Workshop on automatic and semiautomatic speaker recognition
Il convegno sarà preceduto da un workshop satellite, della durata di mezza giornata, tenuto da Michael Jessen (Bundeskriminalamt, Wiesbaden, Germania, Language and Audio Division).
In the first part of the workshop an overview will be given of the historical development of methods/systems that quantify comparisons between speakers in terms of a likelihood ratio. Some of the methods make use of input from acoustic phonetics (semiautomatic speaker recognition), others are based on cepstral coefficients, familiar from speech technology (automatic speaker recognition). The overview includes a demonstration of semiautomatic methods as well as automatic methods in their different historical stages (GMM-UBM, i-vector, x-vector) using the software VOCALISE. In the second part of the workshop participants have the opportunity to gain practical experience in the use of a speaker recognition system. Information regarding temporary licenses and access to speaker corpora will be provided directly to the workshop participants.