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.