Neural dynamics underlying human vocalization

Human speech allows to convey information via different execution forms. Thus, on the behavioral level, speech content is independent from production and execution forms. However, it remains unclear whether such a content dimension can be found on the neural level and whether it is possible to dissociate it from the motor dimension. To address this, we recorded magnetoencephalography (MEG) in human subjects that performed two rule-based vocalization tasks. In the first experiment, content (one of two vowels) and production (overt or covert) were instructed separately and in random order. Applying multivariate pattern analysis (MVPA), we were able to decode both content and production independently and several seconds before vocalization behavior. The temporal dynamics of information reflected cue order. In the second experiment, we used three vowels to assess effects of motor effort. We found that, in line with an abstract neuronal representation, content information indeed generalized across production types. Furthermore, substantial decoding of all vowels, even those with similar motor effort, and little cross-decoding between content and production information suggested content represenations to be largely independent from motor effort. Together, our results provide insights into the neural dynamics underlying human vocalization and open a new window for noninvasive speech research in health and disease.