2021/03/11 by Jaswanth Reddy Katthi, Sriram Ganapathy, Katthi, Jaswanth Reddy +1
Computer Science · Neuroscience · #Audio and Speech Processing (eess.AS) #Blind Source Separation Techniques #EEG and Brain-Computer Interfaces #FOS: Biological sciences #FOS: Computer and information sciences #FOS: Electrical engineering #Neural dynamics and brain function #Quantitative Methods (q-bio.QM) #Sound (cs.SD) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2103.06478
openalex publication_date 2021/03/11 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
The normalization of brain recordings from multiple subjects responding to\nthe natural stimuli is one of the key challenges in auditory neuroscience. The\nobjective of this normalization is to transform the brain data in such a way as\nto remove the inter-subject redundancies and to boost the component related to\nthe stimuli. In this paper, we propose a deep learning framework to improve the\ncorrelation of electroencephalography (EEG) data recorded from multiple\nsubjects engaged in an audio listening task. The proposed model extends the\nlinear multi-way canonical correlation analysis (CCA) for audio-EEG analysis\nusing an auto-encoder network with a shared encoder layer. The model is trained\nto optimize a combined loss involving correlation and reconstruction. The\nexperiments are performed on EEG data collected from subjects listening to\nnatural speech and music. In these experiments, we show that the proposed deep\nmulti-way CCA (DMCCA) based model significantly improves the correlations over\nthe linear multi-way CCA approach with absolute improvements of 0.08 and 0.29\nin terms of the Pearson correlation values for speech and music tasks\nrespectively.\n