Abstract
In forensic phonetic casework, it is recommended to perform speaker comparison using a combination of acoustic measurements and auditory analyses (see, e.g., Nolan, 1999). Differences in the examined parameters between two analyzed voices lend support to the hypothesis that different speakers are probably involved, while similar ranges of values of a given parameter or parameters lend support to the hypothesis that the voices might belong to one and the same speaker.
However, even if two voices are found to manifest similar values of one or more parameters, what remains to be examined is the typicality of the measured values (Jessen, 2012: 40ff.). This can only be achieved by relating the values to relevant population statistics.
It is clear that if, apart from being similar, the observed values are very typical - in other words, if the given parameter reaches similar values by the majority of the relevant population - the similarity between the two voices under analysis is of little explanatory power. Population statistics are so far available for several acoustic parameters in only few languages, for example the fundamental frequency (F0) in English (Hudson et al., 2007), German (Jessen et al., 2005) or Swedish (Lindh, 2006); and articulation rate in German (Jessen, 2007).
Creating such statistics requires the recording of a large number of speakers from the target population, processing the recordings, and extracting the given acoustic parameters, which constitutes quite a formidable task. Nolan's team has made the greatest stride in this direction with Standard British English and specifically their DyViS database (Nolan et al., 2009).
This paper presents the first steps towards creating a comparable database for the Czech language; at present there are no forensically applicable population statistics for Czech.