Abstract
In this study, 10 pairs of unfamiliar voices were aurally and visually (with spectrograms) compared by 12 examiners to determine whether they belonged to the same speaker. Five decisions could be made: positive identification, positive elimination, probable identification, probable elimination, and unable to make a decision with the samples submitted. Each speaker recorded 9 cue words twice, spoken as a sentence, within a 15 month interval. The pair of speakers in each task was of the same sex. Match and no match constituted 50% each of the total, presented in a random fashion. For each group of examiners there were 40 possible positive correct decisions (20 identifications and 20 eliminations). Group I (professional examiners with >2 years experience) reported 90% correct positive identifications; the remaining 10% corresponded to probable identifications and no decision. Correct positive eliminations were 80%; the remaining 20% corresponded to probable eliminations and no decision. Group II (trainees with <2 years experience) reported 80% correct positive identifications; the remaining 20% corresponded to probable identifications. Correct positive eliminations were 75%; the remaining 25% corresponded to probable eliminations and no decision. No errors were produced by either Group I or II. Group III (novices) reported 40% correct positive identifications; probable identifications and no decision constituted 35%. Correct positive eliminations were 75%; probable eliminations were 20%. The percentages of error for false identifications and false eliminations were 5 and 25, respectively. The collaborative results suggest that professional examiners perform better than trainess and the performance of both groups is far superior to that of novices. Examiners with academic training in speech and hearing sciences and with over 2 years of experience perform better than those lacking academic and/or practical experience. The 15-month interval between the non-contemporary speech samples used in this study did not yield any errors for Groups I and II and no significant difference was found in examiners’ ability to identify or eliminate male or female voices. Group III had more difficulty with female speakers than with male speakers. The results also suggest that trained professional examiners can make very reliable decisions using no more than 9 cue words, regardless of speaker sex, and non-contemporary speech samples up to 15-month intervals. On the basis of this study, the official first action voiceprint method, 45.A01–45.A03, has been adopted as official final action.
AN INTEGRATED INTELLIGENT SYSTEM FOR CONSTRUCTION INDUSTRY: A CASE STUDY OF RAISED FLOOR MATERIAL