Proceedings Fonetik 2009 - Institutionen för lingvistik

Proceedings, FONETIK 2009, Dept. of Linguistics, Stockholm UniversityFalse Acceptances / Foil in Target Absent (TA) and TargetPresent (TP)Nr302520151050JL MM TN KG CF JÅ MS JL MM TN KG CF JÅTA TA TA TA TA TA TA TP TP TP TP TP TPFigurant7-8 y11-12 yAdultsAllFigure 1. False acceptances for each figurant speaker in the 3 age groups and the sum (all) for bothtarget absent (TA) and target present (TP).we are only interested in the general tendenciesof the false acceptances (the picking of thewrong voice) and the true, i.e. correct identifications.In Figure 1 we present the false acceptancesgiven by the different age groups and alltogether.In Figure 1 it is very clear that false acceptanceis biased toward certain speakers such asspeaker CF followed by MM and JL. It is noticeablethat correct acceptances in TP was 27and that can explain the decrease in FA for MMand JL, however, the degree of FA for speakerCF is even higher in TP (28).N/second76543210Articulation RateCF JÅ JL KG MS MM NS TN PoCSpeakerFigure 2. Articulation rate (produced syllables persecond) for the speakers in the line-up.In Figure 2 we can see that the target (PoC)was produced with a fast articulation rate. Severalspeakers follow with rather average valuesaround 5 syllables per second. The speakerwith the highest AR compared to PoC is CF. InFigure 3 we take a closer look at pausing.ARPauses tend to increase in duration with higharticulation rate (Goldman-Eisler, 1961).N/min (%)454035302520151050Pausing for Line-up SpeakersCF JÅ JL KG MS MM NS TN PoCSpeakersPausdur/minPaus_N/minPaus_%Figure 3. Pausing (pause duration per minute, numberof pauses per minute and percentage pause fromtotal utterance duration) for the speakers in the lineup.The pausing measurement shows a bias towardsspeaker CF, which might explain some of thefalse acceptances.The perception test resultsBoth listening tests separately (controlled anduncontrolled) show significant inter-rateragreement (Cronbach’s alpha = 0.98 for thecontrolled and 0.959 for the uncontrolled test).When both datasets are pooled the inter-rateragreement remains at the same high level (alpha= 0.975) indicating that listeners in bothsubgroups have judged the voices the sameway. This justifies using the pooled data from188