Fusing linguistic and acoustic information for automated forensic speaker comparison.

Verifying the speaker of a speech fragment can be crucial in attributing a crime to a suspect. The question can be addressed given disputed and reference speech material, adopting the recommended and scientifically accepted likelihood ratio framework for reporting evidential strength in court. In forensic practice, usually, auditory and acoustic analyses are performed to carry out such a verification task considering a diversity of features, such as language competence, pronunciation, or other linguistic features. Automated speaker comparison systems can also be used alongside those manual analyses. State-of-the-art automatic speaker comparison systems are based on deep neural networks that take acoustic features as input. Additional information, though, may be obtained from linguistic analysis. In this paper, we aim to answer if, when and how modern acoustic-based systems can be complemented by an authorship technique based on frequent words, within the likelihood ratio framework. We consider three different approaches to derive a combined likelihood ratio: using a support vector machine algorithm, fitting bivariate normal distributions, and passing the score of the acoustic system as additional input to the frequent-word analysis. We apply our method to the forensically relevant dataset FRIDA and the FISHER corpus, and we explore under which conditions fusion is valuable. We evaluate our results in terms of log likelihood ratio cost (Cllr) and equal error rate (EER). We show that fusion can be beneficial, especially in the case of intercepted phone calls with noise in the background.

From data to a validated score-based LR system: A practitioner's guide.

Likelihood ratios (LRs) are a useful measure of evidential strength. In forensic casework consisting of a flow of cases with essentially the same question and the same analysis method, it is feasible to construct an 'LR system', that is, an automated procedure that has the observations as input and an LR as output. This paper is aimed at practitioners interested in building their own LR systems. It gives an overview of the different steps needed to get to a validated LR system from data. The paper is accompanied by a notebook that illustrates each step with an example using glass data. The notebook introduces open-source software in Python constructed by NFI (Netherlands Forensic Institute) data scientists and statisticians.

Objectifying evidence evaluation for gunshot residue comparisons using machine learning on criminal case data.

Comparative gunshot residue analysis addresses relevant forensic questions such as 'did suspect X fire shot Y?'. More formally, it weighs the evidence for hypotheses of the form H1: gunshot residue particles found on suspect's hands are from the same source as the gunshot residue particles found on the crime scene and H2: two sets of particles are from different sources. Currently, experts perform this analysis by evaluating the elemental composition of the particles using their knowledge and experience. The aim of this study is to construct a likelihood-ratio (LR) system based on representative data. Such an LR system can support the expert by making the interpretation of the results of electron microscopy analysis more empirically grounded. In this study we chose statistical models from the machine learning literature as candidates to construct this system, as these models have been shown to work well for large and high-dimensional datasets. Using a subsequent calibration step ensured that the system outputs well-calibrated LRs. The system is developed and validated on casework data and an additional validation step is performed on an independent dataset of cartridge data. The results show that the system performs well on both datasets. We discuss future work needed before the method can be implemented in casework.

Establishing phone-pair co-usage by comparing mobility patterns.

In forensic investigations it is often of value to establish whether two phones were used by the same person during a given time period. We present a method that uses time and location of cell tower registrations of mobile phones to assess the strength of evidence that any pair of phones were used by the same person. The method is transparent as it uses logistic regression to discriminate between the hypotheses of same and different user, and a standard kernel density estimation to quantify the weight of evidence in terms of a likelihood ratio. We further add to previous theoretical work by training and validating our method on real world data, paving the way for application in practice. The method shows good performance under different modeling choices and robustness under lower quantity or quality of data. We discuss practical usage in court.