Bayesian Strategies for Likelihood Ratio Computation in Forensic Voice Comparison with Automatic Systems
This work addresses forensic voice comparison for legal applications, but it is incremental as it builds on existing generative Gaussian models.
The paper tackled the problem of improving likelihood ratio computation in forensic voice comparison by proposing anchoring strategies and a fully-Bayesian Gaussian model to handle sparse training scores, resulting in the Bayesian model outperforming a Maximum-Likelihood approach with high robustness.
This paper explores several strategies for Forensic Voice Comparison (FVC), aimed at improving the performance of the LRs when using generative Gaussian score-to-LR models. First, different anchoring strategies are proposed, with the objective of adapting the LR computation process to the case at hand, always respecting the propositions defined for the particular case. Second, a fully-Bayesian Gaussian model is used to tackle the sparsity in the training scores that is often present when the proposed anchoring strategies are used. Experiments are performed using the 2014 i-Vector challenge set-up, which presents high variability in a telephone speech context. The results show that the proposed fully-Bayesian model clearly outperforms a more common Maximum-Likelihood approach, leading to high robustness when the scores to train the model become sparse.