Speaker Diarization using Unsupervised Compensation of Within-Speaker Variability

This paper presents a novel framework for unsupervised compensation of within-speaker variability in the context of speaker diarization. Audio session is divided into overlapping short segments, each one parameterized by a GMM-supervector. For each session independently within-speaker variability is estimated in an unsupervised manner, and is compensated using the nuisance attribute projection (NAP) method. The proposed method was evaluated in the context of speaker diarization in two-speaker telephone conversations and achieved a speaker error rate of 2.8% which is a 54% relative error reduction compared to a baseline BIC-based system.