• Data collection and annotation
• Data preprocessing & feature engineering
• Model architecture design
• Model training and optimization
• Quantitative and qualitative evaluation

Emotional Framework

Valence: Positive ↔ Negative; Arousal: High energy ↔ Low energy

The implementation results include:

• A functional demo of the developed graphical user interface
• Four generated MIDI files by our custom-built autoregressive transformer model
• Complete thesis documentation (Spanish PDF)

Demo

Generated midi

Excited (High Arousal, Positive Valence)

Anger (High Arousal, Negative Valence)

Sadness (Low Arousal, Negative Valence)

Calm (Low Arousal, Positive Valence)

Thank you to html-midi-player for the MIDI visualizer