Abstract: This paper describes the speaker diarization system submitted by EURECOM for the Albayzin 2016 speaker diarization evaluation. This evaluation consists of segmenting broadcast audio documents according to different speakers and attributing those segments to the speaker who uttered them, without any prior information about the speaker identities nor their number. EURECOM system is based on the binary key speaker modelling, an efficient and compact speech and speaker representation. The proposed system does not require external training data: the test data itself is used for estimating the resources needed. The system delivered a diarization error rate (DER) of 11.93% on the development set.

Download: