As the years go by, the interaction between humans and machines seems to gain more and more importance for many different reasons, whether it's taken into consideration personal or commercial use. On a time where technology is reaching many parts of our lives, it's important to keep thriving for a healthy progress and help not only to improve but also to maintain the benefits that everyone gets from it. This relationship can be tackled through many points, but here the focus will be on the mind. Emotions are still a mystery. The concept itself brings up serious questions because of its complex nature. Till the date, scientists still struggle to understand it, so it's crucial to pave the right path for the growth on technology on the aid of such topic. There is some consensus on a few indicators that provide important insights on mental state, like words used, facial expressions, voice. The context of this work is on the use of voice and, based on the field of Automatic Speech Emotion Recognition, it is proposed a full pipeline of work with a wide scope by resorting to sound capture and signal processing software, to learning and classifying through algorithms belonging on the Semi Supervised Learning paradigm and visualization techniques for interpretation of results. For the classification of the samples,using a semi-supervised approach with Neural Networks represents an important setting to try alleviating the dependency of human labelling of emotions, a task that has proven to be challenging and, in many cases, highly subjective, not to mention expensive. It is intended to rely mostly on empiric results more than theoretical concepts due to the complexity of the human emotions concept and its inherent uncertainty, but never to disregard prior knowledge on the matter.