Something to be prepared for when pursuing something like this is signals being diversely reflected and transformed, which could make it frustrating to identify them coming from one single source, unless say there is a clear path to it.
This can likely be handled in a similar manner, and I'm sure somebody with actual experience in the domain would be pretty familiar with doing so. All problems are solvable but we have to start with the ones in front of us to get anywhete. ICA is a pretty old form of source separation and doesn't handle timeshifting or media that cause nonlinear transformations. Still may be good enough if used in an informed way with a willingness to transform the data to look like plain linear mixing.
