Design Choices for X-vector Based Speaker Anonymization
Brij Mohan Lal Srivastava
(1)
,
Natalia Tomashenko
(2)
,
Xin Wang
(3)
,
Emmanuel Vincent
(4)
,
Junichi Yamagishi
(3)
,
Mohamed Maouche
(5)
,
Aurélien Bellet
(1)
,
Marc Tommasi
(6)
Brij Mohan Lal Srivastava
- Fonction : Auteur
- PersonId : 1057878
Natalia Tomashenko
- Fonction : Auteur
- PersonId : 17002
- IdHAL : natalia-tomashenko
- IdRef : 223393304
Xin Wang
- Fonction : Auteur
- PersonId : 761950
- ORCID : 0000-0002-3891-2684
Emmanuel Vincent
- Fonction : Auteur
- PersonId : 1256
- IdHAL : emmanuelv
- ORCID : 0000-0002-0183-7289
- IdRef : 089360176
Mohamed Maouche
- Fonction : Auteur
- PersonId : 19280
- IdHAL : mohamed-maouche
- ORCID : 0000-0001-6473-7173
Aurélien Bellet
- Fonction : Auteur
- PersonId : 9877
- IdHAL : aurelien-bellet
- ORCID : 0000-0003-3440-1251
- IdRef : 17653136X
Marc Tommasi
- Fonction : Auteur
- PersonId : 399
- IdHAL : marc-tommasi
- ORCID : 0000-0003-2838-4408
- IdRef : 121846385
Résumé
The recently proposed x-vector based anonymization scheme converts any input voice into that of a random pseudo-speaker. In this paper, we present a flexible pseudo-speaker selection technique as a baseline for the first VoicePrivacy Challenge. We explore several design choices for the distance metric between speakers, the region of x-vector space where the pseudo-speaker is picked, and gender selection. To assess the strength of anonymization achieved, we consider attackers using an x-vector based speaker verification system who may use original or anonymized speech for enrollment, depending on their knowledge of the anonymization scheme. The Equal Error Rate (EER) achieved by the attackers and the decoding Word Error Rate (WER) over anonymized data are reported as the measures of privacy and utility. Experiments are performed using datasets derived from LibriSpeech to find the optimal combination of design choices in terms of privacy and utility.
Format du dépôt | Fichier |
---|---|
Type de dépôt | Communication dans un congrès |
Titre |
en
Design Choices for X-vector Based Speaker Anonymization
|
Résumé |
en
The recently proposed x-vector based anonymization scheme converts any input voice into that of a random pseudo-speaker. In this paper, we present a flexible pseudo-speaker selection technique as a baseline for the first VoicePrivacy Challenge. We explore several design choices for the distance metric between speakers, the region of x-vector space where the pseudo-speaker is picked, and gender selection. To assess the strength of anonymization achieved, we consider attackers using an x-vector based speaker verification system who may use original or anonymized speech for enrollment, depending on their knowledge of the anonymization scheme. The Equal Error Rate (EER) achieved by the attackers and the decoding Word Error Rate (WER) over anonymized data are reported as the measures of privacy and utility. Experiments are performed using datasets derived from LibriSpeech to find the optimal combination of design choices in terms of privacy and utility.
|
Auteur(s) |
Brij Mohan Lal Srivastava
1
, Natalia Tomashenko
2
, Xin Wang
3
, Emmanuel Vincent
4
, Junichi Yamagishi
3
, Mohamed Maouche
5
, Aurélien Bellet
1
, Marc Tommasi
6
1
MAGNET -
Machine Learning in Information Networks
( 432650 )
- France
2
LIA -
Laboratoire Informatique d'Avignon
( 100376 )
- 339 Chemin des Meinajaries Agroparc BP 1228 84911 Avignon cedex 9
- France
3
NII -
National Institute of Informatics
( 6501 )
- 2-1-2 Hitotsubashi, Chiyoda-ku, Tokyo 101-8430
- Japon
4
MULTISPEECH -
Speech Modeling for Facilitating Oral-Based Communication
( 420403 )
- France
5
DRIM -
Distribution, Recherche d'Information et Mobilité
( 394737 )
- France
6
Université de Lille
( 374570 )
- EPE Université de Lille. -- 42 rue Paul Duez, 59000 Lille
- France
|
Langue du document |
Anglais
|
Vulgarisation |
Non
|
Comité de lecture |
Oui
|
Invité |
Non
|
Audience |
Internationale
|
Actes |
Oui
|
Titre du congrès |
INTERSPEECH 2020
|
Date début congrès |
2020-10-25
|
Date fin congrès |
2020-10-29
|
Ville |
Shanghai
|
Pays |
Chine
|
Domaine(s) |
|
Organisateur du congrès |
|
Projet(s) ANR |
|
Projet(s) Européen(s) |
|
Collaboration/Projet |
|
Mots-clés |
en
VoicePrivacy challenge, speaker anonymization, voice conversion, x-vectors, PLDA
|
arXiv Id | 2005.08601 |
Origine :
Fichiers produits par l'(les) auteur(s)
Loading...