White Rose University Consortium logo
University of Leeds logo University of Sheffield logo York University logo

Assessing the strength of non-contemporaneous forensic speech evidence

Rhodes, Richard William (2012) Assessing the strength of non-contemporaneous forensic speech evidence. PhD thesis, University of York.

[img]
Preview
Text
R_W_Rhodes_-_Assessing_the_strength_of_non-contemporaneous_forensic_speech_evidence.pdf
Available under License Creative Commons Attribution-Noncommercial-No Derivative Works 2.0 UK: England & Wales.

Download (5Mb)

Abstract

The aim of this thesis is to assess the impact of long term non-contemporaneity on the strength of forensic speech evidence. Speakers experience age-related changes to the voice over long delays and this time also presents the opportunity for social factors to vary. These changes are shown to impact on speech parameters used in forensic analyses. Using longitudinal data from the Up documentary series, this thesis analyses the effects of aging on forensically useful acoustic parameters in eight speakers at five seven-year intervals between ages 21 and 49. The investigation reveals significant age-related changes in real-time across adulthood. Frequencies of the first three formants in monophthongs /i: ɪ e a ɑ: ʌ ɒ ʊ & u:/ and diphthongs /eɪ & aɪ/ show comprehensive reduction. For monophthongs, F1 exhibits mean change of 8.5%, greater than F2 and F3 at 3.7% and 2.2% respectively. Vowel quality also impacts on magnitude of change in each formant. Estimations based on this data suggest that vocal tract extension and restricted articulator movement are probable drivers for acoustic change, operating on different timelines. Counter-examples to this aging pattern can generally be explained by social factors, as a result of mobility or in accordance with mainstream changes in a variety. Strength of evidence estimates for these non-contemporaneous data are calculated using a numerical likelihood ratio (LR) approach. Age-related changes result in weaker and fewer correct LRs with greater length delays. Cubic coefficients of diphthong formants are investigated in line with a formant dynamic approach. These LR tests show promising results and resilience to aging, especially in F1; tentatively suggesting that, for these speakers, some speaker-specific behaviour pervades in spite of physiological changes. This analysis raises several questions with regards to applying an overtly numerical LR approach where there is apparent mismatch between forensic samples. The effect of aging on an ASR system (BATVOX) is also tested for six male subjects. The system measures Mel-frequency cepstral coefficient (MFCC) parameters that reflect the physical properties of the vocal tract. Predicted degradation of the system’s performance with increasing age is apparent. The reduction in performance is significant, varies between speakers, and is striking in longer delays for all speakers. The degradation in strength of evidence for acoustic data from monophthongs and formant dynamic coefficients, as well as that for the ASR system, demonstrates that aging presents a real problem for forensic analysis in non-contemporaneous cases. Furthermore, aging also presents issues for speech databases for the purpose of assessing strength of evidence, where further research into distributions of parameters in different age groups is warranted.

Item Type: Thesis (PhD)
Keywords: Forensic speech analysis, aging, non-contemporaneity, formant frequencies, ASR, likelihood ratio, mobility
Academic Units: The University of York > Language and Linguistic Science (York)
Identification Number/EthosID: uk.bl.ethos.572374
Depositing User: Mr Richard William Rhodes
Date Deposited: 20 May 2013 10:31
Last Modified: 08 Sep 2016 13:02
URI: http://etheses.whiterose.ac.uk/id/eprint/3935

You do not need to contact us to get a copy of this thesis. Please use the 'Download' link(s) above to get a copy.
You can contact us about this thesis. If you need to make a general enquiry, please see the Contact us page.

Actions (repository staff only: login required)