Audio-visual speaker diarization using fisher linear semi-discriminant analysis