Advances in Speaker Diarization

Abstract: Speaker diarization is the task of segmenting speech into speaker coherent segments and clustering the resulting segments according to speaker identity.

Our approach is to do speaker diarization by learning within speaker variability either explicitly from labeled training data or implicitly in an unsupervised manner. Furthermore, we show how side information such as the identity of one of the speakers can be integrated into the diarization scheme.

תאריך: 
22/05/2013 - 15:00
מרצה: 
Hagai Aronowitz
דוא"ל להרשמה: 
Affiliation: 
IBM Research Lab, Haifa
מיקום: 
building 1103, Room 329