Národní úložiště šedé literatury Nalezeno 2 záznamů.  Hledání trvalo 0.01 vteřin. 
Speaker Recognition Based on Long Temporal Context
Fér, Radek ; Matějka, Pavel (oponent) ; Černocký, Jan (vedoucí práce)
This work deals with temporal features for automated speaker recognition. We give overview of currently known temporal feature extraction methods and afterwards, we propose and preliminarily evaluate a general phoneme-level temporal feature extraction scheme based on factor analysis i-vector paradigm. Much effort has been made to reasonably represent temporal context and make speaker recognition systems more robust, namely speech prosody modeling. Our approach does not explicitly model any temporal parameters of speech, rather it uses the occurrence of neighboring frames as a source of temporal information. We test and analyze this method on standard evaluation database NIST SRE 2008. The results indicate, however, that for speaker recognition, no useful gain can be obtained using this technique. We describe and discuss this discovery at the end.
Speaker Recognition Based on Long Temporal Context
Fér, Radek ; Matějka, Pavel (oponent) ; Černocký, Jan (vedoucí práce)
This work deals with temporal features for automated speaker recognition. We give overview of currently known temporal feature extraction methods and afterwards, we propose and preliminarily evaluate a general phoneme-level temporal feature extraction scheme based on factor analysis i-vector paradigm. Much effort has been made to reasonably represent temporal context and make speaker recognition systems more robust, namely speech prosody modeling. Our approach does not explicitly model any temporal parameters of speech, rather it uses the occurrence of neighboring frames as a source of temporal information. We test and analyze this method on standard evaluation database NIST SRE 2008. The results indicate, however, that for speaker recognition, no useful gain can be obtained using this technique. We describe and discuss this discovery at the end.

Chcete být upozorněni, pokud se objeví nové záznamy odpovídající tomuto dotazu?
Přihlásit se k odběru RSS.