Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

Are Representations Used for Talker Identification Available for Talker Normalization?

James S. Magnuson (1), Reiko A. Yamada (1), Howard C. Nusbaum (2)

(1) ATR Human Information Processing Laboratories, Kyoto, Japan
(2) University of Chicago Department of Psychology, Chicago, IL, USA

Contextual tuning theories of talker normalization state that listeners can use information about a talker's vocal characteristics stored in working memory to recognize that talker's speech [8], We investigated whether people can use information about a familiar talker's voice, stored in long-term memory [10], in the same way. That is, whether people can circumvent talker normalization processes when listening to familiar talkers by referencing the representations they use for talker identification. We presented subjects with stimuli produced by familiar and unfamiliar talkers in a monitoring paradigm that typically results in faster performance in a single-talker condition than a multiple-talker condition. We found the typical normalization effect for both familiar and unfamiliar talkers, suggesting that even if talker representations used for identification are compatible with those used for normalization, they cannot be retrieved more quickly than the representations used for normalization can be computed. We verified subjects ability to identify familiar talkers in a second experiment, and found that familiarity facilitated both accuracy and response time in the identification task. We discuss the implications of the results for theories of talker normalization and talker identification.

Full Paper

Bibliographic reference.  Magnuson, James S. / Yamada, Reiko A. / Nusbaum, Howard C. (1994): "Are representations used for talker identification available for talker normalization?", In ICSLP-1994, 1175-1178.