Probabilistic random projections and speaker verification

Chong Lee Ying, Andrew Teoh Beng Jin

Research output: Chapter in Book/Report/Conference proceedingConference contribution

5 Citations (Scopus)


Biometrics is susceptible to non-revocable and privacy invasion problems. Multiple Random Projections (MRP) was introduced as one of the cancellable biometrics approaches in face recognition to tackle these issues. However, this technique is applicable only to ID fixed length biometric feature vector but failed in varying size feature, such as speech biometrics. Besides, simple matching metric that used in MRP unable to offer a satisfactory verification performance. In this paper, we propose a variant of MRP, coined as Probabilistic Random Projections (PRP) in text-independent speaker verification. The PRP represents speech feature in 2D matrix format and speaker modeling is implemented through Gaussian Mixture Model. The formulation is experimented under two scenarios (legitimate and stolen token) using YOHO speech database. Besides that, desired properties such as one-way transformation and diversity are also examined.

Original languageEnglish
Title of host publicationAdvances in Biometrics - International Conference, ICB 2007, Proceedings
PublisherSpringer Verlag
Number of pages10
ISBN (Print)9783540745488
Publication statusPublished - 2007
Event2007 International Conference on Advances in Biometrics, ICB 2007 - Seoul, Korea, Republic of
Duration: 2007 Aug 272007 Aug 29

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume4642 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Other2007 International Conference on Advances in Biometrics, ICB 2007
Country/TerritoryKorea, Republic of

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)


Dive into the research topics of 'Probabilistic random projections and speaker verification'. Together they form a unique fingerprint.

Cite this