2022P Poster: The speaker in speech – linguistic context in forensic speaker comparisons

In forensic speaker comparisons, one or more disputed speech samples of an unknown speaker are typically compared to one or more speech samples of a suspect. One concern in such analyses is the effect of phonetic-linguistic context on the speech sounds sampled for comparison; many phonetic studies show that contexts affect the acoustic realization of speech sounds. For example, stressed syllables evoke more canonical pronunciations, utterance-initial speech sounds are typically produced with more effort, and talking over the telephone changes speech behaviour.

To better understand the relevance of these acoustic effects for forensic speaker comparisons, we have been studying how segmental features that characterize speakers’ voices depend on what speakers are saying (NWO VIDI project 276-75-010). More specifically, how does the speaker-specificity of speech sounds depend on its direct context (Heeren, 2020, Smorenburg & Heeren, 2020, 2021), on the speech channel (Smorenburg & Heeren, 2022), and on the language spoken (De Boer & Heeren, 2020; De Boer, Quené & Heeren, 2022). In this overview presentation we would like to share some of our main project results obtained on existing research databases (CGN, Oostdijk, 2000; D-LUCEA, Orr & Quené, 2017; WYRED, Gold et al., 2018), and show how our findings are relevant for and translate to forensically-realistic speech data taken from the NFI-FRIDA corpus (Van de Vloed et al., 2020).

References

Boer M.M. de, Quené H. & Heeren W.F.L. (2022), Long-term within-speaker consistency of filled pauses in native and non-native speech, JASA Express Letters 2(3): 035201.
Boer M.M. de & Heeren W.F.L. (2020), Cross-linguistic filled pause realization: the acoustics of uh and um in native Dutch and non-native English, Journal of the Acoustical Society of America 148(6): 3612-3622.
Gold, E., Ross, S., & Earnshaw, K. (2018). The “West Yorkshire Regional English Database”: Investigations into the generalizability of reference populations for forensic speaker comparison casework. In Proceedings of INTERSPEECH (Vol. 2018–Sept., pp. 2748–2752)
Heeren W.F.L. (2020), The Effect of Word Class on Speaker-dependent Information in the Standard Dutch Vowel /aː/, Journal of the Acoustical Society of America 148(4): 2028-2039.
Oostdijk, N. (2000). “Het corpus Gesproken Nederlands,” (The Spoken Dutch Corpus), Nederlandse Taalkunde 5, 280–284.
Orr, R., and Quené, H. (2017). “ D-LUCEA: Curation of the UCU Accent Project data,” in CLARIN in the Low Countries, edited by J. Odijk and A. van Hessen ( Ubiquity Press, Berkeley), pp. 177–190.
Smorenburg B.J.L. & Heeren W.F.L. (2020), The distribution of speaker information in Dutch fricatives /s/ and /x/ from telephone dialogues, Journal of the Acoustical Society of America 147(2): 949-960.
Smorenburg B.J.L. & Heeren W.F.L. (2021), Acoustic and speaker variation in Dutch /n/ and /m/ as a function of phonetic context and syllabic position, Journal of the Acoustical Society of America 150(2).
Smorenburg L. & Heeren W.F.L. (2022), The effect of linguistic contexts on the acoustics and strength-of-evidence of /s/. Proceedings 30th annual conference of the International Association for Forensic Phonetics and Acoustics, p. 21-22.
Van der Vloed, D., Bouten, J., Kelly, F., & Alexander, A. (2018). NFI-FRIDA–Forensically realistic interdevice audio database and intial experiments. In 27th Annual Conference of the International Association for Forensic Phonetics and Acoustics (IAFPA) (pp. 25-27).