Learning to Recognize Talkers From Natural, Sinewave,and Reversed Speech SamplesSonya M. SheffertCentral Michigan UniversityDavid B. PisoniIndiana University BloomingtonJennifer M. FellowesColumbia UniversityRobert E. RemezBarnard CollegeIn 5 experiments, the authors investigated how listeners learn to recognize unfamiliar talkers and howexperience with specific utterances generalizes to novel instances. Listeners were trained over severaldays to identify 10 talkers from natural, sinewave, or reversed speech sentences. The sinewave signalspreserved phonetic and some suprasegmental properties while eliminating natural vocal quality. Incontrast, the reversed speech signals preserved vocal quality while distorting temporally based phoneticproperties. The training results indicate that listeners learned to identify talkers even from acoustic signalslacking natural vocal quality. Generalization performance varied across the different signals and de-pended on the salience of phonetic information. The results suggest similarities in the phonetic attributesunderlying talker recognition and phonetic perception.When a talker produces an utterance, the listener simultaneouslyapprehends the linguistic form of the message as well as thenonlinguistic attributes of the talker’s unique vocal anatomy andpronunciation habits. Anatomical and stylistic differences in artic-ulation convey an array of personal or indexical qualities, such aspersonal identity, sex, approximate age, ethnicity, personality,intentions or emotional state, level of alcohol intoxication, andfacial expression (see Bricker & Pruzansky, 1976; Chin & Pisoni,1997; Cook & Wilding, 1997; Doddington, 1985; Kreiman, 1997;Scherer, 1986; Tartter, 1980; Walton & Orlikoff, 1994).Personal characteristics play an important role in communica-tive interactions. This is especially true for listeners who areunable to use indexical attributes available in other modalities as aresult of neurological impairments in face recognition ( prosopag-nosia: Benton & Van Allen, 1968; Bodamer, 1947; Damasio,Damasio, & Van Hoesen; 1982) or visual impairments (Bull,Rathborn, & Clifford, 1983; Yarmey, 1986). Over the course ofa lifetime, listeners acquire very detailed and enduring know-ledge about many different talkers. The ability to recognize atalker begins in utero (Hepper, Scott, & Shahidullah, 1993) and de-velops rapidly throughout infancy and childhood (DeCasper &Fifer, 1980; Jusczyk, Hohne, Jusczyk, & Redanz, 1993; Mandel,Jusczyk, & Pisoni, 1995), reaching adult levels of proficiency byage 10 (Mann, Diamond, & Carey, 1979).An extensive literature on human talker recognition dates to thework of McGehee (1937), who examined the reliability of ear-witness testimony, and the studies by Peters (1955) and Pollack,Pickett, and Sumby (1954), who examined laboratory effects oflinguistic content on talker recognition. This literature describesthe effects of acoustic, procedural, and individual attributes thataffect the recognition and discrimination of unfamiliar talkers (seeBricker & Pruzansky, 1976; Clifford, 1980; Hecker, 1971;Kreiman, 1997; and Read & Craik, 1995, for reviews).In contrast, much less is known about how a listener recognizesa familiar talker beyond the benchmarks that reveal perceptual,cognitive, and neural differences in the classification of familiarand unfamiliar talkers (Papc¸un, Kreiman, & Davis, 1989; Schmidt-Nielsen & Stern, 1985; Schweinberger, Herholz, & Sommer, 1997;Van Lancker & Canter, 1982; Van Lancker & Kreiman, 1987; VanLancker, Kreiman, & Cummings, 1989). Moreover, few studieshave examined how a listener becomes familiar with a talker(Legge, Grosmann, & Pieper, 1984; Nygaard & Pisoni, 1998;Nygaard, Sommers, & Pisoni, 1994). These studies show thatrepeated or extended exposure to a talker’s speech increases alistener’s sensitivity to talker-specific attributes, improving theEditor’s Note. Carol Fowler served as the action editor for this article.—DARSonya M. Sheffert, Psychology Department, Central Michigan Univer-sity; David B. Pisoni, Department of Psychology, Indiana UniversityBloomington; Jennifer M. Fellowes, Department of Psychiatry, ColumbiaUniversity; Robert E. Remez, Department of Psychology, Barnard College.A portion of these findings was presented at the 134th Annual Meetingof the Acoustical Society of America, December 1997, San Diego, Cali-fornia. This research was supported by National Institute on Deafness andOther Communicative Disorders Grant DC00111 to Indiana UniversityBloomington and Grant DC00308 to Barnard College.We thank Luis Hernandez for technical support, Nathan Large forassisting with data collection for Experiments 3–5, and Dalia Shoretz andRebecca Piorkowski for lending scholarly and technical assistance to theproject. We give special thanks to Vivian Tartter and two anonymousreviewers for their valuable comments on earlier versions of this article.Correspondence concerning this article should be addressed to Sonya M.Sheffert, Psychology Department, Central Michigan University, Sloan Hall214, Mount Pleasant, Michigan 48859. E-mail: [email protected] of Experimental Psychology: Copyright 2002 by the American Psychological Association, Inc.Human Perception and Performance2002, Vol. 28, No. 6, 1447–14690096-1523/02/$5.00 DOI: 10.1037//0096-1523.28.6.144714471 LINE SHORTability to differentiate familiar from unfamiliar talkers. Left un-specified, however, are the properties of the speech signal that aremost relevant for learning and recognizing familiar talkers fromnovel utterances.The research described in this article investigated the recogni-tion of familiar talkers, examining the contribution of differenttalker-specific properties of a speech signal to perceptual learning.To set the task in this experimental design, we trained our listenersto identify different talkers using signals that were acousticallymodified to preserve different properties that were arguably talker-specific. Listeners heard sentence-length natural, sinewave, orreversed speech samples. Their knowledge of the talker was thenassessed using generalization tests in which a novel set of natural,sinewave, or reversed speech samples were used and listeners wereasked again to identify the talkers. Our intention was to permit acomparison of the attributes available in the learning conditionsand in the generalization tests with those proposed in severalclassic and recent accounts of individual
View Full Document
Unlocking...