Análise e reconhecimento de padrões cognitivos em escutas musicais e sonoros em áudios

We are involved in an environment full of sounds around us. Studying and analyzing the impacts that musical practice causes and showing mathematically that this practice provides significant cognitive effects on the human brain are the main motivations of this thesis. In more detail, the aim of this thesis was to develop a methodology capable of characterizing the cortical activation patterns generated during the register of Electroencephalogram (EEG) signals through pattern recognition techniques in statistics, in addition to analyzing the acoustic features commonly employed in this context, in order to reveal whether they are statistically relevant. A computational framework was initially developed to address a 2 group classification problem based on data from EEG signals extracted from volunteer musicians and non-musicians during an auditory task, to predict whether a particular person is a musician or not. The results showed that it is possible to classify the sampled groups with accuracy ranging from 69.2% to 93.8%, allowing not only a better description of the neural activation patterns that characterize the musician and non-musician volunteers, but also highlighting how these patterns they change in the transition regions and decision boundaries that separate the sampled groups, indicating a plausible linear separation between these groups. Additionally, as another original contribution of this thesis, the audio signals from a public and internationally referenced database containing 1000 musical excerpts with 10 different genres were analyzed to investigate numerical similarities between the short-term acoustic features extracted from the audios and commonly explored in related literature. The results obtained show a similar cluster behavior among these features for all analyzed music, regardless of the musical genre. It was then possible to discuss in an unprecedented way the relationship between the way the acoustic features of songs are described in the literature and how they are grouped statistically, revealing that the information we use to cognitively process these sound features is implicitly statistical. Although all the methods described and implemented in this thesis are based on EEG signals, it is believed that they can be extended to other types of multivariate cognitive signals, such as, for example, functional Magnetic Resonance Imaging (fMRI), allowing a greater cortical and sub-cortical understanding of the functioning of our brain during listening