Proceedings Fonetik 2009 - Institutionen för lingvistik

More documents

Recommendations

Info

Proceedings, FONETIK 2009, Dept. of Linguistics, Stockholm UniversityPerception of voice similarity and the results of avoice line-upJonas LindhDepartment of Philosophy, Linguistics and Theory of Science, University of Gothenburg, SwedenAbstractThe perception of voice similarity is not thesame as picking a speaker in a line-up. Thisstudy investigates the similarities and differencesbetween a perception experiment wherepeople judged voice similarity and the resultsfrom voice line-up experiment. Results give usan idea about what listeners do when they tryto identify a voice and what parameters play animportant role. The results show that there aresimilarities between the voice similarity judgmentsand the line-up results.They differ, however,in several respects when we look atspeaking parameters. This finding has implicationsfor how to consider the similarities betweenfoils and suspects when setting up a lineupas well as how we perceive voice similaritiesin general.IntroductionAural/acoustic methods in forensic speakercomparison cases are common. It is possible todivide speaker comparison into 2 differentbranches depending on the listener. The 1 st isthe expert witness’ aural examination of speechsamples. In this case the expert tries to quantifyand assess the similarities/dissimilarities betweenspeakers based on linguistic, phonologicaland phonetic features and finally evaluatethe distinctiveness of those features (French &Harrison, 2007). The 2 nd branch is the speakercomparison made by naive listeners, for examplevictims of a crime where they heard avoice/speaker, but could not see the perpetrator.In both cases, some kind of voice quality wouldbe used as a parameter. However, it is not thoroughlyinvestigated if this parameter can beseparated from so called articulation or speakingparameters such as articulation rate (AR) orpausing, which are parameters that have shownto be useful parameters when comparing speakers(Künzel, 1997). To be able to study thiscloser a web based perception experiment wasset up where listeners were asked to judgevoice similarity in a pairwise comparison test.The speech was played backwards to removespeaking characteristics and force listeners toconcentrate on voice quality similarity. Thespeech material used in the present study wasoriginally produced for an ear witness studywhere 7 speaker line-ups were used to testvoice recognition reliability in ear witnesses.The speakers in that study were male andmatched for general speaker characteristics likesex, age and dialect. The results from the earwitness study and the judgments of voice similaritywere then compared. It was found, forexample, that the occurrence of false acceptances(FA) was not randomly distributed butsystematically biased towards certain speakers.Such results raise obvious questions like: Whywere these particular speakers chosen? Aretheir speaker characteristics particularly similarto those of the intended target? Would an auralvoice comparison test single out the samespeakers? The results give implications on theexistence of speech characteristics still presentin backward speech. It can also be shown thatspeakers that are judged as wolves (term fromspeaker verification, where a voice is rathersimilar to many models) can be picked moreeasily in a line-up if they also possess speechcharacteristics that are similar to the target.MethodTo be able to collect sufficiently large amountsof data, two different web tests were designed.One of the web based forms was only releasedto people that could insure a controlled environmentin which the test was to take place.Such a controlled environment could for examplebe a student lab or equivalent. A secondform was created and published to as manypeople as possible throughout the web, a socalleduncontrolled test group. The two groups'results were treated separately and later correlatedto see whether the data turned out to besimilar enough for the results to be pooled.The ear witness studyTo gain a better understanding of earwitnessperformance a study was designed in whichchildren aged 7-8 and 11-12 and adults servedas informants. A total of 240 participants were186
Proceedings, FONETIK 2009, Dept. of Linguistics, Stockholm Universityequally distributed between the three agegroups and exposed to an unfamiliar voice.Each participant was asked to come along withan experimenter to a clothes shop where theystopped outside a fitting cubicle. Behind thecurtain they could here an unfamiliar voiceplanning of a crime (PoC). The recording theyheard was played with a pair of high qualityloudspeakers and was approximately 45 secondslong. After two weeks, the witnesses wereasked to identify the target-voice in a line-up (7voices). Half of the witnesses were exposed toa target-present line-up (TP), and the other halfto a target-absent line-up (TA). The line-up wasalso played to the witness on loudspeakers froma computer and the voices presented on a powerpoint slide. First an excerpt from a recording ofa city walk of a bout 25 seconds was played.After that a shorter part of the excerpt of about12-15 seconds was used. First they had to saywhether they thought the voice was present inthe line-up, and if so, they pointed the voiceout. Secondly they were asked about their confidenceand what they remembered from whatthe voice in the cubicle had said. This was doneto see whether it was possible to predict identificationaccuracy by analyzing memory forcontent (Öhman, Eriksson & Granhag, 2009).To be able to quantify speaking parameters,pausing and articulation rate was measured. Articulationrate is here defined as produced syllablesexcluded pausing. Pauses are defined asclearly measurable silences longer than 150 ms.The test materialThe recordings consisted of spontaneousspeech elicited by asking the speakers to describea walk through the centre of Gothenburg,based on a series of photos presented to them.The 9 (7 plus 1 in TA + target) speakers wereall selected as a very homogeneous group, withthe same dialectal background (Gothenburgarea) and age group (between 28–35). Thespeakers were selected from a larger set of 24speakers on the basis of a speaker similarityperception test using two groups of undergraduatestudents as subjects. The subjects hadto make similarity judgments in a pairwisecomparison test where the first item was alwaysthe target speaker intended for the line-up test.Subjects were also asked to estimate the age ofthe speakers. The recordings used for thesetests were 16 kHz /16 bit wave files.The web based listening testsThe listening tests had to be made interactiveand with the results for the geographically dispersedlisteners gathered in an automatic manner.Google docs provide a form to create webbased question sheets collecting answers in aspreadsheet as you submit them and that wasthe form of data collection we chose to use forthe perception part of the study. However, ifone cannot provide a controlled environment,the results cannot be trusted completely. As ananswer to this problem two equal web basedlistening tests were created, one intended for aguaranteed controlled environment and oneopenly published test, here referred to as uncontrolled.The two test groups are here treatedseparately and correlated before being mergedin a final analysis.In the perception test for the present study,9 voices were presented pair-wise on a webpage and listeners were asked to judge the similarityon a scale from 1 to 5, where 1 was saidto represent “Extremely similar or same” and 5“Not very similar”. Since we wanted to minimizethe influence of any particular language orspeaking style, the speech samples were playedbackwards. The listeners were also asked tosubmit information about their age, first languageand dialectal background (if Swedishwas their first language). There was also aspace where they could leave comments afterthe completion of test and some participantsused this opportunity. The speech samples usedin the perception test were the first half of the25 second samples used in the earwitness lineups,except for the pairs where both sampleswere from the same speaker. In these cases theother item was the second half of the 25 secondsamples. Each test consisted of 45 comparisonsand took approximately 25 minutes to complete.32 (7 male, 25 female) listeners performedthe controlled listening test and 20 (6male, 14 female) the uncontrolled test.Results and DiscussionThe results will be presented separately in thefirst 2 paragraphs and then the comparison isdone with a short discussion in the last section.The overall results of the ear w itnessstudyThe original purpose of the study was to compareperformance between the age groups. Here187
Page 1 and 2:
Department of LinguisticsProceeding
Page 3 and 4:
Proceedings, FONETIK 2009, Dept. of
Page 5 and 6:
Page 7 and 8:
Page 9 and 10:
Page 11 and 12:
Page 13 and 14:
Page 15 and 16:
Page 17 and 18:
Page 19 and 20:
Page 21 and 22:
Page 23 and 24:
Page 25 and 26:
Page 27 and 28:
Page 29 and 30:
Page 31 and 32:
Page 33 and 34:
Page 35 and 36:
Page 37 and 38:
Page 39 and 40:
Page 41 and 42:
Proceedings, FOETIK 2009, Dept. of
Page 43 and 44:
Page 45 and 46:
Page 47 and 48:
Page 49 and 50:
Page 51 and 52:
Page 53 and 54:
Page 55 and 56:
Page 57 and 58:
Page 59 and 60:
Page 61 and 62:
Page 63 and 64:
Page 65 and 66:
Page 67 and 68:
Page 69 and 70:
Page 71 and 72:
Page 73 and 74:
Page 75 and 76:
Page 77 and 78:
Page 79 and 80:
Page 81 and 82:
Page 83 and 84:
Page 85 and 86:
Page 87 and 88:
Page 89 and 90:
Page 91 and 92:
Page 93 and 94:
Page 95 and 96:
Page 97 and 98:
Page 99 and 100:
Page 101 and 102:
Page 103 and 104:
Page 105 and 106:
Page 107 and 108:
Page 109 and 110:
Page 111 and 112:
Page 113 and 114:
Page 115 and 116:
Page 117 and 118:
Page 119 and 120:
Page 121 and 122:
Page 123 and 124:
Page 125 and 126:
Page 127 and 128:
Page 129 and 130:
Page 131 and 132:
Page 133 and 134:
Page 135 and 136: Proceedings, FOETIK 2009, Dept. of
Page 137 and 138: Proceedings, FONETIK 2009, Dept. of
Page 185: Proceedings, FONETIK 2009, Dept. of
Page 227: Department of LinguisticsPhonetics
show all

Proceedings Fonetik 2009 - Institutionen för lingvistik

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?