ICPhS XVII Regular Session Hong Kong, 17-21 August 2011 663 SPEAKING UNDER COVER: THE EFFECT OF FACE-CONCEALING GARMENTS ON SPECTRAL PROPERTIES OF FRICATIVES Natalie Fecher & Dominic Watt Department of Language and Linguistic Science, University of York, York, UK natalie.fecher@york.ac.uk; dominic.watt@york.ac.uk ABSTRACT This paper firstly reports on the design of an audio- visual ‘face cover’ corpus. High-quality audio and video recordings were taken of 10 speakers reading phonetically-controlled stimuli under various face disguise conditions. Possible articulatory, acoustic and perceptual effects of the masks in a forensic context are introduced. Secondly, preliminary results of a spectral analysis of voiceless fricatives, taken from a subset of speakers, are presented. Keywords: forensic phonetics, acoustics, disguise, fricatives, spectral moments 1. INTRODUCTION Forensic speech science (FSS) depends, among other things, on information about a speaker’s speech and language gathered from questioned audio or video material, the quality and quantity of which is often limited. To account for discrepancies between real-life casework and the empirical research that underpins it, FSS experts are increasing their efforts to simulate more realistic scenarios when conducting experiments. A three-year project has been initiated as part of the interdisciplinary Marie Curie Initial Training Network ‘Bayesian Biometrics for Forensics’ [1], focusing on the influence of forensically-relevant face-concealing garments 1 (henceforth FCGs) on the physiological, acoustic and linguistic levels of the speech chain. The project investigates both human and machine performance during speech and speaker recognition tasks under visually and acoustically degraded conditions. Where an FCG obstructs the talker’s face we could plausibly anticipate effects in three domains: 1.1. Misarticulation and compensation In the first domain (speech production), we might expect FCGs to interfere in various ways with speech articulation. Firstly, misarticulations could be attributed to physiological and somatosensory effects, such as lip/nose contact, restricted jaw elevation [5] and skin stretching [4, 10]. As the same FCGs were used for all speakers in this study, the subjects’ head sizes determined how tightly some FCGs were attached to the speakers’ faces and articulators. Simultaneously, each subject might reveal idiosyncratic articulatory compensation strategies, like an increase in vocal effort [2]. For certain FCGs, compensatory phenomena could also result from the speakers’ ears being covered, impairing auditory self- monitoring. As physio-logical and acoustic events in the vocal tract are interdependent, effects in this domain will alter the acoustic signal. For instance, a perturbed lip protrusion in [ʃ] due to a mask being in contact with the speaker’s lips may shorten the front tube of the vocal tract, leading to frequency shifts. 1.2. Acoustic damping effects The different mask materials will modify the acoustic properties of the signal by affecting the sound transmission/absorption characteristics to varying degrees. The FCGs are assumed to act like a low-pass filter, attenuating the level of sound energy in higher frequency bands [2, 8, 17]. 1.3. Impaired recognition and visual speech FCGs impose a new level of complexity in the listener’s search for perceptual cues in the signal. We propose that the factors in (i) and (ii) impose significant cognitive demands in audio-visual (AV) speech perception. Impaired intelligibility of AV stimuli is anticipated because of interference with speech production and the acoustic signal [8, 17], compounded by impoverished visual cues (see e.g. [12, 13, 14]). In this paper we report on the design of an AV database as well as preliminary findings from the acoustic domain, i.e. the influence of FCGs on the spectral properties of voiceless fricatives.