The speaker in speech – the interdependence of linguistic and indexical information

Samenvatting

Forensic speaker comparisons assess whether the speaker in a questioned recording, say a threatening telephone call, is the same as in a recording of the suspect. Until recently, speaker comparisons focused on the similarity of speech features in the questioned and suspect samples, but an international paradigm shift in the evaluation of forensic evidence is now requiring experts to also include the typicality of speech features in the evaluation. This means that the distributions of relevant speech features, which are used to compare samples, must be known and quantified within and between speakers. The necessary population statistics, however, are mostly lacking. More fundamentally, the underlying relationship between speaker-dependent and linguistic information is poorly understood.

To discover how speaker-specific information is distributed across the speech signal and to make a significant contribution to population statistics, this project studies the relation between linguistic and speaker-dependent, i.e. indexical, information in speech. This is done by investigating how and why the speaker-specificity of speech sounds depends on different aspects of the spoken message: the speech sound’s prosodic, phrasal or syllabic position, the word it occurs in, the speech channel, and the language spoken. Since single speech features insufficiently discriminate among speakers, we investigate not only potential new features, but also how features combine into optimally speaker-specific models. To answer these research questions we undertake comparative analyses using available large corpora of Dutch and English speech, including face-to-face and telephone conversation.

This project will further our fundamental understanding of the interaction between the indexical and linguistic layers of information in speech. The resulting modelling of multi-lingual and multi-dimensional within-speaker and between-speaker variation of phonetic-acoustic speech features will constitute a timely and critical contribution to the advancement of the state of the art in forensic speech science.

Output

Publieksinformatie

Kenmerken

Projectnummer

276-75-010

Hoofdaanvrager

Dr. W.F.L. Heeren

Verbonden aan

Universiteit Utrecht, Faculteit Geesteswetenschappen, Letteren

Uitvoerders

Dr. W.F.L. Heeren, Dr. W.F.L. Heeren, L. Smorenburg, L. Smorenburg, M.M. de Boer

Looptijd

01/10/2017 tot 01/11/2022