Submitted by Roman Derunets 36 Pisets: A Robust Speech Recognition System for Lectures and Interviews Novosibirsk State University 82 3