Channel variability synthesis in i-vector speaker recognition

Ahmed Isam  Ahmed; John Chiverton; David Ndzi; Victor Becerra

doi:10.1049/cp.2017.0362

Channel variability synthesis in i-vector speaker recognition

Ahmed Isam Ahmed, John Chiverton, David Ndzi, Victor Becerra

School of Computing, Engineering and Physical Sciences

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

1 Citation (Scopus)

Abstract

In this paper, we are tackling a practical problem which can be faced when establishing an i-vector speaker recognition system with limited resources. This addresses the problem of lack of development data of multiple recordings for each speaker. When we only have one recording for each speaker in the development set, phonetic variability can be simply synthesised by dividing the recordings if they are of sufficient length. For channel variability, we pass the recordings through a Gaussian channel to produce another set of recordings, referred to here as Gaussian version recordings. The proposed method for channel variability synthesis produces total relative improvements in EER of 5%.

Original language	English
Title of host publication	IET 3rd International Conference on Intelligent Signal Processing (ISP 2017)
Publisher	IET
Pages	1-6
Number of pages	6
ISBN (Electronic)	978-1-78561-708-9
ISBN (Print)	978-1-78561-707-2
DOIs	https://doi.org/10.1049/cp.2017.0362
Publication status	Published - 4 Dec 2017
Event	IET 3rd International Conference on Intelligent Signal Processing - Savoy, London, United Kingdom Duration: 4 Dec 2017 → 5 Dec 2017 https://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=8329306

Conference

Conference	IET 3rd International Conference on Intelligent Signal Processing
Abbreviated title	ISP 2017
Country/Territory	United Kingdom
City	London
Period	4/12/17 → 5/12/17
Internet address	https://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=8329306

Keywords

multi-condition training
session variability
i-vector

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1049/cp.2017.0362Licence: Unspecified

Cite this

@inproceedings{c37db5780db84468830ab5a41d7438ba,

title = "Channel variability synthesis in i-vector speaker recognition",

abstract = "In this paper, we are tackling a practical problem which can be faced when establishing an i-vector speaker recognition system with limited resources. This addresses the problem of lack of development data of multiple recordings for each speaker. When we only have one recording for each speaker in the development set, phonetic variability can be simply synthesised by dividing the recordings if they are of sufficient length. For channel variability, we pass the recordings through a Gaussian channel to produce another set of recordings, referred to here as Gaussian version recordings. The proposed method for channel variability synthesis produces total relative improvements in EER of 5%.",

keywords = "multi-condition training, session variability, i-vector",

author = "Ahmed, {Ahmed Isam} and John Chiverton and David Ndzi and Victor Becerra",

year = "2017",

month = dec,

day = "4",

doi = "10.1049/cp.2017.0362",

language = "English",

isbn = "978-1-78561-707-2",

pages = "1--6",

booktitle = "IET 3rd International Conference on Intelligent Signal Processing (ISP 2017)",

publisher = "IET",

address = "United Kingdom",

note = "IET 3rd International Conference on Intelligent Signal Processing, ISP 2017 ; Conference date: 04-12-2017 Through 05-12-2017",

url = "https://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=8329306",

}

TY - GEN

T1 - Channel variability synthesis in i-vector speaker recognition

AU - Ahmed, Ahmed Isam

AU - Chiverton, John

AU - Ndzi, David

AU - Becerra, Victor

PY - 2017/12/4

Y1 - 2017/12/4

N2 - In this paper, we are tackling a practical problem which can be faced when establishing an i-vector speaker recognition system with limited resources. This addresses the problem of lack of development data of multiple recordings for each speaker. When we only have one recording for each speaker in the development set, phonetic variability can be simply synthesised by dividing the recordings if they are of sufficient length. For channel variability, we pass the recordings through a Gaussian channel to produce another set of recordings, referred to here as Gaussian version recordings. The proposed method for channel variability synthesis produces total relative improvements in EER of 5%.

AB - In this paper, we are tackling a practical problem which can be faced when establishing an i-vector speaker recognition system with limited resources. This addresses the problem of lack of development data of multiple recordings for each speaker. When we only have one recording for each speaker in the development set, phonetic variability can be simply synthesised by dividing the recordings if they are of sufficient length. For channel variability, we pass the recordings through a Gaussian channel to produce another set of recordings, referred to here as Gaussian version recordings. The proposed method for channel variability synthesis produces total relative improvements in EER of 5%.

KW - multi-condition training

KW - session variability

KW - i-vector

U2 - 10.1049/cp.2017.0362

DO - 10.1049/cp.2017.0362

M3 - Conference contribution

SN - 978-1-78561-707-2

SP - 1

EP - 6

BT - IET 3rd International Conference on Intelligent Signal Processing (ISP 2017)

PB - IET

T2 - IET 3rd International Conference on Intelligent Signal Processing

Y2 - 4 December 2017 through 5 December 2017

ER -

Channel variability synthesis in i-vector speaker recognition

Abstract

Conference

Keywords

UN SDGs

Access to Document

Fingerprint

Cite this