Short- and Long-Term Speech Features for Hybrid HMM-i-Vector based Speaker Diarization System.