885095967

1:15

9SP2. The intelligibility of words, sentences, and continuous discourse using the articulation index. Rory DePaolis (Graduate Próg. in Acoust., Penn State Univ., University Park, PA 16803)

The relation betwccn the intdhgibility of speech stimulus with widely varying message redundancy was measured. Using the original articulation index (Al) methodology [N. R. French and J. C. Steinberg, J. Acoust. Soc. Am. 19, 90-119 (1947)], freąuency importance func-tions and transfer functions (Al versus percent correct) were deter-mined for one speaker speaking 616 PB-50 words, 200 meaningful SPIN sentences, and 44 7th-grade reading level continuous discourse, (CD) passages. Thirty subjects were instructed to write down each word and to estimale the pcrcentage of each sentence and CD passage that they heard correctly. The stimulus was degraded with 4 noise and 11 filtered conditions. The results demonstrate a trend towanl less significance of high-frequency cues as message redundancy inereases. There is also evidence to recommend the use of speech typc specific frequcncy importance functions when calculating the Al.

1:30

9SP3. A probability model for distributions of speech intelligibility data. Caldwell P. Smith (Consultant, 378 Chicopce Row, Groton, MA 01450)

It was determined that the compound Poisson probability distribu-tion as described by William Feller in An Introduction to Probability Theory and Its Applications, 2nd ed. [(Wiley, New York, 1975), pp. 270-273] is a valid model for distributions of speech intelligibility scores from diagnostic rhyme tests. This was established from details of scores from 110 multispcaker tests of a variety of speech processing conditions. Probability models were constructed by first converting feature scores to integers representing frequencies of errors in listener responses, and calculating means and variances of those distributions. Variance of a compound Poisson distribution is equal to the mean divided by p, and in this corpus of data the value of p tended to remain relatively fixed at an average value of 0.129, with thc consequence that distributions were esscntially defined by mean values and dispersions a linear function of means. In these measures, variance averaged 7.75 times the mean, with the average value of this cocfficient varying over a limited rangę with different speech processing conditions: for LPC processors, the average was 8.37; for wideband processors, 7.12; for processors in tandem, 6.75; and for speech in Gaussian noise, 8.06. Partitioned into separate data sets for voiced and unvoiced feature scores, the same trends were ob-served, but with coefficients approximately 15% largcr with voiced data, and approximately 20% smaller with unvoiced data.

sis. Ideas for futurę research, including a cross-language study of eon tractions used in conversation, will be proposed.

2:00

9SP5. Time course of bemispberic differences in spoken word recognition. Edward T. Auer and Paul A. Luce (Lang. Perccpt. Lab., Dept. of Psychol., State Univ. of New York, Buffalo, NY 14260)

Previous research has demonstrated differences between hemi-spheric processing of auditorily and visually presented lexically ambig-uous words. The present study further examined the time course of hemispheric processing of lexically ambiguous spoken words. Lexically ambiguous primes (e.g., BANK) were presented binaurally. Fłfty or 500 ms later, a target word was then presented monaurally to the left or right ear for a speeded lexical decision response (i.e., WORD-NONWORD). Target words were either (1) related to the dominant mcaning of the ambiguous prime (BANK-MONEY), (2) related to thc subordinate meaning of the prime (BANK-RIVER), or (3) unre-lated to the prime. No significant diffcrential effeets of facilitation were found in any condition. In addition, only the fastest subjects in the 50-ms ISI expcriment showed a significant right-car advantage. Impli-cations of these results will be discussed in terms of modality specificity and temporal versus spatial stimulus array.

2:15

9SP6. Speeded comparisons of spoken words. Paul A. Luce (Lang. Percept. Lab., Dept. of Psychol., State Univ. of New York, Buffalo, NY 14260)

This research evaluated the claim that spoken words are recognized as soon as they diverge from all other words in memory. Pairs of eon son an t-vowe 1 -consonan t (CVC) words were presented for a speeded SAME-DIFFERENT response. Subjects were instructed to respond SAME or DIFFERENT as soon as they could determine if the second word of the pair was the same as or different from the first word. Reaction times were measured from the onset of the second word. The overlap between the two members of the pairs of spoken words was systematically manipulated to determine if reaction times to say DIFFERENT were a simple function of the point at which the two words difTered or whether overlapping segments after the divergencc point would affect reaction times. The results are consistent with the notion that the temporal window for mapping information onto representations of spoken words in memory spans morę than a feature or segment and that spoken words are not necessarily recognized as soon as they diverge from other words in memory.

1:45

9SP4. Multimode database and its preliminary results. Karen Wallace, Josephine Homa, Yukiko Monzen, and Noriko Umeda (Dept. of Linguistics ISALS, New York Univ., 719 Broadway, New York, NY 10003)

Data from four speaking modes—word-list reading, text reading, conversation, and reading of sentences occurring in the convcrsation— were collected from American English speakera (five małe, fivc femalc). Each speaker spoke for approximately 5 h. Smaller amounts of similar data are being collected for Japanese and Spanish. The principal objec-tive is to better understand speech activities and their underlying rules by capturing them in speaking modes that exhibit very different ranges of variation. The reading of sentences spoken in conversation is part i c-ularly useful, because it allows us to directly compare acoustic attributes of linguistic forms in formal versus spontaneous speech. Scveral re-search projects that contrast conversation and morę formal modes (two of which will be presented at this meeting) are being carried out. These include formant frequencies of unstressed vowełs [Wallace]; acoustic properties of /a/s [Homa]; prosodic attributes of stressed and unstressed vowels; perception and acoustic correlates of degrees of empha-

2010 J. Acoust Soc. Am.. Vol. 89. No. 4, Pt. 2. April 1991

2:30

9SP7. Some lexical effeets in a generalized phoneme monitoring task. Scott E. Lively and David B. Pisoni (Speech Res. Lab., Dept. of Psychol., Indiana Univ., Bloomington, IN 47405)

A long-standing concem in thc spoken word recognition literaturę has been whether usc of thc lexicon is nccessary to complete a phoneme monitoring task. Eimas et ai [J. Mcm. Lang. 29(2), 160-180 (1990)], for example, found no )exical effeets in a phoneme monitoring task until monitoring responses were accompanied by lexical decisions or noun-verb categorizations. Frauenfelder and Segui [Mcm. Cog. 17(2), 134— 140 (1989)], in contrast, found facilatory priming effeets in a monitoring task when the position of the target phoneme varied randomly from trial to trial. Tlie current study adopts Frauenfelder and Segui*s generalized phoneme monitoring paradigm in a speeded phoneme classifica-tion task. Subjects participating in a blocked coiKlition demonstrated word frequency and density effeets only for blocks of trials in which targets occurred word finally. No !cxical effeets were observed for responses to word initial targets. Subjects in a mixed condition showed frequency and density effeets for word finał targets and frequency effeets for word initial targets. In terms of Cutler’s race model, thc data indi-

121st Meeting: Acoustical Society of America 2010