Decoding Auditory Attention from Multivariate Neural Data using Cepstral Analysis

Sammanfattning: Very little is known about the remarkable ability of humans to separate a single sound source from a dense mixture of sound sources in a crowded background, known as the cocktail-party scenario. Better understanding could lead to a breakthrough for the next-generation of hearing aids to have the ability to be cognitively controlled. A key finding in the field is that human cortical activity has been shown to follow the speech envelope. However, in these experimental results, the correlation coefficients between the EEG and speech envelope are very low, on the order of r = 0.1-0.2. Also, classification rates are not yet 100%. The aim of this project is to investigate whether cepstral analysis can be used as a more robust mapping between speech and EEG. Our preliminary results show correlations on the order of r > 0.5. This thesis will give a insight into the method we are developing, our current results, and the expected future results and applications in hearing aids.