Speech/Non-Speech Detection in Meetings from Automatically Extracted low Resolution Visual Features