creativeAI: visualizing music emotions

Nowadays Artificial Intelligence (AI) is a widespread tool which recently has consistently came into our lives due to the enhancement of the Deep Learning technique made possible by means the growth of the hardware computational power. Despite its support on 'hard' tasks as object classification, geo-localization or text-completion, AI contributes to some other 'soft' tasks related to art, creativity and human emotions. However, as of now, there are few studies which aims to deal with these aspects using a multimodal approach, yet as humans we know that these subjects don't have a single communication channel. This work stems from the audio emotion recognition task, which is exploited in order to connect sound information to visual one through a generative process. At first a detailed analysis of the sound-emotion and visual-emotion existing datasets took place together with the alignment of both in order to have a shared set of emotion categories. Consequently different formalization of sound have been used and compared to perform the emotion recognition task and finally the sound emotion classifier output is used to condition a generative model to obtain new creative images providing the desired sentiment.