Speaker recognition regardless of context and language on a fixed set of competitors