Detail publikace

13 years of speaker recognition research at BUT, with longitudinal analysis of NIST SRE

MATĚJKA, P. PLCHOT, O. GLEMBEK, O. BURGET, L. ROHDIN, J. ZEINALI, H. MOŠNER, L. SILNOVA, A. NOVOTNÝ, O. DIEZ SÁNCHEZ, M. ČERNOCKÝ, J.

Originální název

13 years of speaker recognition research at BUT, with longitudinal analysis of NIST SRE

Anglický název

13 years of speaker recognition research at BUT, with longitudinal analysis of NIST SRE

Jazyk

en

Originální abstrakt

In this paper, we present a brief history and a "longitudinal study" of all important milestone modelling techniques used in text independent speaker recognition since Brno University of Technology (BUT) first participated in the NIST Speaker Recognition Evaluation (SRE) in 2006-GMM MAP, GMM MAP with eigen-channel adaptation, Joint Factor Analysis, i-vector and DNN embedding (x-vector). To emphasize the historical context, the techniques are evaluated on all NIST SRE sets since 2004 on a time-machine principle, i.e. a system is always trained using all data available up till the year of evaluation. Moreover, as user-contributed audiovisual content dominates nowadays Internet, we representatively include the Speakers In The Wild (SITW) and VOiCES challenge datasets in the evaluation of our systems. Not only we present a comparison of the modelling techniques, but we also show the effect of sampling frequency.

Anglický abstrakt

In this paper, we present a brief history and a "longitudinal study" of all important milestone modelling techniques used in text independent speaker recognition since Brno University of Technology (BUT) first participated in the NIST Speaker Recognition Evaluation (SRE) in 2006-GMM MAP, GMM MAP with eigen-channel adaptation, Joint Factor Analysis, i-vector and DNN embedding (x-vector). To emphasize the historical context, the techniques are evaluated on all NIST SRE sets since 2004 on a time-machine principle, i.e. a system is always trained using all data available up till the year of evaluation. Moreover, as user-contributed audiovisual content dominates nowadays Internet, we representatively include the Speakers In The Wild (SITW) and VOiCES challenge datasets in the evaluation of our systems. Not only we present a comparison of the modelling techniques, but we also show the effect of sampling frequency.

Dokumenty

BibTex


@article{BUT162674,
  author="Pavel {Matějka} and Oldřich {Plchot} and Ondřej {Glembek} and Lukáš {Burget} and Johan Andréas {Rohdin} and Hossein {Zeinali} and Ladislav {Mošner} and Anna {Silnova} and Ondřej {Novotný} and Mireia {Diez Sánchez} and Jan {Černocký}",
  title="13 years of speaker recognition research at BUT, with longitudinal analysis of NIST SRE",
  annote="In this paper, we present a brief history and a "longitudinal study" of all
important milestone modelling techniques used in text independent speaker
recognition since Brno University of Technology (BUT) first participated in the
NIST Speaker Recognition Evaluation (SRE) in 2006-GMM MAP, GMM MAP with
eigen-channel adaptation, Joint Factor Analysis, i-vector and DNN embedding
(x-vector). To emphasize the historical context, the techniques are evaluated on
all NIST SRE sets since 2004 on a time-machine principle, i.e. a system is always
trained using all data available up till the year of evaluation. Moreover, as
user-contributed audiovisual content dominates nowadays Internet, we
representatively include the Speakers In The Wild (SITW) and VOiCES challenge
datasets in the evaluation of our systems. Not only we present a comparison of
the modelling techniques, but we also show the effect of sampling frequency.",
  address="NEUVEDEN",
  chapter="162674",
  doi="10.1016/j.csl.2019.101035",
  edition="NEUVEDEN",
  howpublished="print",
  institution="NEUVEDEN",
  number="63",
  volume="2020",
  year="2020",
  month="september",
  pages="1--15",
  publisher="NEUVEDEN",
  type="journal article in Web of Science"
}