SCCN Talk

Detection of Perceived and Imagined Rhythm from ERPs

A computational method is presented for identifying, on the basis of ERP signals, which rhythm a subject is listing too or imagining. Preprocessing was conducted by independent component analysis. The resulting un-mix matrix was applied to both single trials and average templates, concentrating the information relevant to rhythm processing into a few channels. These were spectrally separated by a filter bank before template matching was conducted, using correlation as similarity measure. A stepwise classification based on these correlations was able to correctly identify a single perceptual trial out of five presented rhythms in about half of the cases. Identifying imagined rhythm was harder, but the method still performed significantly better than chance. The results demonstrate that, even though well-structured rhythms are about as easy to detect as difficult ones during perception, detection during imagery is much more successful for simple patterns.