Investigating the use of visual focus of attention for audio-visual speaker diarisation