Our aim is to fill the gap between computer music and musical emotions. We use a 2D emotional space (2D ES) both to analyze listener's emotions captured from physiological measure and to drive a music generator with a Reinforcement Learning algorithm. We present hereafter the Reinforcement Learning algorithm and the different parts of our framework used as input (emotions classification with Emotiv EEG headset and Kirlian photography device, to improve the machine learning by using data that have no correlation between them), and output of the algorithm (rhythm and scale/chords generation). Some current and future applications are also briefly depicted.