Recognizing Emotions from Voice: A Prototype.

Over time, the interaction between humans and machines has become increasingly important for both personal and commercial use. As technology continues to permeate various aspects of our lives, it is essential to seek healthy progress and not only improve, but also maintain the benefits that technology brings. While this relationship can be approached from many angles, this discussion focuses on emotions. Emotions remain a complex and enigmatic concept that remains a mystery to scientists. As such, it is crucial to pave the right way for the development of technology that can help understand emotions. Some indicators, such as word use, facial expressions and voice, provide important information about mental states. This work focuses on voice and proposes a comprehensive process for automatic emotion recognition by speech. The pipeline includes sound capture and signal processing software, algorithms for learning and classification within the Semi-Supervised Learning paradigm and visualisation techniques to interpret the results. For classifying the samples, a semi-supervised approach using Neural Networks is adopted to reduce the reliance on human emotion labelling, which is often subjective, difficult and costly. Empirical results carry more weight than theoretical concepts, given the complexity and uncertainty inherent in human emotions, but prior knowledge in this domain is not ignored.