Score Normalization for Text-Independent Speaker Verification Systems

By Roland Auckenthaler, M. Carey, H. Lloyd-Thomas. 2000

882 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Signal Processing

Citations in the corpus

End-to-end text-dependent speaker verification (2015, background): “Results are reported in terms of equal error rate (EER), without and with t-norm score normalization [25].”
Meta-Recognition: The Theory and Practice of Recognition Score Analysis (2011, methodology): “Cohort analysis [10], [11], [12], [ 13 ], [14], [15] is a postverification (one versus one matching, as opposed to recognition’s one versus IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL.”
Deep bottleneck features for i-vector based text-independent speaker verification (2015, methodology): “Tnorm is an example of such methods [6].”
Feature warping for robust speaker verification (2001, methodology): “An improvement on the likelihood ratio test combined with H-Norm is the inclusion of Test segment Normalisation (TNorm) [13].”
Within-class covariance normalization for SVM-based speaker recognition (2006, methodology): “However, besides applying the feature transformation to the input feature vectors, our system differs from the MLLR-SVM sy stem used in [4] in the following ways: 1) our system does not ap ply rank normalization [3] to the input feature vectors and 2 ) our system does not apply TNORM [13] to the out”
A Tutorial on Text-Independent Speaker Verification (2004, methodology): “Still based on the estimate of mean and variance parameters to normalize impostor score distribution, test-normalization (Tnorm), proposed in [44], differs from Znorm by the use of impostor models instead of test speech signals.”
An Introduction to Application-Independent Evaluation of Speaker Recognition Systems (2007, neutral): “target and non-target distributions, respectively [ 10 ,11].”
HMM-based on-line signature verification: Feature extraction and signature modeling (2007, methodology): “09 teriori target-dependent score normalization techniques, namely: no score normalization, z-norm (Auckenthaler et al., 2000), and Target-Dependent Score Normalization based on individual EERs (EER-TDSN) as described in Section 4.”
Support vector machines for speaker and language recognition (2006, methodology): “Other methods for GMMs including cohort normalization [21] and TNorm [22] are closer to the proposed SVM method; although, the latter method (TNorm) 7”
Learning Human Identity From Motion Patterns (2015, background): “As a final step, zt-score normalization [2] is performed to compensate for inter-session and inter-person variations and reduce the overlap between the distribution of scores from authentic users and impostors.”
Heterogeneous information fusion: A novel fusion paradigm for biometric systems (2011, neutral): “A generalized form of T-norm was reported in [10].”
Large-scale evaluation of multimodal biometric authentication using state-of-the-art systems (2005, background): “Previously, test normalization (T-norm) [11] that can be thought of as adaptive normalization considering impostor scores was proposed.”
Combining evidence from residual phase and MFCC features for speaker recognition (2006, methodology): “The performance has improved to 14% when normalization techniques [model normalization (ZNorm) and TNorm] are employed [18].”
Application-independent evaluation of speaker detection (2004, background): “1 Scaling and shifting is a side-effect of some score normalization schemes such as T-norm [16].”
Finding Difficult Speakers in Automatic Speaker Recognition (2011, methodology): “The final normalization of interest is test normalization, or T-norm, which generates scores for a test utterance against impostor models (in addition to the target model), in order to estimate the impostor score distribution’s statistics [4].”
Deep neural networks for small footprint text-dependent speaker verification (2014, result): “The EER results with t-norm [22] for score normalization are consistently much better than with the raw scores.”