Speaker-invariant suprasegmental temporal features in normal and disguised speech

Abstract

Introduction Voice disguise refers to the voluntary act of changing one"s voice as a means to mask identity. In some forensic cases, perpetrators disguise their voices. The German Federal Police Office (BKA) reports that cases involving voice disguise range between 12.6% (Wagner, 1998) and 22.9% (Braun, 2009). Current research distinguishes between two types of voice disguise: electronic voice disguise, e.g. a perpetrator applies an electronic voice changer, and non-electronic voice disguise (Masthoff, 1996; Künzel, 2000; Perrot et al., 2007). Existing research on the former type recognizes the critical role of electronic voice disguise on voice identification: an artificial change in voice fundamental frequency, for example, negatively affects identification performance (Clark and Foulkes, 2007). Non-electronic disguise, however, occurs more frequently and can typically be broken up into (a) a change in voice source characteristics (e.g. raised pitch, lowered pitch, whispered speech), (b) a change in resonance features (e.g. foreign object in vocal tract, hypo-/hypernasality, tissue held to mouth), (c) a change in language, accent, and dialect (Künzel, 2000; Perrot et al., 2007; Neuhauser, 2011a). Previous research has shown that voice disguise can lead to high withinspeaker variability, affecting various acoustic parameters. Hollien and Majewksi (1977) reported high within-speaker variability in long-term spectra under various voice disguise conditions (disguise of the speaker"s own choice). Face covering as a means of voice disguise has been reported to affect spectral characteristics of fricatives (Fecher and Watt, 2011) and the amplitude of the overall signal (Llamas et al., 2009). Hirson and Duckworth (1993) showed that the intentional application of creaky voice as a disguise affects spectral characteristics of /s/. A further strategy for disguising one"s voice is to imitate another speaker. Eriksson and Wretling (1997) and Endres et al. (1971) reported that speakers particularly aim at matching f0 and formant frequencies when asked to imitate another speaker. Most often, this f0 and formant matching is successful. High within-speaker variability-such authentic or disguised voices-makes it difficult for forensic practitioners to draw conclusions about the speakers" identity. It is thus desirable in forensic cases that the acoustic features examined reveal maximum between-speaker and minimum within-speaker variability (Nolan, 2009). How suprasegmental temporal features behave in voice-disguised speech is currently unknown. The objective of this contribution is to test if and how temporal features that are not restricted to a single segment but rather to the more global temporal organization of speech remain speaker-invariant in one form of disguised speech, namely in cases where a different dialect is imitated. Such features have shown promising trends in this respect: preliminary research in the context of voice disguise has shown that suprasegmental temporal features such as the amount of voicing in the s...

Perceptual Evaluation of the Effectiveness of Voice Disguise by Age Modification

“…Lombard reflex under noisy environments), whereas deliberate modification is actively enforced by the speaker so that he or she is fully aware of it. Voice acting, disguise and impersonation [6,7,8] are good examples of this.…”

Voice Disguise vs. Impersonation: Acoustic and Perceptual Measurements of Vocal Flexibility in Non Experts

“…Particularly documented are modifications in fundamental frequency [1,4,5,6,7], formant frequencies mean and variability [1,8,9,10], and speech rate [6,10,11]. On the other hand, the temporal organization of speech may be more resistant to vocal disguise [12,13].…”

Multiple Frame Rates for Feature Extraction and Reliable Frame Selection at the Decision for Speaker Identification Under Voice Disguise

“…Features related to the fundamental frequency F0, outperformed other features. In a recent study [21], the effect of voice disguise on suprasegmental features has been studied. It showed low variability within speakers and high variabilty between speakers.…”