End-to-end DNN Speaker recognition system

Hlavní řešitel:Plchot Oldřich
Spoluřešitelé:Matějka Pavel
Další řešitelé:Novotný Ondřej, Silnova Anna
Agentura:Google
Zahájení:2017-03-07
Ukončení:2018-09-06
Klíčová slova:speaker recognition, DDN system
Anotace:
Text-independent speaker verification (SV) is currently the only bastion in the domain of speech data mining that resists the massive attack of deep neural networks (DNNs). We have already seen the end-to-end DNN approach to yield very good performance in the area of text-dependent SV and DNNs have been very successful in the related domain of spoken language recognition. In text-independent SV, features and frame-alignment based on DNNs has helped too, but so far, to end-to-end DNN system has matched the performance of iVectors. In this project, we will depart from existing DNN approaches for SV and advance towards fullDNN systems, mainly with the help of recently introduced Sequence-summarizing Neural Networks (SSNN). We will work on big and diverse datasets by utilizing large amount of publicly available data in combination with exploring efficient input feature representations.

Publikace

2017PLCHOT Oldřich, MATĚJKA Pavel, SILNOVA Anna, NOVOTNÝ Ondřej, DIEZ Sánchez Mireia, ROHDIN Johan A., GLEMBEK Ondřej, BRÜMMER Niko, SWART Albert du Preez, PRIETO Jesús J., GARCIA Perera Leibny Paola, BUERA Luis, KENNY Patrick, ALAM Jahangir a BHATTACHARYA Gautam. Analysis and Description of ABC Submission to NIST SRE 2016. In: Proceedings of Interspeech 2017. Stockholm: International Speech Communication Association, 2017, s. 1348-1352. ISSN 1990-9772.
 SILNOVA Anna, BURGET Lukáš a ČERNOCKÝ Jan. Alternative Approaches to Neural Network based Speaker Verification. In: Proceedings of Interspeech 2017. Stockholm: International Speech Communication Association, 2017, s. 1572-1575. ISSN 1990-9772.

Vaše IPv4 adresa: 54.227.104.40
Přepnout na IPv6 spojení

DNSSEC [dnssec]