Codificação e decodificação da resposta cerebral à música

• Ano de defesa: 2017
• Autor(a) principal: Höfle, Sebastian
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Rio de Janeiro; Brasil; Instituto Alberto Luiz Coimbra de Pós-Graduação e Pesquisa de Engenharia; Programa de Pós-Graduação em Engenharia Biomédica; UFRJ
• Palavras-chave em Português: Engenharia biomédica; Regressão linear múltipla; Música; CNPQ::ENGENHARIAS::ENGENHARIA BIOMEDICA
• Link de acesso: http://hdl.handle.net/11422/10187

The neural correlates of listening to music have been investigated in several ways. However, mapping ongoing brain activity during naturalistic music listening combined with detailed models of musical features is an emerging approach. The socalled “encoding models” allow capturing the effects of multiple stimulus variables on brain responses that can be used subsequently to decode or identify stimuli from brain activity. This work applies methods for encoding and decoding brain activity in response to naturalistic music listening. First, musical features are extracted from the audio signal and mapped to brain activity, measured by functional magnetic resonance imaging. Building on this mapping, multivariate spatial representations are decoded in order to identify a specific music which is listened as the activity is measured. Further, a systematic investigation reveals internal parameters that maximize model performance. Among the most important parameters are the ideal point in the spatial dimension and the consideration of entropy of the music pieces, resulting in maximum accuracies of up to 95%.