Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2022 Mar 7:4:100223.
doi: 10.1016/j.fsisyn.2022.100223. eCollection 2022.

Validations of an alpha version of the E3 Forensic Speech Science System (E3FS3) core software tools

Philip Weber  1   2 , Ewald Enzinger  1   2   3 , Beltrán Labrador  4 , Alicia Lozano-Díez  4 , Daniel Ramos  4 , Joaquín González-Rodríguez  4 , Geoffrey Stewart Morrison  1   2
Affiliations

Validations of an alpha version of the E3 Forensic Speech Science System (E3FS3) core software tools

Philip Weber et al. Forensic Sci Int Synerg. .

Abstract

This paper reports on validations of an alpha version of the E3 Forensic Speech Science System (E3FS3) core software tools. This is an open-code human-supervised-automatic forensic-voice-comparison system based on x-vectors extracted using a type of Deep Neural Network (DNN) known as a Residual Network (ResNet). A benchmark validation was conducted using training and test data (forensic_eval_01) that have previously been used to assess the performance of multiple other forensic-voice-comparison systems. Performance equalled that of the best-performing system with previously published results for the forensic_eval_01 test set. The system was then validated using two different populations (male speakers of Australian English and female speakers of Australian English) under conditions reflecting those of a particular case to which it was to be applied. The conditions included three different sets of codecs applied to the questioned-speaker recordings (two mismatched with the set of codecs applied to the known-speaker recordings), and multiple different durations of questioned-speaker recordings. Validations were conducted and reported in accordance with the "Consensus on validation of forensic voice comparison".

Keywords: Forensic voice comparison; Likelihood ratio; Validation; x-vector.

PubMed Disclaimer

Conflict of interest statement

Dr Morrison is the Director of Forensic Evaluation Ltd, and Dr Weber and Dr Enzinger have worked for Forensic Evaluation Ltd on a contract basis.

Figures

Fig. 1
Fig. 1
High-level architecture of E3FS3α.
Fig. 2
Fig. 2
Simplified schematic of the architecture of the ResNet used for extracting x-vectors.
Fig. 3
Fig. 3
Simplified schematic of the feature vectors and the input layer of the ResNet used for extracting x-vectors. Only one channel is shown.
Fig. 4
Fig. 4
Simplified schematic of the architecture of a block within the ResNet used for extracting x-vectors. Only one channel is shown.
Fig. 5
Fig. 5
Simplified schematic of the final stages of the ResNet used for extracting x-vectors. In this figure “ × ” indicates matrix multiplication.
Fig. 6
Fig. 6
Tippett plot of the results of validating E3FS3α using the forensic_eval_01 data.
Fig. 7
Fig. 7
Cllr values for case-specific conditions. (a) Male speakers. (b) Female speakers.
Fig. 8
Fig. 8
Tippett plots of the results of validating E3FS3α using case-specific data (female Australian-English speakers and questioned-speaker condition μ-law + G.729a) at different questioned-speaker net-speech durations.
See this image and copyright information in PMC

References

    1. Morrison G.S. Distinguishing between forensic science and forensic pseudoscience: testing of validity and reliability, and approaches to forensic voice comparison. Sci. Justice. 2014;64:245–256. doi: 10.1016/j.scijus.2013.07.004. - DOI - PubMed
    1. Morrison G.S., Enzinger E., Ramos D., González-Rodríguez J., Lozano-Díez A. In: Handbook of Forensic Statistics. Banks D.L., Kafadar K., Kaye D.H., Tackett M., editors. Chapman and Hall/CRC; 2020. Statistical models in forensic voice comparison; pp. 451–497. ch. 20. - DOI
    1. Essential Scientific Literature for Human-Supervised Automatic Approaches to Forensic Speaker Recognition. Organization of Scientific Area Committees for Forensic Science; 2021. Speaker Recognition Subcommittee of the Organization of Scientific Area Committees for Forensic Science.https://www.nist.gov/document/essentialscientificliteratureforhuman
    1. Morrison G.S., Enzinger E., Hughes V., Jessen M., Meuwly D., Neumann C., Planting S., Thompson W.C., van der Vloed D., Ypma R.J., Zhang C., Anonymous A., Anonymous B. Consensus on validation of forensic voice comparison. Sci. Justice. 2021;61:299–309. doi: 10.1016/j.scijus.2021.02.002. - DOI - PubMed
    1. Weber P., Enzinger E., Morrison G.S. E3 Forensic Speech Science System (E3FS3): technical report on design and implementation of software tools. 2022. http://e3fs3.forensic-voice-comparison.net/ Available at. - PMC - PubMed

LinkOut - more resources