Tensorflow implementation of speaker recognition with x-vector topology

Tensorflow implementace rozpoznávání mluvčího s x-vector topologií

software

Use of the result by another entity is possible without acquiring a license (the result is not licensed)

The licensor does not require a license fee for the result

Zeinali Hossein, Ph.D. (DCGM)
Burget Lukáš, doc. Ing., Ph.D. (DCGM)
Rohdin Johan Andréas, M.Sc., Ph.D. (DCGM)
Stafylakis Themos
Černocký Jan, prof. Dr. Ing. (DCGM)

Speaker recognition, speaker embedding, DNN, x-vectors, retraining

This is a Tensorflow implementation of x-vector topology (speaker embedding). It uses Kaldi toolkit for data processing. We train the model using Tensorflow and also extract speaker embeddings (x-vectors) using it. This allow to train or retrain the system to the particular customer specific domain or provides the ability to modify the topology or training schema to achieve better performance for the specific domain.  This software is a result of Czech Ministry of Interior project  "Dolování infoRmAcí z řeči Pořízené vzdÁlenými miKrofony - DRAPÁK", No. VI20152020025 (https://www.fit.vut.cz/research/project/1009/)

https://github.com/BUTSpeechFIT/x-vector-kaldi-tf

• pdf zeinali_icassp2019_0006141.pdf 197 kB

Information mining in speech acquired by distant microphones, MV, Bezpečnostní výzkum České republiky 2015-2020, VI20152020025, 2015-2020, completed

Speech Data Mining Research Group BUT Speech@FIT (RG SPEECH)

Department of Computer Graphics and Multimedia (DCGM)