Určení kdy kdo mluví

Speaker Diarization

Developed speaker diarization system is based on standard Bayesian Information Criterion segmentation and clustering algorithm. Phoneme recognizer is used for speech and non-speech detection. System is still not finished nor extensively tested, five possible solutions for segmentation and clustering are compared. No metric was used for obtained results, but at the first look they seem to be promising.

diarization, speaker segmentation, BIC, Bayesian information criterion

ENT, Petr. Určení kdy kdo mluví. Brno, 2008. Semestral project. Brno University of Technology, Faculty of Information Technology. 2008-01-08. Supervised by Matějka Pavel. Available from: https://www.fit.vut.cz/study/thesis/6939/