Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

What is the relevant population for forensic voice comparison evidence?

52 views

Published on

Hughes, V. (2014) What is the relevant population for forensic voice comparison evidence? Paper presented at Department of Language and Linguistic Science Postgraduate Presentation Day, University of York, UK. 30 May 2014.

Published in: Education
  • Be the first to comment

  • Be the first to like this

What is the relevant population for forensic voice comparison evidence?

  1. 1. What is the relevant population for forensic voice comparison evidence? Vincent Hughes Department of Language and Linguistic Science Postgraduate Presentation Day 30th May 2014
  2. 2. 2 1.1 Forensic voice comparison example: offender sample (unknown speaker) I’ve come to see the lady at number two (I’m fro)m the Home Care I’ve come to collect her sheet suspect sample (known speaker) (I’ve) come to see [name removed] from number two Home Care come to collect the sheets PG Presentation Day 30th May 2014
  3. 3. 3 1.1 Forensic voice comparison question for the court (judge/ jury): is the person on the criminal recording the same as the person on the suspect recording? • to answer this question an expert is asked to conduct a comparison of the two samples: – componential auditory-acoustic ling-phon analysis – used by 74% of experts worldwide (French & Gold 2011) PG Presentation Day 30th May 2014
  4. 4. 4 1.2 Auditory-acoustic analysis feature notes Vowels English 24; different patterns for specific phonological environments; acoustic features (formants) Consonants English 20; different patterns for specific phonological environments; acoustic features (fricatives and stop bursts); segment durations Vocal setting Laver VPA scheme: 38 separate elements Intonation constrained by phonology& discourse Pitch mean, range, s.d. ...
  5. 5. 5 1.2 Auditory-acoustic analysis feature notes Articulation rate speed of speaking Rhythm timing by syllable/stress Tone for languages with contrastivetone Connected speech processes assimilation,elision... Discourse/ Pragmatics discourse markers, turn-taking,, code switching… Non-linguistic audible breathing, throat-clearing, clicking, hesitation phenomena...
  6. 6. 6 1.3 Voice as a biometric • no vocal equivalent of a fingerprint/DNA profile – plasticityof vocal tract (high within- speaker variation) – external influences: health, environment, intoxicants, learned behaviour (style, dialect, accent etc.) – samples will always differ, even if the same person says the same thing • therefore we can’t talk about finding a ‘match’ between the suspect and offender (unlike on CSI) PG Presentation Day 30th May 2014
  7. 7. 7 1.4 Likelihood ratio (LR) • the role of the expert is to evaluate the evidence under the competing hypotheses: – prosecution = same speaker – defence = different speakers • LR = indicates whether the evidence supports the prosecution or the defence: – together with a gradient assessment of the strength/ weight of evidence ✓ PG Presentation Day 30th May 2014
  8. 8. 8 1.4 Likelihood ratio • similarity: how similar are the suspect and offender samples to each other? • typicality: how common or rare are the features that we find in the suspect and offender samples? – e.g. pronunciation of /r/ in words like car and farm = common in Scotland/ (relatively) rare in England ✓ PG Presentation Day 30th May 2014
  9. 9. 9 1.5 Relevant population • typicality = dependent on patterns in the “relevant population” (Aitken & Taroni 2004) but… paradox = without knowing who the offender is we can’t know (for sure) the population of which (s)he is a member PG Presentation Day 30th May 2014
  10. 10. 10 2.1 Current approach logical relevance • factors which affect the distribution a variable in the population (determine sub-populations) - approach used in DNA analysis (ethnicity) • Rose (2004: 4): – speakers of same sex (incorrectly termed ‘gender’) and language as the offender – reflected in the majority of LR-based studies (e.g. Morrison 2009) and in casework (Rose 2013) PG Presentation Day 30th May 2014
  11. 11. 11 2.1 Current approach PG Presentation Day 30th May 2014 but… numerous sources of potentially relevant between-speaker variation: • regional background • sex (and gender) • ethnicity • age • socio-economic class • social networks etc.
  12. 12. 12 2.2 Emperimental data • based on acoustic analysis of /aɪ/ in words like price, bike, sight… • using two relevant populations: a. matched: SSBE b. mixed: general British English PG Presentation Day 30th May 2014 1 2 3 1 2 3
  13. 13. 13 PG Presentation Day 30th May 2014 results using the mixed population: • c. 20% more DS pairs classified as SS - i.e. innocent classified as guilty • c. 5% more SS pairs classified as DS - i.e. guilty classified as innocent • evidence on average between 10 and 100 times weaker compared with the results using the matched population
  14. 14. 14 3. Conclusions • definition of the relevant population has a big effect on justice - understanding of sociolinguistic variation can help to improve the quality of forensic science • but… we still have the paradox – solution = multiple LRs based on different assumptions about the offender – this approach is used in DNA analysis PG Presentation Day 30th May 2014
  15. 15. Thanks! Questions? Acknowledgements: Paul Foulkes, Erica Gold, Peter French, Dom Watt, Ashley Brereton, FSS Research Group (York) 15

×