Likelihood-ratio calibration using prior-weighted proper scoring rules

Prior-weighted logistic regression has become a standard tool for calibration in speaker recognition. Logistic regression is the optimization of the expected value of the logarithmic scoring rule. We generalize this via a parametric family of proper scoring rules. Our theoretical analysis shows how different members of this family induce different relative weightings over a spectrum of applications of which the decision thresholds range from low to high. Special attention is given to the interaction between prior weighting and proper scoring rule parameters. Experiments on NIST SRE'12 suggest that for applications with low false-alarm rate requirements, scoring rules tailored to emphasize higher score thresholds may give better accuracy than logistic regression.

• Publication year

  2013

• Venue

  Interspeech

• Publication date

  2013-07-30

• Fields of study

  Mathematics, Computer Science

• Identifiers
  DOI 10.21437/Interspeech.2013-470arXiv 1307.7981
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Numerical Optimization

  2018cited by this paper
• The BOSARIS Toolkit: Theory, Algorithms and Code for Surviving the New DCF

  2013cited by this paper
• The role of proper scoring rules in training and evaluating probabilistic speaker and language recognizers

  2012cited by this paper
• The Role of Score Calibration in Speaker Recognition

  2012cited by this paper
• Discriminatively trained Probabilistic Linear Discriminant Analysis for speaker verification

  2011cited by this paper
• Measuring, refining and calibrating speaker and language information extracted from speech

  2010influential reference
• Fusion of Heterogeneous Speaker Recognition Systems in the STBU Submission for the NIST Speaker Recognition Evaluation 2006

  2007cited by this paper
• Strictly Proper Scoring Rules, Prediction, and Estimation

  2007cited by this paper
• Loss Functions for Binary Class Probability Estimation and Classification: Structure and Applications

  2005cited by this paper
• Application-independent evaluation of speaker detection

  2004cited by this paper
• Transforming classifier scores into accurate multiclass probability estimates

  2002influential reference
• Optimal Statistical Decisions

  1970cited by this paper
• VERIFICATION OF FORECASTS EXPRESSED IN TERMS OF PROBABILITY

  1950cited by this paper
• The NIST Year 2012 Speaker Recognition Evaluation Plan 1 I

  year unknowncited by this paper

• A Copula-Based Generative Score-Level Fusion Model for Speaker Verification

  2025cites this paper
• Analysis of the ABC Classification Backends for NIST SRE24

  2025cites this paper
• Revisiting and Improving Scoring Fusion for Spoofing-aware Speaker Verification Using Compositional Data Analysis

  2024cites this paper
• A Generative Approach to Condition-Aware Score Calibration for Speaker Verification

  2023cites this paper
• The Distributions of Uncalibrated Speaker Verification Scores: A Generative Model for Domain Mismatch and Trial-Dependent Calibration

  2023cites this paper
• Analysis and Comparison of Classification Metrics

  2022cites this paper
• Impostor Score Statistics as Quality Measures for the Calibration of Speaker Verification Systems

  2022cites this paper
• Investigation of Different Calibration Methods for Deep Speaker Embedding based Verification Systems

  2022cites this paper
• A Generative Model for Duration-Dependent Score Calibration

  2021cites this paper
• A Speaker Verification Backend with Robust Performance across Conditions

  2021influential citation
• On the Distribution of Speaker Verification Scores: Generative Models for Unsupervised Calibration

  2021cites this paper
• Machine Learning for Speaker Recognition

  2020cites this paper
• A Speaker Verification Backend for Improved Calibration Performance across Varying Conditions

  2020cites this paper
• A Discriminative Condition-Aware Backend for Speaker Verification

  2019cites this paper
• Partial AUC Metric Learning Based Speaker Verification Back-End

  2019cites this paper
• Tied Normal Variance–Mean Mixtures for Linear Score Calibration

  2019cites this paper
• Normal Variance-Mean Mixtures for Unsupervised Score Calibration

  2019cites this paper
• Speaker Verification by Partial AUC Optimization With Mahalanobis Distance Metric Learning

  2019cites this paper
• DNN-Based Score Calibration With Multitask Learning for Noise Robust Speaker Verification

  2018cites this paper
• Joint PLDA for Simultaneous Modeling of Two Factors

  2018cites this paper
• Implementation and assessment of a likelihood ratio approach for the evaluation of LA-ICP-MS evidence in forensic glass analysis.

  2017cites this paper
• Improving classifier fusion via Pool Adjacent Violators normalization

  2016cites this paper
• Robust discriminative training against data insufficiency in PLDA-based speaker verification

  2016cites this paper
• Posterior calibration and exploratory analysis for natural language processing models

  2015cites this paper
• An information theory based data-homogeneity measure for voice comparison

  2015cites this paper
• Robust speaker verification

  2014cites this paper
• Discriminative PLDA training with application-specific loss functions for speaker verification

  2014cites this paper
• Bayesian calibration for forensic evidence reporting

  2014cites this paper
• A comparison of linear and non-linear calibrations for speaker recognition

  2014influential citation
• Generative modelling for unsupervised score calibration

  2013cites this paper
• Proceedings of Biometric Technologies in Forensic Science

  2013cites this paper
• The distribution of calibrated likelihood-ratios in speaker recognition

  2013cites this paper