Winfried Krebber Vigo Rinas
[Inst. for Commun. Eng., Aachen Univ. of Technol. RWTH), Melatener Str. 23, D-52056 Aachen, Germany]
Motivated by the well-known identification problems in telephone communication, this study investigates the influence of filtering on the aural differentiation of related speakers e.g., father and son). Speech samples of couples of related speakers were presented to 30 listeners in 14 sessions 1 couple/session). Each session started with a reference sample to acquaint the listeners with one speaker of the couple. Subsequently the listeners received samples of both speakers and had to detect the samples belonging to the acquainted speaker. Additionally they gave information about the certainty of their decision using the terms ``surely,'' ``probably,'' or possibly.'' The following speech samples were used: voiced vowels and unknown) sentences, unfiltered and filtered with several bandpass filters; voiced vowels without higher formants only F1 left), and whispered vowels. In an additional opinion test the similarity of both voices of each couple was rated. The rates of false identifications depend clearly on the filter bandwidth. For voiced vowels there is a good correlation between the subjective similarity and the rates of false identifications, but not for sentences. Speech transmitted by usual telephone handsets, applied with an acoustical leakage to the ear, results in very high false identification rates up to 45%).