Proceedings Fonetik 2009 - Institutionen för lingvistik

Proceedings, FONETIK 2009, Dept. of Linguistics, Stockholm University6) for which they showed the highest level ofcorrect answers to the identification question.Summary and conclusionsIn this study the original event consisted of adialogue between two persons and, similarly,the recordings for the foils were a dialogue.This is an important feature of this study andsomething that contributes to increasing theecological validity in this research area sinceprevious research often has used monologuereadings of text both as the original events andas the recognition stimuli. To what extent thisfeature of the study influenced the results is notclear since we did not have a comparison conditionin this context.Characteristic voice features had an impactupon the listeners in this study. The results, sofar, are expected since the participants seem tobe confused and thought that the voice of foil 4was the target speaker, compared with the otherfoils in the lineup. Foil 4 was the most alikeconcerning pitch and speaking style. It mightbe that the speaking style and the forced voicewas a kind of hang-up for the listeners. Eventhough all male speakers had almost the samedialect and the same age as the target speaker,there were obvious differences in their voicesand speech behavior. The listeners were nottold what to focus on when listening the firsttime. As noted above, we don’t know if the useof a dialogue with a forensic content had an effectupon the result. The recordings in thelineup were completely different in their content.In brief, the results in this study suggest thatprominent characteristic features in voice andspeech are important in an earwitness identificationsituation. In a forensic situation it wouldbe important to be aware of characteristic featuresin the voice and speech.Turning next to the realism in the participants’confidence judgments it is of interestthat the participants in this study over all, incontrast to some other studies on earwitnesses(e.g., Olsson et al, 1998), showed only a modestlevel of overconfidence. However, a recentreview of this area shows that the level of realismfound depends on the specific measureused and various specific features of the voicesinvolved. For example, more familiar voicesare associated with better realism in the confidencejudgments (Yarmey, 2007). Had a differentmixture of voices been used in the presentstudy, the general level of realism in the O/Uconfidencemeasure might have been different.We next discuss the variation between thefoils with respect to their level of overconfidence.It can be discerned from Table 2 that thelevel of overconfidence follows the respectivefoil’s level of accuracy. When the level of identificationaccuracy is high, the level of O/Uconfidenceis lower, or even turns into underconfidence.Thus, a contributing reason to thevariation in overconfidence between the foils(in addition to the similarity of the foils’ voicesto that of the target), may be that the participantsexpected to be able to identify a foil asthe target and when they could not do so thisresulted in less confidence in their answer. Anotherspeculation is that the participants’ generalconfidence level may have been the mostimportant factor. In practice, if these speculationsare correct it is possible that the differentspeech features of the foils’ voices did not contributevery much to the participants’ level ofconfidence or degree of overconfidence. Insteadthe participants’ confidence may be regulatedby other factors as speculated above.The results for the slope measure showedthat the participants evidenced some ability toseparate correct from incorrect answers bymeans of their confidence judgments for thetwo foils 3 and 6, that is, the foils for which theparticipants showed the highest level of accuracyin their identifications. These two foilswere also the foils that may be argued to beperceptually (i.e., “experientially”) most separatefrom the target voice. For the other fourfoils the participants did not evidence any abilityat all to separate correct from incorrect identificationanswers by means of their confidencejudgments.Finally, given that they hold in future research,the results showed that earwitnesses’confidence judgments do not appear to be avery reliable cue as to the correctness of thetheir identifications, at least not in the situationinvestigated in this study, namely the context oftarget-absent lineups when the target voice occurin dialogues both in the original event andin the foils’ voice sample. The results showedthat although the average level of overconfidencewas fairly modest when computed overall foils, the level of over-underconfidence varieda lot between the different foils. Still itshould be noted that for those two foils wherethe participants had the best accuracy level theyalso tended to give higher confidence judg-184