Fisher vectors in PLDA speaker verification system

The goal of this paper is to examine the Fisher Vector and incorporate this vector in the PLDA based speaker verification system. The PLDA based system utilizes the Supervector of Statistics extracted from a Gaussian Mixture Model (adopted from the speaker adaptation task) to collect the information...

Full description

Saved in:
Bibliographic Details
Main Authors: Zajic, Zbynek, Hruz, Marek
Format: Conference Proceeding
Language:English
Subjects:
Covariance matrices
Data models
Feature extraction
Fisher Vector
Gaussian mixture model
iVector
Linear discriminant analysis
Maximum likelihood estimation
PLDA
Speaker Verification
Supervector
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The goal of this paper is to examine the Fisher Vector and incorporate this vector in the PLDA based speaker verification system. The PLDA based system utilizes the Supervector of Statistics extracted from a Gaussian Mixture Model (adopted from the speaker adaptation task) to collect the information about a speaker from a dataset. We compare the efficiency of the PLDA based speaker verification system using Supervector of Statistics and the same system with Fisher vector. The experimental results of these two approaches to the verification task and the fusion of these two systems indicate that the Fisher Vector brings almost the same information to the PLDA verification process as the Supervector of Statistics when sufficient data are available.
ISSN:2164-5221
DOI:10.1109/ICSP.2016.7878044