Biometric authentication relies on the use of biometric templates. A person seeking to access a biometrically controlled system must first identify himself/herself and the system will then verify a match with his/her stored template, created during an enrolment procedure. Such systems are intended for long-term deployment. Hence invariance to human ageing is an important element of any biometric system. The majority of human biometrics such as face, fingerprint, handwriting, iris and voice are subject to ageing effects. Other chapters in this book discuss many of these modalities at length. Here we will focus on speech. Speech is the least intrusive biometric from a user perspective. It can be used independently, or combined with other modalities such as face in multimodal systems.

Chapter Contents:

• 9.1 Introduction
• 9.2 Speech changes with age
• 9.3 Age in speaker verification
• 9.4 Speaker ageing data
• 9.4.1 Ageing UBM database
• 9.5 Ageing speaker verification experiment
• 9.5.1 Feature extraction and the GMM-UBM system
• 9.5.2 Ageing speaker verification evaluation
• 9.6 Effect of quality variation on speaker verification
• 9.6.1 SNR
• 9.6.2 Skewness
• 9.6.3 Kurtosis
• 9.6.4 UBM likelihood
• 9.6.5 Wnorm
• 9.6.6 Quality evaluation
• 9.7 Ageing-quality stacked classifier
• 9.7.1 Stacked classifier framework
• 9.7.2 Stacked classifier experimental evaluation
• 9.8 Conclusions
• References

Inspec keywords: authorisation; speech recognition; biometrics (access control); geriatrics

Other keywords: user perspective; speech-based biometric systems; biometric authentication; stored template; ageing impact; least intrusive biometric; biometric templates; multimodal systems; biometrically controlled system; long-term deployment

Subjects: Speech processing techniques; Speech recognition and synthesis; Data security