Abstract
Voice quality parameters have not been investigated to a great extent in technical speaker identification tasks, in spite of the fact that forensic phoneticians appear to make rather frequent use of voice quality in their casework (Nolan, 2005; Gold & French, 2011). The main reason for the lack of acoustic investigations appears to be the fact that the presence of especially laryngeal voice quality features is compromised in telephone speech (Nolan, 2005).
In addition, the plasticity of our voice production mechanism allows for great stylistically conditioned variability, and it is mostly voice quality which is affected. Yet we believe that there still is space for acoustic examinations of speaker specificity of voice quality, especially of its short-term correlates which reflect spectral slope by comparing the amplitudes of various events in the acoustic spectrum (Hanson et al., 2001).
The motivation for using the parameters H1*-H2*, H1*-A1*, H1*-A2*, H1*-A3* and H2*-H4* is twofold: first, it appears that some of them yield favourable rates of intra-speaker stability and inter-speaker variability (Vaňková and Skarnitzl, 2014); second, low frequencies relevant for H1 are actually not filtered out by the Adaptive Multi-Rate (AMR) codec, which is the current standard in mobile telephony (Guillemin and Watson, 2008; Vaňková and Bořil, submitted).