Verbal information verification for high-performance speaker authentication.

Qin Chao.Thesis (M.Phil.)--Chinese University of Hong Kong, 2005.Includes bibliographical references (leaves 77-82).Abstracts in English and Chinese.Chapter 1 --- Introduction --- p.1Chapter 1.1 --- Overview of Speaker Authentication --- p.1Chapter 1.2 --- Goals of this Research --- p.6Chapter 1.3 --- Thesis Outline --- p.7Chapter 2 --- Speaker Verification --- p.8Chapter 2.1 --- Introduction --- p.8Chapter 2.2 --- Front-End Processing --- p.9Chapter 2.2.1 --- Acoustic Feature Extraction --- p.10Chapter 2.2.2 --- Endpoint Detection --- p.12Chapter 2.3 --- Speaker Modeling --- p.12Chapter 2.3.1 --- Likelihood Ratio Test for Speaker Verification --- p.13Chapter 2.3.2 --- Gaussian Mixture Models --- p.15Chapter 2.3.3 --- UBM Adaptation --- p.16Chapter 2.4 --- Experiments on Cantonese Speaker Verification --- p.18Chapter 2.4.1 --- Speech Databases --- p.19Chapter 2.4.2 --- Effect of Endpoint Detection --- p.21Chapter 2.4.3 --- Comparison of the UBM Adaptation and the Cohort Method --- p.22Chapter 2.4.4 --- Discussions --- p.25Chapter 2.5 --- Summary --- p.26Chapter 3 --- Verbal Information Verification --- p.28Chapter 3.1 --- Introduction --- p.28Chapter 3.2 --- Utterance Verification for VIV --- p.29Chapter 3.2.1 --- Forced Alignment --- p.30Chapter 3.2.2 --- Subword Hypothesis Test --- p.30Chapter 3.2.3 --- Confidence Measure --- p.31Chapter 3.3 --- Sequential Utterance Verification for VIV --- p.34Chapter 3.3.1 --- Practical Security Consideration --- p.34Chapter 3.3.2 --- Robust Interval --- p.34Chapter 3.4 --- Application and Further Improvement --- p.36Chapter 3.5 --- Summary --- p.36Chapter 4 --- Model Design for Cantonese Verbal Information Verification --- p.37Chapter 4.1 --- General Considerations --- p.37Chapter 4.2 --- The Cantonese Dialect --- p.37Chapter 4.3 --- Target Model Design --- p.38Chapter 4.4 --- Anti-Model Design --- p.38Chapter 4.4.1 --- Role of Normalization Techniques --- p.38Chapter 4.4.2 --- Context-dependent versus Context-independent Antimodels --- p.40Chapter 4.4.3 --- General Approach to CI Anti-modeling --- p.40Chapter 4.4.4 --- Sub-syllable Clustering --- p.41Chapter 4.4.5 --- Cohort and World Anti-models --- p.42Chapter 4.4.6 --- GMM-based Anti-models --- p.44Chapter 4.5 --- Simulation Results and Discussions --- p.45Chapter 4.5.1 --- Speech Databases --- p.45Chapter 4.5.2 --- Effect of Model Complexity --- p.46Chapter 4.5.3 --- Comparisons among different Anti-models --- p.47Chapter 4.5.4 --- Discussions --- p.48Chapter 4.6 --- Summary --- p.49Chapter 5 --- Integration of SV and VIV --- p.50Chapter 5.1 --- Introduction --- p.50Chapter 5.2 --- Voting Method --- p.53Chapter 5.2.1 --- Permissive Test vs. Restrictive Test --- p.54Chapter 5.2.2 --- Shared vs. Speaker-specific Thresholds --- p.55Chapter 5.3 --- Support Vector Machines --- p.56Chapter 5.4 --- Gaussian-based Classifier --- p.59Chapter 5.5 --- Simulation Results and Discussions --- p.60Chapter 5.5.1 --- Voting Method --- p.60Chapter 5.5.2 --- Support Vector Machines --- p.63Chapter 5.5.3 --- Gaussian-based Classifier --- p.64Chapter 5.5.4 --- Discussions --- p.66Chapter 5.6 --- Summary --- p.67Chapter 6 --- Conclusions and Suggested Future Works --- p.68Chapter 6.1 --- Conclusions --- p.68Chapter 6.2 --- Summary of Findings and Contributions of This Thesis --- p.70Chapter 6.3 --- Future Perspective --- p.71Chapter 6.3.1 --- Integration of Keyword Spotting into VIV --- p.71Chapter 6.3.2 --- Integration of Prosodic Information --- p.71Appendices --- p.73Chapter A --- A Cantonese VIV Demonstration System --- p.73Bibliography --- p.7