Continuous speech recognition using an on-line speaker adaptation method based on automatic speaker clustering