SPEAKER RECOGNITION

This research is concerned with defining a per ceptual space within which listeners locate voices, to the end that the effects of manipu lating speaker, hardware, and listener charac teristics can be measured, and eventually, that specifications for elements of the communication system can be prepared to produce the desired system characteristics. In the experiments taped speech samples were rated by listeners using Osgoods semantic differential method. Previous study indicated only four basis dimen sions were required to account for ratings given speakers on a large number of characteristics. In a second experiment, a reduced number of characteristics, selected from the original list as best representing the four necessary factors, was used by listeners to rate speakers from AFCRLs speaker library. The experimental design allowed examination of the effects on ratings due to differences between listeners, due to repe tion of the rating task, and to order of speaker presentation. Results of these examinations and the following are presented The adequacy of original factors to account for listeners ratings The differentiation between speakers The reliability of ratings The familiarity of previously unheard voices.