Osgoode Hall Law Journal Volume 52 Issue 3 Volume 52, Issue 3 (Summer 2015) Article 3 1-3-2016 Earwitness Evidence: The Reliability of Voice Identifications Christopher Sherrin The University of Western Ontario Follow this and additional works at: https://digitalcommons.osgoode.yorku.ca/ohlj Part of the Criminal Procedure Commons Article This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License. Citation Information Sherrin, Christopher. "Earwitness Evidence: The Reliability of Voice Identifications." Osgoode Hall Law Journal 52.3 (2015) : 819-862. https://digitalcommons.osgoode.yorku.ca/ohlj/vol52/iss3/3 This Article is brought to you for free and open access by the Journals at Osgoode Digital Commons. It has been accepted for inclusion in Osgoode Hall Law Journal by an authorized editor of Osgoode Digital Commons. Earwitness Evidence: The Reliability of Voice Identifications Abstract While much attention has been paid to the frailties of eyewitness evidence, little attention has been given to the reliability of voice identification evidence, even though such “earwitness” evidence has been tendered in several wrongful conviction cases. The author reviews the empirical literature on the reliability of earwitness evidence and compares it to the principal factors used by Canadian criminal courts to assess earwitness testimony. The author concludes that earwitness evidence often can be quite unreliable and that the courts have not always properly assessed its reliability, offering some suggestions for reform. Keywords Voice--Identification; Evidence (Law); Canada Creative Commons License This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License. This article is available in Osgoode Hall Law Journal: https://digitalcommons.osgoode.yorku.ca/ohlj/vol52/iss3/3 819* Faculty of Law, Western University. My thanks to Ian Scott for his comments on an earlier version of this article. Any remaining errors are my own.Earwitness Evidence: The Reliability of Voice IdentificationsCHRISTOPHER SHERRIN*While much attention has been paid to the frailties of eyewitness evidence, little attention has been given to the reliability of voice identification evidence, even though such “earwitness” evidence has been tendered in several wrongful conviction cases. The author reviews the empirical literature on the reliability of earwitness evidence and compares it to the principal factors used by Canadian criminal courts to assess earwitness testimony. The author concludes that earwitness evidence often can be quite unreliable and that the courts have not always properly assessed its reliability, offering some suggestions for reform.Alors que l’on se préoccupe beaucoup de la fragilité des preuves apportées par des témoins oculaires, on néglige trop souvent la fiabilité des preuves découlant de l’identification de voix, même si de telles preuves ont parfois mené à des condamnations injustifiées. L’auteur passe en revue la documentation empirique touchant la fiabilité des preuves apportées par les témoins auditifs et compare ces dernières aux principaux facteurs utilisés au Canada par les cours criminelles pour évaluer les témoins auditifs. L’auteur conclut que les témoignages auditifs sont souvent très peu fiables et que les cours n’ont pas toujours évalué correctement leur fiabilité, en apportant quelques suggestions de changement.(2015) 52 OSGOODE HALL LAW JOURNAL820ON 3 NOVEMBER 2006, two armed men, one white and one black, invaded the southwestern Ontario home of Duane and Deirdre Hicks.1 They tied up the occupants, demanded money, and ransacked the house for electronics. The black intruder even took Mr. Hicks to the bank to withdraw money. The police quickly charged Joseph Webber with the robbery, alleging that he was the white intruder. That intruder wore a balaclava during the robbery, but the area around his eyes remained exposed. At Webber’s trial, Mr. Hicks testified that he was able to recognize Webber by his eyes, the way he walked, and his voice. Hicks had the opportunity during the robbery to see and hear the white intruder when he was only a few feet away. Hicks had known Webber for many years, had occasionally socialized with him, and two years earlier had hired him to do work on the Hicks’ house. Hicks also testified that the black intruder called his partner Joe and that the white intruder mentioned that he had a drug debt to pay, something that was of interest given that two weeks before the robbery Webber had asked Hicks for a loan to repay a drug debt. Based on this evidence, the trial judge convicted Webber and sentenced him to seven and a half years’ imprisonment. The only problem was that Webber was innocent. The police themselves discovered this by happenstance when another man named Justin Parry confessed to the crime in June 2008. The police corroborated the confession, located the accomplice (who 1. Most of the information in this paragraph was obtained from the facta filed in the Ontario Court of Appeal by the Crown and defence in R v Webber. See R v Webber, 2010 ONCA 4, 194 ACWS (3d) 873 (Factum of the Appellant); R v Webber, 2010 ONCA 4, 194 ACWS (3d) 873 (Factum of the Respondent).I. THE (IN)ACCURACY OF VOICE IDENTIFICATION .............................................................................. 823II. VOICE IDENTIFICATION EVIDENCE IN CANADIAN LAW .................................................................. 826III. FACTORS BEARING UPON THE RELIABILITY OF VOICE IDENTIFICATIONS .................................... 829A. Familiarity ........................................................................................................................... 829B. Exposure Length ................................................................................................................ 834C. Retention Interval ............................................................................................................... 836D. Confidence .......................................................................................................................... 838E. Attention ............................................................................................................................. 840F. Abnormal Speaking ............................................................................................................ 842G. Telephone Transmission .................................................................................................... 845H. Unfamiliar Accents and Languages ................................................................................... 847I. Distinctiveness ................................................................................................................... 851IV. CONCLUSION AND RECOMMENDATIONS ........................................................................................ 854V. APPENDIX A: AMERICAN WRONGFUL CONVICTION CASES BASED PARTLY ON VOICE SHERRIN, EARWITNESS EVIDENCE 821also identified Parry as the white intruder), and found no connection between Webber and Parry. The Crown eventually asked the Ontario Court of Appeal to acquit Webber. The court did so on 7 January 2010, commenting that “the Crown acknowledges that the fresh evidence overwhelmingly shows the appellant did not commit these crimes.”2Joseph Webber’s case exemplifies the familiar frailties of eyewitness identification, but it also highlights another much less discussed phenomenon: the frailties of voice identification, sometimes called “earwitness” evidence. Earwitness evidence is less common than eyewitness evidence, and it is often heard in combination with eyewitness evidence, but it can be important and even critical in criminal prosecutions. Earwitness evidence is routinely needed when the Crown relies on evidence of intercepted communications, as in many conspiracy and narcotics trials. It can also be crucial in cases where the perpetrator wore a disguise, the victim was blindfolded or otherwise unable to make reliable observations, or the offence was committed over the telephone. The evidence is admissible in Canadian courts under the general rules of admissibility. When tendered through a lay witness,3 it is not considered opinion evidence, and there are no special preconditions to admissibility.4 The contribution of mistaken eyewitnesses to wrongful convictions has been well documented. The contribution of mistaken earwitnesses has not. But Joseph Webber is not the only victim of incorrect voice identification. At Guy Paul Morin’s second trial, the victim’s mother identified Morin’s voice as the one that uttered “Help me, help me, Oh God, help me” on the night of the victim’s funeral—testimony that the prosecution used as evidence of consciousness of guilt. The Commissioner examining Morin’s wrongful conviction concluded that the voice 2. R v Webber, 2010 ONCA 4 at para 1, 194 ACWS (3d) 873.3. Expert evidence of voice identification is sometimes tendered, but it is beyond the scope of this article. I will be dealing solely with non-expert identification evidence.4. See R v Williams (1995), 23 OR (3d) 122, 98 CCC (3d) 160 (CA) [Williams]; R v Braumberger, 62 WWR 285, [1967] BCJ No 10 (QL) (CA); R v Rahimi, 2015 SKCA 85 at paras 67-68, [2015] SJ No 385. The only precondition, if one can call it that, is that the Crown must lay a foundation for the witness’s ability to identify the voice (by adducing evidence that the witness had more than fleeting exposure to the voice outside of the criminal encounter in circumstances where the witness could connect the voice to a particular person). See R v Portillo (2003), 174 OAC 226 at paras 40-42, 176 CCC (3d) 467 (CA) [Portillo].(2015) 52 OSGOODE HALL LAW JOURNAL822identification evidence was patently unreliable.5 At least seventeen individuals have been wrongly convicted in the United States based in part on faulty voice identification evidence.6 In at least five cases, the voice identification evidence was critical to the prosecution’s case. One of England’s oldest documented wrongful convictions was the product of mistaken voice identification. William Hulet was convicted of high treason based substantially on the identification of his voice as that of the masked executioner who hanged King Charles I. Another individual (the regular hangman) subsequently confessed, and Hulet was exonerated.7As all of this illustrates, voice identification is not infallible. In fact, those who have studied its reliability have variously characterized the identification task, at least as it relates to unfamiliar voices, as “remarkably difficult”8 and “error prone”9 where performance is “very poor,”10 “extremely inaccurate,”11 and “likely to produce high false identifications.”12 This may seem surprising given the common experience we all have of easily recognizing voices. But care must be taken not to generalize from experience with familiar voices in everyday 5. Ontario, The Commission on Proceedings Involving Guy Paul Morin: Report, ch 5 (Toronto: Ministry of the Attorney General, 1998) at 978. Ivan Henry may be another Canadian example. The Crown’s case against him certainly included voice identification evidence. See R v Henry, 2010 BCCA 462 at paras 114-32, 294 BCAC 96 [Henry]. The case was also very weak, an alternative perpetrator has been identified, and Henry is sometimes included in the list of Canada’s wrongly convicted. See e.g. Joan McEwen, Innocence on Trial: The Framing of Ivan Henry (Victoria: Heritage House, 2014). But it is arguable that he has not been proven factually innocent.6. See the names and sources listed in Appendix A, below. As is so often true, there may be some debate about the factual innocence of some of these individuals, but I limited my list to cases where there was fairly strong, even if not always indisputable, evidence of innocence.7. Lawrence M Solan & Peter M Tiersma, “Hearing Voices: Speaker Identification in Court” (2003) 54:2 Hastings LJ 373 at 393-94, citing TB Howell, A Complete Collection of State Trials and Proceedings for High Treason and other Crimes and Misdemeanors from the Earliest Period to the Year 1783, vol 5 (London: TC Hansard, 1816) at 1145, 1185.8. Sarah V Stevenage, Amy Howland & Anna Tippelt, “Interference in Eyewitness and Earwitness Recognition” (2011) 25:1 Appl Cognit Psychol 112 at 112.9. A Daniel Yarmey, “The Psychology of Speaker Identification and Earwitness Memory” in RCL Lindsay et al, eds, The Handbook of Eyewitness Psychology: Volume II: Memory for People (Mahwah, NJ: Lawrence Erlbaum Associates, 2007) 101 at 102 [Yarmey, “Psychology of Speaker Identification”].10. Lisa Öhman, Anders Eriksson & Pär Anders Granhag, “Angry Voices from the Past and Present: Effects on Adults’ and Children’s Earwitness Testimony” (2013) 10:1 J Investigative Psychol & Off Profiling 57 at 59 [Öhman, Eriksson & Granhag, “Angry Voices”].11. Howard Saslove & A Daniel Yarmey, “Long-Term Auditory Memory: Speaker Identification” (1980) 65:1 J Appl Psychol 111 at 114.12. A Daniel Yarmey, A Linda Yarmey & Meagan J Yarmey, “Face and Voice Identifications in Showups and Lineups” (1994) 8:4 Appl Cognit Psychol 453 at 463.SHERRIN, EARWITNESS EVIDENCE 823situations to assumptions about identification accuracy with less familiar voices in other contexts.13 Indeed, one should not even assume that identification of familiar voices is inevitably accurate.14In recent years, Canadian criminal courts have demonstrated an appreciation of both the fallibility of earwitness testimony and some of the reasons for it. However, their analysis has been informed by experience, intuition, and assumption rather than empirical study. That is not to say that the courts have always got it wrong; on the contrary, they have often got it right. But their assessment of earwitness testimony only can be improved by an understanding of the results of the scientific study of voice identification. The point of this article is to offer some assistance in that regard. I discuss the results of the empirical studies. I compare the ways in which the approach of the courts has coincided with and deviated from those results. I offer some thoughts on how courts could alter and ultimately improve their assessment of earwitness testimony.I. THE (IN)ACCURACY OF VOICE IDENTIFICATIONAs stated above, many researchers studying voice identification have expressed concerns over its reliability. This stems in large part from the fact that studies have so often produced very low accuracy rates. One study, for example, obtained a “hit rate” of 9%, meaning that only 9% of the test subjects were able to correctly recognize a voice they had previously heard (the target voice) from amongst a group of other voices in a voice lineup.15 Some other studies have obtained hit rates, collapsed across experimental conditions, of 12%,16 13%,17 15–20%,18 13. Brian R Clifford, “Voice Identification by Human Listeners: On Earwitness Reliability” (1980) 4:4 L & Hum Behav 373 at 374.14. See Part III(A), below.15. Yarmey, Yarmey & Yarmey, supra note 12 at 458. The hit rate was 28% (lower than chance) in the voice showup condition where test subjects heard only a single voice and were asked if they recognized it as one they had heard five minutes before.16. Öhman, Eriksson & Granhag, “Angry Voices,” supra note 10 at 63. This was chance level performance.17. Lisa Öhman, Anders Eriksson & Pär Anders Granhag, “Mobile Phone Quality vs. Direct Quality: How the Presentation Format Affects Earwitness Identification Accuracy” (2010) 2:2 Eur J Psychol Appl Legal Context 161 at 172 [Öhman, Eriksson & Granhag, “Mobile Phone Quality”]. This was chance level performance.18. Nils Olsson, Peter Juslin & Anders Winman, “Realism of Confidence in Earwitness Versus Eyewitness Identification” (1998) 4:2 J Experimental Psychol Appl 101 at 107.(2015) 52 OSGOODE HALL LAW JOURNAL82419–20%,19 and 24%.20 It is not uncommon for hit rates to be lower than the percentage of test subjects who incorrectly indicate that the target voice is not present in the lineup21 or who choose the voice of a lineup foil.22Of even greater concern is the “false alarm rate” obtained in numerous studies: the percentage of test subjects who, when presented with a set of voices, incorrectly identify someone other than the speaker they had previously heard. An inaccurate positive identification, after all, can lead to the prosecution of the wrong person. Test subjects routinely select foils in a “target-present” lineup (one that includes the previously-heard voice).23 More importantly, they very often believe they recognize a voice in a “target-absent” lineup (one that does not contain the previously-heard voice). In such lineups, studies have obtained false 19. Lisa Öhman, Anders Eriksson & Pär Anders Granhag, “Enhancing Adults’ and Children’s Earwitness Memory: Examining Three Types of Interviews” (2013) 20:2 Psychiatry Psychol & L 216 at 223 [Öhman, Eriksson & Granhag, “Enhancing”].20. José H Kerstholt et al, “Earwitnesses: Effects of Accent, Retention and Telephone” (2006) 20:2 Appl Cognit Psychol 187 at 192 [Kerstholt et al, “Earwitnesses”]. This was the hit rate when test subjects were allowed to indicate they did not know if the target was present amongst the voices (which 24% of them indicated). When subjects were forced to make a selection, the hit rate increased to 34%.21. See e.g. Öhman, Eriksson & Granhag, “Enhancing,” supra note 19 at 224; Olsson, Juslin & Winman, supra note 18 at 106-07; Öhman, Eriksson & Granhag, “Angry Voices,” supra note 10 at 63; Lisa Öhman, Anders Eriksson & Pär Anders Granhag, “Overhearing the Planning of A Crime: Do Adults Outperform Children As Earwitnesses?” (2011) 26:2 J Police & Crim Psychol 118 at 123 [Öhman, Eriksson & Granhag, “Overhearing the Planning”].22. See e.g. Öhman, Eriksson & Granhag, “Enhancing,” supra note 19 at 224; Olsson, Juslin & Winman, supra note 18 at 106-07; Yarmey, Yarmey & Yarmey, supra note 12 at 458; Öhman, Eriksson & Granhag, “Angry Voices,” supra note 10 at 63; Öhman, Eriksson & Granhag, “Overhearing the Planning,” supra note 21 at 123.23. See Öhman, Eriksson & Granhag, “Enhancing,” supra note 19 at 224; Olsson, Juslin & Winman, supra note 18 at 106-07; Yarmey, Yarmey & Yarmey, supra note 12 at 458; Öhman, Eriksson & Granhag, “Angry Voices,” supra note 10 at 63; Öhman, Eriksson & Granhag, “Overhearing the Planning,” supra note 21 at 123.SHERRIN, EARWITNESS EVIDENCE 825alarm rates, across experimental conditions, of 99%,24 98%,25 85%,26 43–93%,27 70%,28 67%,29 56%,30 53%,31 and around 50%.32 Those figures do not translate to the likelihood that an innocent suspect will be identified in a real-life situation, since many of the false alarms would be foil identifications ignored by the police, but they do illustrate the propensity of earwitnesses to misidentify voices. Furthermore, studies that pre-selected a particular voice to serve as the innocent suspect in a voice lineup have still found substantial “misidentification rates,” 24. Lori R Van Wallendael et al, “‘Earwitness’ Voice Recognition: Factors Affecting Accuracy and Impact on Jurors” (1994) 8:7 Appl Cognit Psychol 661 at 666. Some of the test subjects had heard the target voice only a short time before the lineup was administered. All test subjects were cautioned (albeit subtly) that the target might not be present in the lineup.25. Yarmey, Yarmey & Yarmey, supra note 12 at 458. Test subjects were exposed to the voice during an innocuous encounter that lasted about fifteen seconds. Testing occurred about five minutes later. Subjects were warned that the target might not be present in the lineup.26. Harriet Mary Jessica Smith & Thom Baguley, “Unfamiliar Voice Identification: Effect of Post-Event Information on Accuracy and Coice Ratings” (2014) 5:1 J Eur Psychol Students 59 at 62, 64. Some of the test subjects were given misleading information about the pitch of the target voice prior to selecting a voice from the lineup, but the misinformation had no effect on overall accuracy.27. Sarah V Stevenage, Gabriella Clarke & Allan McNeill, “The ‘Other-Accent’ Effect in Voice Recognition” (2012) 24:6 J Cognit Psychol 647 at 650.28. A Daniel Yarmey, “Earwitness Descriptions and Speaker Identification” (2001) 8:1 Forensic Linguistics 113 at 118 [Yarmey, “Earwitness Descriptions”].29. Axelle C Philippon et al, “Earwitness Identification Performance: The Effect of Language, Target, Deliberate Strategies and Indirect Measures” (2007) 21:4 Appl Cognit Psychol 539 at 545. That is the false alarm rate for voices speaking a familiar language. The rate for voices speaking an unfamiliar language was even higher, at 93%. All test subjects had heard the target voice about thirty minutes earlier, were instructed to pay attention to the voice, and were told that the target voice may not be present in the lineup.30. Charles P Thompson, “A Language Effect in Voice Identification” (1987) 1:2 Appl Cognit Psychol 121 at 126.31. Öhman, Eriksson & Granhag, “Overhearing the Planning,” supra note 21 at 123.32. José H Kerstholt et al, “Earwitnesses: Effects of Speech Duration, Retention Interval and Acoustic Environment” (2004) 18:3 Appl Cognit Psychol 327 at 331 [Kerstholt et al, “Effects of Speech Duration”]; Kerstholt et al, “Earwitnesses,” supra note 20 at 192. The false alarm rates were 51% and 50%, respectively. In both studies, the test subjects knew the experiment concerned voice identification and were warned that the target voice might not be in the lineup.(2015) 52 OSGOODE HALL LAW JOURNAL826such as 13–14%,33 19%34 and 22%.35 Those rates were over and above all of the foil identifications.Not all studies have obtained poor accuracy rates. Indeed, some have obtained fairly good rates. For specific lineups, one occasionally sees hit rates of or in excess of 90% and false alarm rates of less than 10%.36 More commonly, one can find hit rates in the 35–65% range and false alarm rates in the 20–40% range, at least in certain experimental conditions (albeit sometimes with poorer rates in other conditions).37 Some researchers in the 1980s posited an overall accuracy rate of 60–70%,38 although more recent estimates are much less optimistic.39 In the end, the variety of results and methodologies in the different earwitness studies 33. A Daniel Yarmey, “Earwitness Identification Over the Telephone and in Field Settings” (2003) 10:1 Int’l J Speech Language & L 62 at 69 [Yarmey, “Earwitness Identification”].34. Yarmey, Yarmey & Yarmey, supra note 12 at 458. Subjects were tested within minutes of hearing the target voice and were told that the voice might not be in the lineup.35. Tara L Orchard & A Daniel Yarmey, “The Effects of Whispers, Voice-Sample Duration, and Voice Distinctiveness on Criminal Speaker Identification” (1995) 9:3 Appl Cognit Psychol 249 at 254. This is the false identification rate collapsed across all experimental conditions but only for non-distinctive voices. The equivalent rate for distinctive voices was 6%. Test subjects were told to listen carefully to the target voice for purposes of later identification. They were also told that the target voice might not be present in the lineup.36. See e.g. Orchard & Yarmey, ibid at 253-54 (perfect scores in one of the eight-minute exposure conditions); Van Wallendael et al, supra note 24 at 666 (in two out of four target-present lineups, although the authors cautioned that the targets in those lineups had distinctive voices, and extremely high false alarm rates were achieved in the target-absent lineups, usually 100%).37. See e.g. Judith P Goggin et al, “The Role of Language Familiarity in Voice Identification” (1991) 19:5 Memory & Cognition 448 at 451 (most hit rates between 35% and 65%, and false alarm rates lower than 40%); Susan Cook & John Wilding, “Earwitness Testimony: Never Mind the Variety, Hear the Length” (1997) 11:2 Appl Cognit Psychol 95 at 101 (overall hit rate of 38% for unfamiliar voices); Yarmey, “Earwitness Descriptions,” supra note 28 at 118 (hit rates of 47% and 55% and false alarm rates of 33% and 38% in target-present lineups); Kerstholt et al, “Effects of Speech Duration,” supra note 32 at 331 (hit rate of 42% and false alarm rate of 24% in target-present lineup, albeit with a 51% false alarm rate in the target-absent lineup).38. See e.g. Ray Bull & Brian R Clifford, “Earwitness Voice Recognition Accuracy” in Gary L Wells & Elizabeth A Loftus, eds, Eyewitness Testimony: Psychological Perspectives (Cambridge: Cambridge University Press, 1984) 92 at 120; Thompson, supra note 30 at 125.39. See e.g. Öhman, Eriksson & Granhag, “Angry Voices,” supra note 10 at 67. Öhman, Eriksson & Granhag noted that poor performance in the study “concurs with the suggestion that real-life conditions will result in earwitness performance at chance level.” See also Daniel Read & Fergus IM Craik, “Earwitness Identification: Some Influences on Voice Recognition” (1995) 1:1 J Experimental Psychol Appl 6. Read & Craik noted that “[i]n more realistic circumstances, … the probability that the correct speaker will be chosen is no better than chance” (ibid at 16).SHERRIN, EARWITNESS EVIDENCE 827probably makes it impossible to come up with an overall, generally applicable, accuracy rate. Accuracy is very much a product of specific factors and specific circumstances. Earwitnesses can be reliable.40 In a variety of situations, however, they can be quite inaccurate and prone to misidentify. As a result, one must endorse the conclusion of many researchers that courts should exercise great caution and care before placing reliance on earwitness testimony.41II. VOICE IDENTIFICATION EVIDENCE IN CANADIAN LAWIn recent years, Canadian criminal courts have recognized the need to proceed cautiously when it comes to voice identification evidence. In the 2015 case of R v Dodd, for example, the Ontario Court of Appeal commented that such evidence “ought to be treated with extreme caution … .”42 Courts in other provinces, both trial and appellate, have adopted a similar position.43This speaks well of the Canadian position with respect to earwitness evidence. But four qualifications must be added. First, one can certainly find cases where courts have treated voice identification evidence with something less than “extreme caution.” The 2006 case of R v Campbell is illustrative.44 Campbell was charged with robbing a video store. The issue in the case was identity. The store clerk was the only person to give identification evidence. The robber was previously unknown to her, and she interacted with him on the date in question for five to ten minutes. A month later, she claimed to see him at a local mall. She recognized him by his appearance and his voice. The trial judge cautioned himself regarding the frailties of eyewitness evidence but said nothing about the weaknesses of earwitness evidence. On the contrary, he only used the victim’s voice identification to help overcome any weaknesses with her visual identification.45 On appeal, Campbell claimed that his conviction was 40. Yarmey, “Psychology of Speaker Identification,” supra note 9 at 102.41. See e.g. Kerstholt et al, “Effects of Speech Duration,” supra note 32 at 334; Öhman, Eriksson & Granhag, “Angry Voices,” supra note 10 at 68.42. 2015 ONCA 286 at para 79 (available on QL) [Dodd]. From the same court, see also R v Clouthier, 2012 ONCA 636 at para 19 (available on QL) [Clouthier]; R v Quidley, 2008 ONCA 501 at para 36, 232 CCC (3d) 255; R v Badgerow, 2008 ONCA 605 at para 9, 237 CCC (3d) 107; R v Brown, [2003] OJ No 4592 (QL) at para 4, 2003 CarswellOnt 4776 (WL Can).43. See e.g. R v Saddleback, 2013 ABCA 250 at paras 18, 35, 556 AR 17 [Saddleback]; R v Pabla, 2013 BCSC 913 at para 23 (available on QL) [Pabla]; R v JE (2011), 100 WCB (2d) 69, [2011] NJ No 38 (QL) at para 27 (Prov Ct) [JE].44. 2006 BCCA 109, 223 BCAC 164.45. Ibid at para 35.(2015) 52 OSGOODE HALL LAW JOURNAL828unreasonable, in part because the trial judge “gave undue weight to [the victim’s] recognition of the appellant’s voice as confirming her identification of him.”46 The British Columbia Court of Appeal said nothing about that submission and only used the earwitness testimony to help justify the reasonableness of the trial judge’s decision.47 I do not claim that Campbell is typical of how courts treat voice identification, but it is also not so anomalous as to be extraordinary. A number of other court decisions have demonstrated some insensitivity to the potential weaknesses of earwitness evidence.48The Ontario Court of Appeal has also taken the position that a warning to the jury about the frailties of voice identification evidence is not always mandatory. In R v Masters, the court stated that it was not convinced a caution was mandatory in the case given the “significant body of other evidence at trial capable of implicating the appellant as one of the speakers on the wiretaps.”49 This curious holding suggests that when the trier of law determines that the corroborative evidence of identification is strong, the trier of fact need not be warned that the direct evidence of identification may be weak. Such an approach not only blurs the roles between the two triers in a case, it also leaves a jury unassisted in assessing the weight to be given to a piece of evidence that the jury may feel is a necessary element in finding proof beyond a reasonable doubt.Canadian courts also sometimes treat earwitness evidence no differently than eyewitness evidence. One sees this in some of the general cautionary statements. In the 2012 case of R v Clouthier, for example, the Ontario Court of Appeal wrote that “voice identification evidence, like any identification evidence, ought to be treated with extreme caution … .”50 The same court more recently commented that “[v]oice identification evidence is even more fraught with dangers than eyewitness identification evidence,”51 but its overall record is mixed. For example, 46. Ibid at para 68.47. Ibid at para 69. Only Finch CJ, in dissent, addressed this issue, but it is safe to assume that the majority agreed since they only said they differed from him in their view of how the trial judge used another piece of evidence.48. See e.g. R v Mackinaw, 2010 ABCA 359 (available on QL) [Mackinaw] (forgiving a trial judge’s complete lack of self-caution regarding important earwitness evidence); R v Wu, 2010 ABCA 337 at paras 29, 48, 52, 266 CCC (3d) 482 [Wu] (responding to a concern over the frailties of voice identification evidence by simply noting that the trial judge acknowledged the witness had no training in voice recognition and that the witness was “subjected to the usual scrubbing of cross-examination”) (ibid at para 52); R v Masters, 2014 ONCA 556, 114 WCB (2d) 628 [Masters] (discussed immediately below).49. Ibid at para 46.50. Supra note 42 at para 19 [emphasis added].51. Dodd, supra note 42 at para 79.SHERRIN, EARWITNESS EVIDENCE 829it has not repudiated its pronouncement in 1995 that “[t]here is no pattern of instruction that the trial judge must give to the jury or to himself, if he is the trier of fact, with respect to this type of identification evidence as distinct from the more usual identification evidence given by an eyewitness.”52 One can also see the same approach in the analysis of voice identification evidence in particular cases. R v Henry, for example, was a prosecution that depended heavily on voice identification as well as, and possibly even more than, visual identification.53 In assessing the reasonableness of the verdict, however, the British Columbia Court of Appeal focused solely on the weaknesses of eyewitness evidence, adding nothing specific about earwitness evidence.54 The court in Henry ultimately declared the verdict unreasonable, but from an analytical standpoint, it was inappropriate to effectively assimilate the two kinds of evidence. Studies have shown that earwitness evidence is even less reliable than eyewitness evidence.55 It must therefore be treated with greater caution than eyewitness evidence, not with the same amount.56The last reason why one cannot be overly sanguine about the Canadian position regarding earwitness evidence is that the courts do not always properly assess the variables that affect the reliability of earwitness testimony. General cautions are helpful, but what really matters is how courts treat the various factors that relate to the trustworthiness of the specific earwitness evidence in a given case. Courts certainly take into account a number of appropriate factors, but they do not always do so with the nuance required. They also sometimes rely on unhelpful factors and fail to consider helpful ones. In the next part of this article, I turn to an analysis of some of the most important factors. I emphasize at the start that in order to analyze the performance of Canadian courts, I have had to isolate their use of a particular factor from the rest of the evidence considered in a given decision. The reader must keep in mind that courts rarely rely solely on voice identification evidence to convict and that any frailties in the analysis of such evidence may not have had a significant impact on the overall validity of the ultimate verdict.52. Williams, supra note 4 at para 22. Cf R v Rowbotham (No 4) (1977), 2 CR (3d) 244 at 262, 33 CCC (2d) 411 (Ont Gen Sess Ct). The court noted that “the jury must be cautioned with respect to the frailties surrounding such evidence in like manner to the caution normally given in the case of visual identification.”53. See Henry, supra note 5 at paras 113-38.54. See especially Henry, ibid at paras 77, 139-41.55. See e.g. Yarmey, Yarmey & Yarmey, supra note 12; Olsson, Juslin & Winman, supra note 18.56. See R v Pinch, 2011 ONSC 5484 at para 76, 97 WCB (2d) 618 [Pinch].(2015) 52 OSGOODE HALL LAW JOURNAL830III. FACTORS BEARING UPON THE RELIABILITY OF VOICE IDENTIFICATIONSA. FAMILIARITYThe factor most commonly taken into account by Canadian courts is almost certainly familiarity. Courts routinely note that an earwitness had significant prior exposure to the accused’s voice and thus was in a good position to recognize it during the commission of the crime.57 Some courts have even divided earwitness evidence into recognition versus identification evidence in an effort to highlight the presumed reliability of the former and distinguish instances where people are asked to identify the voice of a stranger.58Familiarity is an entirely appropriate factor to take into account. Empirical studies have supported the common sense assumption that significant prior exposure to a voice enhances the ability to identify it on a given occasion.59 Indeed, some studies have obtained results of near-perfect accuracy.60It is important, however, not to assume that familiarity necessarily imports accuracy. Familiar voices can be misidentified, sometimes at significant rates. One study obtained a 10% misidentification rate from members of a close social network who had known each other for almost two years and who currently or recently lived in the same residence.61 In other words, the test subjects, hearing the voices of a series of close friends, incorrectly attributed a voice to the wrong person 10% of the time.62 The hit rate was 68%, and the test subjects offered no opinion regarding the identity of the speaker the other 22% of the time. Another study of immediate family members and best friends obtained a 5% 57. See e.g. R v Dobbin (2013), 334 Nfld & PEIR 284 at para 32, [2013] NJ No 73 (QL) (Prov Ct) [Dobbin]; R v Fitur, 2011 MBQB 59 at para 36, 264 Man R (2d) 20; R v Shea, 2010 NSPC 69 at paras 11, 15, 91 WCB (2d) 361. Courts have also taken into account lack of familiarity. See e.g. Clouthier, supra note 42 at para 20; Dodd, supra note 42 at paras 80-81.58. See e.g. R v Herman, 2010 BCSC 1068 at para 98 (available on QL) [Herman]; JE, supra note 43 at para 30.59. See Diana Van Lancker, Jody Kreiman & Karen Emmorey, “Familiar Voice Recognition: Patterns and Parameters” (1983) 13 J Phonetics 19 at 20-21. 60. See e.g. Harry Hollien, Wojciech Majewski & E Thomas Doherty, “Perceptual Identification of Voices Under Normal, Stress and Disguise Speaking Conditions” (1982) 10 J Phonetics 139.61. Paul Foulkes & Anthony Barron, “Telephone Speaker Recognition Amongst Members of a Close Social Network” (2000) 7:2 Forensic Linguistics 180.62. This was based on hearing about nine seconds of speech over the telephone. Test subjects were simply asked to listen to a voice and, if they recognized it, to write down the name of the speaker.SHERRIN, EARWITNESS EVIDENCE 831misidentification rate.63 A judge who resolves the issue of voice identification solely on the basis of familiarity, therefore, proceeds too quickly. The judge also needs to consider the other factors that bear upon reliability. Hearing someone speak only a small number of words, for example, has been shown to affect identification accuracy of even very familiar listeners.64There are also different degrees of familiarity. All the studies mentioned in the preceding two paragraphs used test subjects who had among the greatest degree of familiarity with the speakers’ voices. The accuracy of test subjects who do not have quite the same degree of familiarity is generally lower. The aforementioned study of immediate family members and best friends, who the study categorized as “high-familiar” subjects, compared them to “moderate-familiar” (coworkers, teammates, club-mates, and general friends) and “low-familiar” subjects (casual acquaintances who only spoke with the target for a few minutes on occasion in any week over the preceding year).65 The respective rates were 85%, 79%, and 49% for hits and 5%, 13%, and 23% for misidentifications.66 Another study showed that people who considered themselves to be extremely familiar with a target’s voice, having heard it for more than one hour, identified it only 31% of the time.67 It is not clear that Canadian judges have been adequately alive to the significance of different gradations of familiarity. They have sometimes found and relied on familiarity despite fairly vague evidence of the degree of 63. A Daniel Yarmey et al, “Commonsense Beliefs and the Identification of Familiar Voices” (2001) 15:3 Appl Cognit Psychol 283. As in the Foulkes & Barron study, test subjects were simply asked to listen to a voice and, if they recognized it, to state the name of the speaker. Identifications were made as soon as the listener thought he or she recognized the voice, after hearing anywhere from one word to two minutes of speech. The hit rate was 85%.64. This issue is discussed in Part III(B), below. See R v Meier, 2012 SKPC 41 at paras 23-26, 49-54, 93, 394 Sask R 204. This is an example of a case where earwitness testimony was accepted largely on the basis of familiarity despite the fact that during the crime, the witness, who was highly stressed, only heard the perpetrator utter about fifteen words.65. The study actually referred to the groups as high-, moderate-, and low-familiar “speakers,” but it is clear that it was the familiarity of the listeners with the speakers’ voices that was being measured.66. Yarmey et al, supra note 63 at 287-88.67. Read & Craik, supra note 39 at 11-13. The 31% figure is the hit rate averaged across the three experimental conditions.(2015) 52 OSGOODE HALL LAW JOURNAL832prior exposure.68 They also may have occasionally slotted a witness into a higher category of familiarity than was warranted.69 The exact magnitude of the problem is difficult to discern,70 but hopefully greater familiarity with the relevant empirical literature will encourage proper sensitivity in the future. Finally, there is the issue of expectation. An “elementary principle of psychology is that context and expectations influence an individual’s perceptions and interpretations of what he observes.”71 When circumstances lead people to expect some fact, they tend to perceive that fact in the face of ambiguous information. This can lead to error biased in the direction of the expectation.72 This phenomenon of expectancy bias happens naturally and quite unconsciously, but it can be powerful.73Researchers have suggested that expectancy bias can find application in the context of voice identification. For example, Professor Yarmey has written:68. See e.g. Mackinaw, supra note 48 (finding that the victim recognized the accused’s voice from an uncertain number of past encounters—possibly only one—at inexact dates in the past); R v Sanghera, 2012 BCSC 733, 102 WCB (2d) 361 [Sanghera] (accepting a police officer’s identification evidence even though his personal dealings with the accused “had been relatively limited, including, he estimated, two or three face to face meetings”) (ibid at para 145); R v Lamarche, 2010 YKTC 28 (available on QL) [Lamarche] (“[w]ith respect to Mr. Brown’s ability to identify Mr. Lamarche’s voice over the phone, even though Mr. Brown had difficulty giving specific details about the timing and frequency of his dealings with Mr. Lamarche, … based on his evidence, … he was sufficiently familiar with Mr. Lamarche to be able to recognize his voice”) (ibid at para 26). I remind the reader that comments like these in this article are only directed at the courts’ use of a specific factor (in this case, familiarity) and not at the overall legitimacy of the judgments. It so happens, for example, that the judge in Sanghera ultimately acquitted and the judge in Lamarche relied on significant corroborating evidence.69. See e.g. Dobbin, supra note 57. The court found that an identifying witness was “very familiar” with the accused’s voice after having heard it as little as ten times over the preceding two or more years during the course of a professional social worker-client relationship. In fairness, it is unclear that the court in Dobbin, or courts in any other case, used the term “very familiar” to refer specifically to the highest category of familiarity studied in the psychological literature.70. Courts do sometimes explicitly recognize that there are different degrees of familiarity. See e.g. JE, supra note 43 at para 30. The court noted that “in recognition cases the degree and circumstances of past associations between the witness and the accused play a pivotal role in the weight the trier of fact must give to the evidence.”71. D Michael Risinger et al, “The Daubert/Kumho Implications of Observer Effects in Forensic Science: Hidden Problems of Expectation and Suggestion” (2002) 90:1 Cal L Rev 1 at 12.72. See Keith A Findley & Michael S Scott, “The Multiple Dimensions of Tunnel Vision in Criminal Cases” (2006) 2006:2 Wis L Rev 291 at 308.73. See Saul M Kassin, Itiel E Dror & Jeff Kukucka, “The Forensic Confirmation Bias: Problems, Perspectives, and Proposed Solutions” (2013) 2:1 J Appl Research Memory & Cognition 42.SHERRIN, EARWITNESS EVIDENCE 833Misidentification of a familiar speaker … can occur through witness expectations. Recognition of familiar persons often depends on the closed set of people likely to be encountered in particular settings. People tend to hear who they expect to hear. Thus, if observers (police) expect to hear a particular person answer a telephone, misidentification of a familiar speaker may occur if someone else actually answers the call.74To my knowledge, only one experiment has sought specifically to determine if expectation had an effect on voice identification (finding that it did),75 but the phenomenon is so well established76 that it would be surprising if it did not have an effect. Indeed, some courts have recognized that expectation is a factor that should be taken into account in assessing earwitness testimony.77 The problem is that it is rare to find a case where consideration of expectation actually had an impact, even though the prosecution frequently tenders identification evidence that could have been tainted by expectation. Witnesses, usually (but not always) police officers involved in the investigation of the alleged offences, often testify to voice identifications first made after they know the accused has been arrested or singled out as a suspect—such as when they listen to a post-arrest interview and compare it (and only it) to a voice heard on intercepted communications.78 In such situations, witnesses could be affected by the expectation generated by the arrest or investigation that the voice they will hear is the voice they heard previously.79 Yet judges have commonly assessed, and even accepted, such identification evidence without any clear consideration of the potential biasing 74. Yarmey, “Psychology of Speaker Identification,” supra note 9 at 116.75. Peter Ladefoged, “Expectation Affects Identification by Listening” (1978) 21:4 Language & Speech 373.76. See Kassin, Dror & Kukucka, supra note 73 at 44.77. See e.g. Saddleback, supra note 43 at para 25; R v Willis, 2002 MBCA 138 at para 23, 170 Man R (2d) 22, Twaddle JA, dissenting [Willis]; R v Chan, 2001 BCSC 1180 at para 31, 51 WCB (2d) 225 [Chan].78. See e.g. Wu, supra note 48 at para 29. In Wu, “[k]nown samples of the appellants’ voices were acquired by [the Seargent] in a post-arrest interview.” Many cases have reviewed the constitutionality of the police using the post-arrest procedures to provide an opportunity for voice identification. See R v Oliynyk, 2008 BCCA 132 at paras 19-36, 232 CCC (3d) 411 (and also cases cited therein). R v Badgerow, 2010 ONSC 937 at paras 13-15 (available on QL) [Badgerow, (2010)]. Badgerow, (2010) is an example of a case where lay witnesses, after learning that the accused had been arrested, were asked if they recognized his voice on a tape-recorded phone call.79. They could also be affected by the related psychological phenomenon of motivational bias, where a person’s perceptions and interpretations are tainted by preference for a particular conclusion. See Kassin, Dror & Kukucka, supra note 73 at 45.(2015) 52 OSGOODE HALL LAW JOURNAL834effect of the context.80 Even when judges rely on additional independent evidence of identification (as they often do), the weight they give to the potentially tainted evidence should factor in the impact of expectation. Indeed, when an officer familiar with a voice on intercepted communications is asked to listen only to the accused to see if it is the same voice, the identification process is tantamount to a showup. Showups in the voice identification context have been shown to be especially prone to error.81 Less weight would normally be given to eyewitness identification evidence collected in a showup.82 It is difficult to understand why earwitness evidence should be treated differently.83B. EXPOSURE LENGTHIn assessing the reliability of earwitness testimony, courts have frequently considered the length of time that the witness was exposed to the perpetrator’s voice on the date(s) the crime was committed. The British Columbia Supreme Court in R v Savoy, for example, held that voice identification had not been proven partly because the witness only heard a “few sentences uttered by the robber in the course of a robbery which [the witness] estimates to have had a duration of about a half-minute.”84Empirical studies have generally affirmed that exposure length is related to accuracy. The research results have not been entirely consistent, but on the whole they show that “[t]he longer the opportunity to listen to a speaker, the greater the accuracy of identification.”85 Unfortunately, there is no magical tipping point before which voice identifications are not reliable and after which they are. It 80. See e.g. R v Alcantra, 2009 ABQB 524 at paras 86-89, 24 Alta LR (5th) 248; R v Nestorovski, 2013 ONSC 2523 at paras 32-41 (available on QL).81. Two studies have shown that false alarms are significantly higher in showups than in lineups. See Yarmey, Yarmey & Yarmey, supra note 12 at 459; Yarmey, “Earwitness Identification,” supra note 33 at 70-71. In the former study, the hit rate was lower than that expected by chance.82. R v Miaponoose (1996), 30 OR (3d) 419, [1996] OJ No 3216 (QL) (CA).83. In the context of a civilian earwitness, the Ontario Court of Appeal has commented that a showup procedure was “seriously flawed.” See Clouthier, supra note 42 at para 22.84. 2000 BCSC 296 at para 35, 46 WCB (2d) 67. For two other examples, see Dodd, supra note 42 at para 80; R v Whalen, 2007 NLSCTD 79 at para 65, 266 Nfld & PEIR 286.85. Yarmey, “Psychology of Speaker Identification,” supra note 9 at 120. Kenneth A Deffenbacher et al have suggested that the results of studies not showing an effect of exposure length may be partly attributable to the fact that most of the studies tested immediate rather than (the more forensically relevant) delayed recall. See Kenneth A Deffenbacher et al, “Relevance of Voice Identification Research to Criteria for Evaluating Reliability of an Identification” (1989) 123:2 J Psychol 109 at 111.SHERRIN, EARWITNESS EVIDENCE 835depends in part on familiarity. If one has had significant prior exposure to a voice (along with the identity of the person associated with it), there is a good chance, in ideal conditions, that the listener will be able to recognize it after hearing only a sentence or two.86 If the speaker is a stranger to the listener, one does not usually see hit rates in post-event lineups greater than 50% until more (and sometimes much more) than a minute of exposure during the event.87 False alarm rates for unfamiliar voices in target-absent lineups do not seem to change much as exposure increases from around twenty seconds to eight minutes, often hovering around 50%.88Some early research suggested that variability matters more than length, such that people may be able to identify voices accurately based on relatively short speech samples as long as the samples contain sufficient phonemic variety.89 86. See e.g. I Pollack, JM Pickett & WH Sumby, “On the Identification of Speakers by Voice” (1954) 26:3 J Acoustical Soc Am 403; Peter D Bricker & Sandra Pruzansky, “Effects of Stimulus Content and Duration on Talker Identification” (1966) 40:6 J Acoustical Soc Am 1441; Phil Rose & Sally Duncan, “Naïve Auditory Identification and Discrimination of Similar Voices by Familiar Listeners” (1995) 2:1 Forensic Linguistics 1. Some early research suggested that familiar listeners can accurately identify voices after hearing less than a single word. See e.g. Bricker & Pruzansky, supra note 86. More recent (and methodologically sound) research shows that suggestion to be overly optimistic. See e.g. Yarmey et al, supra note 63 at 295 (finding only a 30% hit rate and a 30% false alarm rate by familiar listeners after hearing one word). In R v Herman, the trial judge accepted that a familiar witness was able to identify a voice based on twice hearing the words “100 times” (supra note 58). That may have been overly optimistic.87. See e.g. A Daniel Yarmey & Eva Matthys, “Voice Identification of an Abductor” (1992) 6:5 Appl Cognit Psychol 367 at 370-71 [Yarmey & Matthys, “Abductor”] (hit rates close to 30% after 18 and 36 seconds of exposure and close to 50% after two and six minutes); Orchard & Yarmey,” supra note 35 at 253 (hit rates mostly below 50% after 30 seconds and mostly above 50% after eight minutes); A Daniel Yarmey, “Voice Identification Over the Telephone” (1991) 21:22 J Appl Soc Psychol 1868 at 1872 [Yarmey, “Over the Telephone”] (hit rate of 48% after 7.8 minutes but only 24% and 30% after 3.2 and 4.3 minutes); Kerstholt et al, “Effects of Speech Duration,” supra note 32 at 333 (hit rates all above 50% after 70 seconds, half below 50% after 30 seconds). Clifford’s study obtained much better hit rates. See Clifford, supra note 13 at 378-79.88. See e.g. Yarmey & Matthys, “Abductor,” supra note 87 at 373 (false alarm rates around 50% after 18 seconds, 36 seconds, two minutes and six minutes); Yarmey, “Over the Telephone,” supra note 87 at 1873-74 (false alarm rates of 48%, 51%, and 44% at 3.2, 4.3, and 7.8 minutes, respectively); Kerstholt et al, “Effects of Speech Duration,” supra note 32 at 333-34 (overall false alarm rate of 51% across 30 and 70 seconds). Curiously, in Orchard & Yarmey’s study, false alarm rates were sometimes worse after eight minutes than after 30 seconds. See Orchard & Yarmey, supra note 35 at 254.89. See e.g. Pollack, Pickett & Sumby, supra note 86 at 406; Bricker & Pruzansky, supra note 86 at 1444.(2015) 52 OSGOODE HALL LAW JOURNAL836To some extent, this is a distinction without a difference: Increased variety will often accompany increased duration. The research findings are also not entirely consistent,90 and the early research may only apply to familiar listeners,91 so it would be unsafe for courts to ignore exposure length. Certainly, as a practical matter, length is easier for a court to gauge than phonemic variety. The research findings regarding variability are more useful in providing reason to be extra cautious when considering short speech samples of limited variability. In R v Aulakh, for example, the evidence disclosed that the perpetrator of a sexual assault repeatedly uttered a single three-word phrase during the attack, which the complainant claimed to recognize from encounters with the accused somewhat earlier in the evening.92 While the corroborating evidence of identification in Aulakh was extremely powerful,93 voice identification evidence of that sort should probably be given no weight.94Another important finding in the research is that people are generally quite poor at estimating the length of time for which they heard someone speak on any particular occasion, often overestimating it by a significant margin.95 In many cases, the only evidence of exposure length will come from the earwitness. Courts should generally treat such evidence with skepticism.C. RETENTION INTERVALA factor listed by some courts as relevant to the assessment of earwitness testimony is retention interval: the amount of time between the criminal event and the 90. See especially Susan Cook & John Wilding, “Earwitness Testimony: Effects of Exposure and Attention on the Face Overshadowing Effect” (2001) 92:4 Brit J Psychol 617 at 622-23.91. Ibid at 618.92. 2012 BCCA 340 at para 19, 326 BCAC 177 [Aulakh].93. It suggested that the accused had exclusive opportunity to commit the offence.94. The trial judge in Aulakh gave the identification evidence some weight, albeit not much. The British Columbia Court of Appeal held that the evidence of exclusive opportunity was determinative. See Aulakh, supra note 92 at paras 38, 88.95. This may be particularly true for short speech samples. See e.g. Orchard &Yarmey, supra note 35 at 257 (mean estimates for 30-second and 480-second speech samples were 102 seconds and 517 seconds, respectively, and 93% of test subjects overestimated the 30-second sample); A Daniel Yarmey & Eva Matthys, “Retrospective Duration Estimates of an Abductor’s Speech” (1990) 28:3 Bull Psychonomic Soc 231 (98% of test subjects overestimated a 72-second speech sample by an average of four to one, and 51% of test subjects underestimated a five-minute speech sample).SHERRIN, EARWITNESS EVIDENCE 837witness’s closest other exposure(s) to the accused’s voice.96 The supposition is that, all things being equal, the reliability of the testimony decreases as the retention interval increases. In truth, consideration of this factor is not terribly common. While some courts may list it as a relevant factor, it is not often given much emphasis in the analysis.97 Some decisions address evidence relevant to retention interval mostly or entirely in the context of discussing other issues.98 Some decisions discount concerns about retention interval.99It is difficult to assess the mixed performance of the courts. The results of empirical studies, which have focused almost entirely on unfamiliar voices, have been somewhat mixed as to the impact of retention interval. Most show no real impact of intervals up to two weeks.100 The results of the few studies that have investigated intervals longer than that have been inconsistent, with one study finding a significant impact after three weeks101 and two others finding no or limited impact after three and even eight weeks.102 To further complicate matters, 96. See e.g. Chan, supra note 77 at para 31; Pabla, supra note 43 at para 23; Pinch, supra note 56 at para 75. The other exposure(s) will occur before the criminal event in the case of a familiar voice and after the event in the case of an unfamiliar voice, for example when the witness hears a voice lineup or showup.97. It occasionally is. See e.g. Saddleback, supra note 43 at paras 14, 26, 29, 34; Dodd, supra note 42 at paras 80-82.98. See e.g. Mackinaw, supra note 48 at 4, 10-13. Evidence relevant to retention interval was mentioned during the analysis of whether, rather than when, the earwitness had heard the accused’s voice before. Evidence of retention interval is often mentioned during the analysis of familiarity. See e.g. Dobbin, supra note 57 at paras 2, 32. Given the paucity of empirical studies of retention interval involving familiar voices (as discussed below), it is hard to say whether the two issues are substantially distinct.99. See e.g. Lamarche, supra note 68. The trial judge found a voice identification to be accurate even though the witness “had difficulty giving specific details about the timing and frequency of his dealings” with the accused (ibid at para 26). There was significant corroborating evidence, but the finding was not dependent on that evidence.100. See e.g. Öhman, Eriksson & Granhag, “Angry Voices,” supra note 10 at 63-64 (no significant effect of a two-week interval on adults, although hits by children decreased); Van Wallendael et al, supra note 24 at 666 (no significant differences across zero-, seven-, and fourteen-day retention periods); Read & Craik, supra note 39 at 12.101. Clifford, supra note 13 at 384-85. Accuracy was not significantly different after two weeks versus one week, but it was significantly different (reduced to chance level) after three weeks.102. Kerstholt et al, “Earwitnesses,” supra note 20 at 193-94 (finding a marginally significant decline in accuracy in a target-present lineup over one-, three-, and eight-week intervals but an improved performance in a target-absent lineup after three- and eight-week intervals versus a one-week interval); A Broeders & A Rietveld, “Speaker Identification by Earwitnesses” in Angelika Braun & Jens-Peter Köster, eds, Studies in Forensic Phonetics (Trier, Germany: Wissenschaftlicher Verlag Trier, 1995) 24 at 34 [on file with author] (little difference in performance after three weeks versus after one week).(2015) 52 OSGOODE HALL LAW JOURNAL838in the real world, intervals longer than two or eight weeks will occur.103 This is an area where the empirical studies do not offer maximal assistance to the courts. The general assumption is that memory for at least an unfamiliar voice, just like for most other things, tends to decline over time,104 but more research is needed into the effect of intervals of several weeks and months for both familiar and unfamiliar voices. For now, courts would probably be wise to assume that after a few weeks, retention interval has an increasingly detrimental, but not necessarily profound, effect on identification accuracy105—albeit one that may be less severe in the case of familiar voices. D. CONFIDENCEAnother factor that has been of some interest to the courts is how confident the earwitness is about his or her identification. Some judgments have been ambivalent about this factor, acknowledging that confidence is not determinative of reliability106 and that a confident witness can be mistaken.107 But the courts have not dismissed confidence as irrelevant, and it does seem to have factored into some decisions regarding reliability.108This approach to confidence is similar to the approach taken in the eyewitness identification context, where the courts have repeatedly cautioned against placing much weight on confidence.109 Curiously, the approach of the courts may even be stricter in the eyewitness context, where the Supreme Court of Canada has taken the position that there is a “very weak link between the confidence level 103. See e.g. R v Anderson, 2015 BCCA 300 at para 7, [2015] BCJ No 1556 (interval of as much as seven years); Mackinaw, supra note 48 at para 4 (interval of possibly “a few months”).104. See e.g. Yarmey, “Psychology of Speaker Identification,” supra note 9 at 121.105. See Kerstholt et al, “Earwitnesses,” supra note 20 at 195.106. See Masters, supra note 48 at para 47; R v Danaii-Asil, 2011 ONSC 2230 at para 20, 93 WCB (2d) 577.107. See Saddleback, supra note 43 at para 18. Arguably, in Pinch, the Ontario Superior Court of Justice went further in saying that “confidence does not substantially correlate to reliability.” See Pinch, supra note 56 at para 75. The Ontario Court of Appeal in Masters, however, seemed to interpret Pinch as saying that “witness confidence in voice identification does not make the identification reliable.” See Masters, supra note 48 at para 47.108. See e.g. R v Sharifi, 96 WCB (2d) 462, [2011] OJ No 3985 (QL) at para 13 (Sup Ct); Sanghera, supra note 68 at paras 145, 176; Anderson, supra note 103. In no case was confidence considered to be the sole determinant of reliability, but it was one factor taken into account.109. See e.g. R v Jack, 2013 ONCA 80 at para 30, 302 OAC 137. The court discussed the importance of “caution[ing] about the diminished correlation between a witness’ confidence level and his or her accuracy” (ibid).SHERRIN, EARWITNESS EVIDENCE 839of a witness and the accuracy of that witness.”110 This is curious because in the eyewitness identification context, empirical studies have shown some (qualified) correlation between confidence and accuracy,111 whereas the same cannot be said in the earwitness context. Numerous studies have investigated the relationship between confidence and voice identification accuracy, and the general conclusion is that there is very little, if any, correlation between the two.112 The findings have not been entirely consistent, but many studies have found no significant correlation113 and many others have found only a modest correlation in only some experimental conditions.114 In a few instances, studies have even found a negative correlation, with confident witnesses being less likely to be correct than unconfident witnesses.115 This is yet another illustration of why earwitness testimony must be treated with greater caution than eyewitness testimony, not 110. R v Hibbert, 2002 SCC 39 at para 52, [2002] 2 SCR 445. Arguably, the courts have sometimes even gone further and recommended that no weight should be given to witness confidence. See e.g. R v MacDonald, 2014 ONCA 610 at para 15 (available on QL); R v Knox (2006), 80 OR (3d) 515 at paras 54-55, 211 OAC 164 (CA).111. See e.g. Neil Brewer & Gary L Wells, “Eyewitness Identification” (2011) 20:1 Curr Directions Psychol Sci 24 at 24-25.112. See Yarmey, “Psychology of Speaker Identification,” supra note 9 at 110. Yarmey noted that “[c]onfidence-accuracy correlations in most studies on voice identification for unfamiliar speakers are nonsignificant or are relatively low.”113. See e.g. Öhman, Eriksson & Granhag, “Overhearing the Planning,” supra note 21 at 124; Yarmey, “Earwitness Descriptions,” supra note 28 at 119; Goggin et al, supra note 37 at 457; Kerstholt et al, “Effects of Speech Duration,” supra note 32 at 335; Kerstholt et al, “Earwitnesses,” supra note 20 at 194-95.114. See e.g. Yarmey, “Earwitness Identification,” supra note 33 at 70 (small but significant correlation in only one of four experimental conditions, with no correlation in the other three); Yarmey, “Over the Telephone,” supra note 87 at 1871-72 (modest correlation found only for target-absent lineup, not for target-present lineup); Thompson, supra note 30 at 128 (correlation only approached or reached significance when outliers were included); Philippon et al, supra note 29 at 547 (significant correlation found for target-absent lineups but not for target-present lineups or overall); Smith & Baguley, supra note 26 at 63 (significant correlation found for male-voice, target-present lineup, but no significant correlation for female-voice or for target-absent lineups); Timothy J Perfect, Laura J Hunt & Christopher M Harris, “Verbal Overshadowing in Voice Recognition” (2002) 16:8 Appl Cognit Psychol 973 at 977.115. Yarmey & Matthys, “Abductor,” supra note 87 at 373 (significant negative correlations found in some experimental conditions relating to voice samples of 18 and 36 seconds, although modest positive correlations found in some experimental conditions relating to longer voice samples); Orchard & Yarmey, supra note 35 at 256-57 (negative correlations found in experiments involving distinctive voices, although small positive correlations found when results were collapsed across all experimental conditions).(2015) 52 OSGOODE HALL LAW JOURNAL840with the same or less caution.116 In general, the safest and wisest position for the courts to adopt is to completely disregard confidence as any indicator of the accuracy of voice identifications.117 There may be one qualification to add to the above. There is some evidence that confidence may have a useful correlation with accuracy when it comes to quite familiar voices. Two studies found a significant correlation between confidence and accuracy for very familiar listeners: family members and close friends.118 Another study found a correlation when using the voices of generally well-known public figures.119 It may be premature to make too much of these findings, though. Methodological differences make an overall assessment of the studies somewhat difficult.120 One of the aforementioned studies also obtained inconsistent results across two experiments.121 And an entirely different study found no significant correlation at all, although that finding may be the result of not using any test subjects who were as familiar as family members and close friends.122 If the courts are to consider confidence when assessing identifications made by familiar earwitnesses, they should only do so in relation to the most familiar category of witnesses. An interesting question not really addressed in the empirical literature is whether lack of confidence should be taken into consideration.123 Courts certainly consider lack of confidence to be a relevant and sometimes important factor undermining the reliability of identifications,124 and the position may be 116. See Olsson, Juslin & Winman,” supra note 18 at 116. They state: “In comparable information-processing circumstances, the forensic system should place more trust in eyewitness than in earwitness confidence.” See also Stevenage, Howland & Tippelt, supra note 8 at 115-17.117. See e.g. Kerstholt et al, “Earwitnesses,” supra note 20 at 196.118. Rose & Duncan, supra note 86 at 14; Yarmey et al, supra note 63 at 289 (only when speakers spoke in a normal tone of voice).119. Van Lancker, Kreiman & Emmorey, supra note 59 at 28-29.120. For example, unlike other studies, Rose & Duncan included several familiar voices (rather than just one) in the lineups and included more familiar voices than foils. See Rose & Duncan, supra note 86.121. In the second of two experiments, Yarmey and his colleagues did not find a significant confidence-accuracy correlation for very familiar speakers (at least using one method of analysis). Indeed, the results of the second experiment were, in one respect, the converse of the results of the first, a result the researchers found “difficult to explain.” See Yarmey et al, supra note 63 at 297.122. Read & Craik, supra note 39 at 16.123. Studies occasionally provide some relevant information but no sustained research into the issue has been undertaken.124. See e.g. Pinch, supra note 56 at para 75; JE, supra note 43 at para 44.SHERRIN, EARWITNESS EVIDENCE 841defensible simply as a function of the burden of proof.125 But as matters currently stand, we do not truly know whether lack of confidence is indicative of lack of accuracy. It may be that confidence of any sort is entirely unrelated to accuracy. Further assistance from earwitness researchers would be welcome.E. ATTENTIONA factor to which the courts make surprisingly little reference is attention: the degree to which the earwitness made a conscious effort during the crime to pay attention to the characteristics of the perpetrator’s voice. The case law lists it as a relevant factor,126 and courts sometimes explicitly consider it.127 But given that it could be relevant to essentially every single voice identification case, it is surprising that it is not more often given more explicit consideration.128Judges would be wise to give greater consideration to earwitness attention. Several studies have determined that subjects who are told in advance that they will be tested on their ability to identify a voice perform better than subjects who are not told.129 The positive effect of attention has not always been found,130 and it is certainly not the case that a witness who specifically attended to the perpetrator’s voice will always be correct (or, conversely, that a witness who did not will always be wrong). But all things being equal, paying attention to the characteristics of a voice improves the chances of later being able to identify it, and vice versa. It may be tempting to apply this conclusion to cases where the perpetrator’s voice is captured on tape and someone—be it a police officer, judge, or jury member—later compares it to the voice of the accused, on the basis that in those cases attention is being closely paid to the characteristics of the voice. That would be unwise. As explained above,131 earwitnesses in those cases can be affected by 125. Although it is not clear that the burden of proof compels a particular response to factors the value, rather than the existence, of which is in question.126. See e.g. Chan, supra note 77 at para 31.127. See e.g. R v Bubar, 2006 NBPC 24 at paras 94, 101, 305 NBR (2d) 19 [Bubar].128. In fairness, courts may sometimes implicitly consider it, perhaps especially in wiretap cases where officers listen to a sample of the accused’s voice for the specific purpose of comparing it to a voice recorded on tape.129. Saslove & Yarmey, supra note 11 at 113; Yarmey, “Earwitness Identification,” supra note 33 at 69; Heather A Armstrong & Stuart J McKelvie, “Effect of Face Context on Recognition Memory for Voices” (1996) 123:3 J General Psychol 259 at 265. Perfect and his colleagues found an effect in the same direction, but it did not quite reach statistical significance. See Perfect, Hunt & Harris, supra note 114 at 976. See also Clifford, supra note 13 at 382-83.130. Cook & Wilding, supra note 90 at 621.131. See Part III(A), above.(2015) 52 OSGOODE HALL LAW JOURNAL842expectancy bias. No study on earwitness attention employed a scenario where test subjects were told in advance whose voice it might be (never mind that it is alleged to be). It is simply not known whether the positive effect of attention would be attenuated or even completely undone by the effect of expectation. Until we know more, the findings relating to the effect of attention should only be applied to cases where earwitnesses attend to the characteristics of a voice without previously being told who the speaker could be—or, perhaps more frequently, to cases where earwitnesses do not specifically attend to the characteristics of a voice. F. ABNORMAL SPEAKINGSomeone perpetrating a crime may not use his or her normal voice during the event. The person may deliberately attempt to disguise it so as not to be identified. She or he may also speak in a way that is unusual for situational reasons: because of the demands of the event (which require yelling or the use of an angry tone), because of the stress and emotionality of the situation,132 or because of intoxication.133 Canadian courts have included disguise as a factor that diminishes the accuracy of voice identification.134 They have also tried to take into account situation-driven changes in voice.135 In both respects, they have been correct to do so. However, it may be that courts should be even more cautious.Four different studies have investigated the effect of having speakers intentionally disguise their voices in almost any way they wanted. The hit rate for extremely familiar listeners was 79% (leaving a 21% error rate).136 The hit rate for unfamiliar listeners in one study was 61%137 but in three others was quite low: Subjects who compared a disguised voice with a speaker’s 132. Bull & Clifford, supra note 38 at 111.133. Yarmey, “Psychology of Speaker Identification,” supra note 9 at 118.134. See e.g. Saddleback, supra note 43 at para 25; Bubar, supra note 127 at para 86; Whalen, supra note 84 at para 65; Pinch, supra note 56 at para 75.135. See e.g. Saddleback, supra note 43 at para 29; Bubar, supra note 127 at paras 86, 93; Herman, supra note 58 at para 106; Pinch, supra note 56 at para 75.136. Hollien, Majewski & Doherty, supra note 60 at 142. Test subjects were classified as “extremely familiar” with the target voices when they not only knew the speakers but also correctly identified all ten of the speakers, without any errors, in a pre-test assessment.137. Alan R Reich & James E Duke, “Effects of Selected Vocal Disguises upon Speaker Identification by Listening” (1979) 66:4 J Acoustical Soc Am 1023 at 1025. This was the hit rate when speakers were permitted to disguise their voices in a manner they thought would best conceal their identity. The hit rates were slightly higher when speakers were directed to use specific forms of disguise, reaching as high as 70% when speakers spoke at an extremely slow rate.SHERRIN, EARWITNESS EVIDENCE 843normal voice made correct identifications at most one-quarter of the time.138 Regrettably, methodological issues make it a little difficult to interpret those results. The authors of the study with the highest hit rate for unfamiliar listeners acknowledged that the rate was likely inflated by the fact that the study simply asked listeners to decide whether individual pairs of speech samples came from the same speaker.139 Conversely, it is possible that the methodology of one of the other studies depressed the hit rates.140 That disguise can have a substantial impact on identification accuracy is supported, however, by the results of two other studies that examined the use of one particular form of disguise—whispering. Whispering can act as a disguise because it conceals vocal characteristics such as pitch, inflection, and intonation.141 When test subjects heard whispered unfamiliar voices, they again generally achieved hit rates less than or equal to 25%.142 False alarm rates were also high when subjects heard a voice lineup,143 and the misidentification rate was over 50% when subjects were simply exposed to an unfamiliar voice and asked if they could identify who was speaking (the presumption being that they could not).144 These findings held true even when test subjects were specifically directed to attend to the voice for purposes of later identification,145 when distinctive 138. Clifford, supra note 13 at 382 (26% hit rate); Hollien, Majewski & Doherty, supra note 60 at 142 (18%, 17%, 25%, and 21%, across four different groups); Read & Craik, supra note 39 at 13-14 (22% and 26%).139. Reich & Duke, supra note 137 at 1027.140. Hollien, Majewski & Doherty, supra note 60 at 141. Listeners were not asked to select a previously heard voice from a voice lineup. Instead, listeners were given pre-test training on the names associated with the ten different unfamiliar speakers and then, in the experiment, later asked to choose the correct name from a supplied list when hearing the voices played on tape. Test subjects may have found it difficult to digest the pre-test information and keep track of the various voice-name associations. That said, subjects were given a preliminary identification test using only undisguised voices, and even those who scored quite well performed poorly in the actual experiment. 141. Yarmey, “Psychology of Speaker Identification,” supra note 9 at 118.142. Orchard & Yarmey, supra note 35 at 253 (across four experimental conditions, hit rates were 25%, 25%, 17%, and 33%, with the highest rate achieved only after subjects heard the whispered voice for eight minutes); Yarmey et al, supra note 63 at 292 (20% hit rate after two minutes of speech).143. Orchard & Yarmey, supra note 35 at 254, 256 (53% to 83% across four experimental conditions). 144. Yarmey et al, supra note 63 at 292 (58% misidentification rate after hearing the whispered voice for two minutes). In Orchard & Yarmey’s study using target-absent voice lineups, lineup members specifically chosen to serve as the innocent suspects were selected 12% of the time across the four lineups.145. Orchard & Yarmey, supra note 35 at 252; Yarmey et al, supra note 63 at 291.(2015) 52 OSGOODE HALL LAW JOURNAL844voices were used,146 and/or when exposure length was substantial.147 Familiar voices did not always substantially change the results. In one study, test subjects hearing very familiar voices achieved a 77% hit rate (and a 15% misidentification rate) but achieved a 35% hit rate and a 39% misidentification rate when hearing moderately familiar voices.148Similar or even worse results have been obtained in experiments where speakers employ tones of voice (anger or emotionality) that may often be used in the commission of crimes. Across six different experiments, the highest hit rate obtained was 33%; most rates were at the level of chance.149 This was true even when test subjects were informed in advance that it was a voice identification test,150 were tested immediately after exposure to the voice,151 and/or were exposed to as much as 40 seconds of speech.152 The one study to record false alarms obtained rates between 43–53%.153 Even using listeners who were mildly familiar with the speakers did not improve the results.154All the results reported above reflect the accuracy rates achieved when test subjects were asked to compare the voice of an individual using an abnormal voice with the voice of an individual using a normal speaking voice. This is important. The available evidence does not indicate that disguised, angry, or emotional voices are inherently more difficult to identify than unemotional voices. The generally poor results were more likely a product of comparing normal to abnormal voices. Hit rates in studies where test subjects hear an abnormal voice both initially and 146. Orchard & Yarmey, supra note 35 at 253-54.147. Ibid (eight minutes); Yarmey et al, supra note 63 at 287 (two minutes).148. Yarmey et al, supra note 63 at 292. A moderately familiar speaker was a co-worker, teammate, club-mate, or general friend of the listener. Pollack, Pickett, and Sumby found that for familiar voices, equivalent hit rates were not obtained until subjects heard speech at least three times as long as a non-whispered sample. See Pollack, Pickett & Sumby, supra note 86 at 405.149. Saslove & Yarmey, supra note 11 at 113, 115 (13-33% hit rate, rates that were sometimes lower than chance); Read & Craik, supra note 39 at 10, 12, 14-15 (19%, 20%, and 22% across three experiments—all not significantly different from chance); Bull & Clifford, supra note 38 at 114 (33%); Öhman, Eriksson & Granhag, “Angry Voices,” supra note 10 at 64, 66 (chance level 5-14%).150. Saslove & Yarmey, supra note 11 at 113.151. Ibid; Öhman, Eriksson & Granhag, “Angry Voices,” supra note 10 at 64.152. Öhman, Eriksson & Granhag, “Angry Voices,” supra note 10 at 61.153. Ibid at 64.154. Read & Craik, supra note 39 at 11-13. Mildly familiar listeners had previously heard the speaker for more than one hour. The study excluded anyone who was familiar enough with a speaker’s voice to identify it on initial exposure.SHERRIN, EARWITNESS EVIDENCE 845at test, for example, were generally (albeit not always) significantly higher.155 One experiment even specifically tested whether voices were differently memorable depending on whether they were emotional, finding no evidence that they were.156This suggests that greater reliance can be placed on identifications by earwitnesses who are exposed to the same kind or tone of voice both during the crime and at another time; the similarity of exposure can assist them in making an accurate comparison and identification.157 Even then, however, the identification can be suspect. The one study that specifically sought to explore the benefits of similar exposures found that hit rates for both unfamiliar and mildly familiar earwitnesses, while improved, still did not usually rise above 50%;158 the rates were even lower when the two exposures were not perfectly identical.159 Regrettably, the initial exposure in that study was quite brief (only four seconds), so the generalizability of its conclusions is uncertain. But even if longer exposures would produce better results (as seems likely), it may not be common in the real world for the various exposures to be similar enough to generate high levels of accuracy. The authorities cannot force a suspect to speak at all,160 never mind in a particular way, and it may be difficult for them to obtain, through observation of the suspect, a voice sample similar in kind or tone to that used during the commission of the offence, especially if it was a deliberately disguised voice. G. TELEPHONE TRANSMISSIONCourts have been concerned about the possibility that hearing a voice over the telephone can interfere with the ability to identify it accurately. In R v Garofalo, for example, it was stated that “[v]oice identification, by itself, is fraught with problems. It is even more uncertain when a voice is identified over the 155. Ibid at 10; Saslove & Yarmey, supra note 11 at 113; Orchard & Yarmey, supra note 35 at 254 (one experimental condition only).156. Read & Craik, supra note 39 at 14-15.157. Some courts seem to have assumed as much. See e.g. JE, supra note 43 at paras 13, 38; Saddleback, supra note 43 at para 29.158. Read & Craik, supra note 39 at 10, 13-14. The most familiar listeners had previously heard the speaker’s voice for more than one hour. All the other listeners had previously heard the voice, if at all, for at most an hour. 159. Ibid at 13-14. In one variation of the experiment, test subjects were exposed to emotional statements both initially and at exposure, but the test statement was a re-recording, by the same speaker, of the original statement rather than a replaying of the original recorded statement. The hit rate never exceeded 38%.160. R v Turcotte, 2005 SCC 50 at para 52, [2005] 2 SCR 519. The suspect also cannot be compelled to participate in a lineup. See Henry, supra note 5 at paras 40-58.(2015) 52 OSGOODE HALL LAW JOURNAL846telephone.”161 Commentators have likewise been concerned, hypothesizing that recognition may be impaired by the limited range of sound frequencies transmitted over the phone.162In addition, concerns have been expressed about distortions introduced when earwitnesses attempt to compare speech heard live with speech heard over the telephone. These concerns are less commonly found in the case law163 (indeed, they frequently appear to be entirely absent),164 but one does occasionally come across indications that courts are at least comforted by evidence that an earwitness purporting to identify a voice heard over the phone had previously heard the accused speak over the same medium.165Surprisingly, neither of these concerns has been validated by empirical research. Only limited research has been undertaken, but the majority of the existing studies have failed to find a significant difference in identification accuracy between voices heard over the telephone and voices heard live.166 This is true using both target present and target absent lineups.167 Furthermore, no research 161. 2012 ONSC 6351 at para 111 (available on QL). Cf Pinch, supra note 56 at para 75. The court referred to “difficulties which the telephone imposes on voice identification because of the loss of acoustic information.” See also Whalen, supra note 84 at para 61.162. See e.g. Yarmey, “Psychology of Speaker Identification,” supra note 9 at 122. Yarmey noted that “[s]peaker identifications involving the use of the telephone may have particular problems to overcome ... . Degradations of the speech signal ... are common because telephone lines typically transmit a band of frequencies between 300 and 3,400 Hertz whereas human voices may contain components up to about 12,000 Hertz.”163. They are found in the literature. See e.g. HA Rathborn, RH Bull & BR Clifford, “Voice Recognition Over the Telephone” (1981) 9 J Police Sci & Admin 280 at 282-84.164. Identifications of voices heard on telephone wiretaps are commonly made based on comparisons with live speech without courts considering the potential effect of witnessing across different mediums. See e.g. Wu, supra note 48 at paras 29-30, 47-53.165. For example, in a case where the accused was found to have uttered a threat to kill over the telephone, the trial judge mentioned that the recipient of the threat had previously spoken with the accused “on the phone and in person, ten to fifteen times.” See Dobbin, supra note 57 at para 2.166. Yarmey, “Earwitness Identification,” supra note 33 at 69, 71; Perfect, Hunt & Harris, supra note 114 at 976; Öhman, Eriksson & Granhag, “Mobile Phone Quality,” supra note 17 at 172-74. Rathborn, Bull, and Clifford, however, found that witnesses who heard telephone voices either at exposure or in the lineup, or both, were significantly less accurate than witnesses who were only exposed to non-telephone voices. See Rathborn, Bull & Clifford, supra note 163 at 283.167. Yarmey, “Earwitness Identification,” supra note 33 at 69.SHERRIN, EARWITNESS EVIDENCE 847currently supports the notion that accuracy is impaired when an earwitness compares a voice heard over the telephone to a voice heard live, or vice versa.168This does not mean that judges should not concern themselves with possible impairments to identification accuracy when voices are heard over the telephone. The quality of a particular transmission may be poor, there may be lots of ambient noise (for example, when someone uses a mobile or public telephone),169 and some individuals may even speak differently over the phone.170 What the empirical studies do suggest, however, is that the mere fact that the communication is over a telephone is not necessarily cause for concern. They also indicate that police forces need not worry too much about specifically using telephone voices when constructing voice lineups,171 and courts need not worry when the police do not use them. All that said, more research is needed before any definitive conclusions can be drawn.172 For instance, it may be important that no study has compared exposure to truly live telephone voices with exposure to truly live voices both initially and at test; subjects are generally exposed only to tape-recorded voices.173 There are very understandable practical reasons for this,174 but given that the studies are trying to ascertain the possible impact of a medium of transmission (the telephone), it is a bit worrying that the experiments have used a second medium of transmission (the tape recorder), which may have its own impact.168. Rathborn, Bull & Clifford, supra note 163 at 283-84 (no significant difference in accuracy between telephone-telephone conditions and telephone-non-telephone conditions); Kerstholt et al, “Earwitnesses,” supra note 20 at 194 (no significant differences between telephone-non-telephone conditions and either telephone-telephone or non-telephone-non-telephone conditions); Öhman, Eriksson & Granhag, “Mobile Phone Quality,” supra note 17 at 175 (the study’s results “imply … that using a mobile phone recorded voice line-up when the voice is originally heard over a mobile phone is not likely to improve identification accuracy”).169. For a related example, see Dodd, supra note 42. A voice was heard over a baby monitor through static and the moaning of another man.170. See Foulkes & Barron, supra note 61 at 182. Foulkes and Barron noted that “[i]t is … well known that some speakers adopt a ‘telephone voice’, modifying their rate of speech, segmental pronunciations, and/or voice quality.”171. The police sometimes do this. See e.g. R v Nichols, [2004] OJ No 6186 (QL) at paras 18-26 (Ct J) [Nichols].172. Kerstholt and his colleagues, for example, were unable to offer an explanation for why the results differed between the Rathborn, Bull, and Clifford study and the remaining empirical studies. See Kerstholt et al, supra note 20 at 196. I am in the same position.173. Yarmey used live voices during initial exposure but tape-recorded voices for the lineups. See Yarmey, “Earwitness Identification,” supra note 33 at 64-67.174. It could be very expensive, time-consuming, and organizationally challenging to use truly live voices. It could also introduce methodological problems in maintaining consistency across tests.(2015) 52 OSGOODE HALL LAW JOURNAL848H. UNFAMILIAR ACCENTS AND LANGUAGESWitnesses are sometimes asked to identify the voice of a criminal perpetrator who spoke with an accent or even in a foreign language. A few judges have been concerned about the opportunity for error in these circumstances. Justice Twaddle of the Manitoba Court of Appeal, for example, once wrote of the “danger, where the accused has an accent, that the witness is identifying the accent rather than the particular voice of the accused.”175 That is, in essence, a concern that people may not be able to distinguish readily between voices speaking in a manner that is unfamiliar to the witness. It is a concern that has been borne out by the empirical research.A number of studies have tested the accuracy of individuals trying to identify the voice of someone who speaks the same language as the listener but with a different accent. With English speech, the listeners tested have come from the United States, England, Scotland, and Australia, and the accents studied have included Spanish, British, Scottish, and Taiwanese. Other experiments have used non-English speech, listeners from other countries, and other accents. In most cases, the bottom line result has been the same: People are worse at identifying accented voices than non-accented voices.176 The difference has not always reached statistical significance (although it often has), but the overall trend is 175. Willis, supra note 77 at para 24.176. Stevenage, Clarke & McNeill, supra note 27 at 650-51 (listeners from Glasgow, Scotland, and Southampton, England significantly worse at identifying speakers from the foreign versus home location in target-present lineups, while listeners from Southampton, but not Glasgow, made significantly more errors with Glaswegian versus Southampton speakers in target-absent lineups); Thea Vanags, Marie Carroll & Timothy J Perfect, “Verbal Overshadowing: A Sound Theory in Voice Recognition?” (2005) 19:9 Appl Cognit Psychol 1127 at 1133-34 (Australian listeners significantly better at identifying speakers from Australia versus England); Kerstholt et al, “Earwitnesses,” supra note 20 at 193 (Dutch listeners better at identifying “ordinary standard-accented” Dutch voices than “regionally and socially strongly marked (The Hague) accent[s],” although the difference was only marginally significant); Goggin et al, supra note 37 at 454 (monolingual English listeners from Texas significantly better at identifying Midwestern American speakers using their normal voice than when they used a heavy Spanish accent, although the same was not true for English-Spanish bilingual listeners); Thompson, supra note 30 at 124, 126 (Midwestern American listeners better at identifying Midwestern American English speakers versus the same speakers using a heavy Spanish accent, although the difference did not reach statistical significance and was only found in a target-present lineup); Alvin G Goldstein et al, “Recognition Memory for Accented and Unaccented Voices” (1981) 17:5 Bull Psychonomic Soc 217 at 218-19 (identification by Midwestern American listeners of Taiwanese-accented versus Midwestern American voices significantly worse when the initial exposure was a single word, although not when it was a single sentence).SHERRIN, EARWITNESS EVIDENCE 849consistent. In many cases, the difference in accuracy has been in the range of 20%. Interestingly, the problem does not just arise when people are hearing accents that are very foreign or different from their own. In fact, the study that obtained the greatest difference in accuracy was one that asked listeners in southern England to identify voices of people from Glasgow.177Studies testing the accuracy of individuals trying to identify voices of people speaking a foreign language that the listener does not understand have obtained similar results. English-speaking listeners have been tested when hearing people speaking Spanish, German, and French. Polish-, German-, Chinese-, and Spanish-speaking listeners have been exposed to people speaking German or English. In most cases, listeners were worse at identifying voices speaking an unfamiliar language than a familiar language.178 A few studies have tried to investigate whether accuracy is improved by partial, albeit incomplete, familiarity 177. Stevenage, Clarke & McNeill, supra note 27 at 650 (50% difference in a target-absent lineup for both correct rejections and false alarms).178. Olaf Köster & Niels O Schiller, “Different Influences of the Native Language of a Listener on Speaker Recognition” (1997) 4:1 Forensic Linguistics 18 at 23-25 (Spanish- and Chinese-speaking listeners significantly worse than German-speaking listeners at identifying a voice speaking German); Goggin et al, supra note 37 at 451-54 (English- and German-speaking listeners both significantly less accurate when hearing the unfamiliar versus the familiar language, and English-speaking listeners also significantly less accurate when identifying Spanish versus English speakers); Neils O Schiller & Olaf Köster, “Evaluation of a Foreign Speaker in Forensic Phonetics: A Report” (1996) 3:1 Forensic Linguistics 176 at 179 (listeners who only spoke English significantly worse than native German speakers at identifying German-speaking voices); Philippon et al, supra note 29 at 545 (English-speaking listeners from the United Kingdom had a substantially higher false alarm rate when hearing a French versus English speaker in both target-present and target-absent lineups, although the hit rate was identical); Thompson, supra note 30 at 124, 126, 128 (in two of three experiments, English-speaking Americans significantly or marginally more accurate in identifying English speakers than Spanish speakers, while no difference was found in the third experiment that only used a target-absent lineup, but the overall false alarm rate in that lineup was 56% and the overall correct rejection rate was only 13%); Hollien, Majewski & Doherty, supra note 60 at 142-44 (Polish-only speakers less accurate than English speakers in identifying a voice speaking English).(2015) 52 OSGOODE HALL LAW JOURNAL850with a language. It does seem to help.179 Unfortunately, it does not always eliminate the difference in accuracy,180 and current research does not indicate how much familiarity is necessary to have a substantial impact.181Decreased accuracy in identifying accented and foreign language-speaking voices is probably attributable to an expertise effect.182 By repeated exposure to a language or accent, listeners learn to perceive, understand, and attend to subtle but important differences between speakers in syntax, pronunciation, emphasis, and so forth. Listeners unfamiliar with the meaning or manner of speech miss those clues and thus make more errors in identification.183 As stated above, judges have sometimes implicitly recognized this problem. Some judges, however, have inappropriately labelled this a problem of “cross-racial” voice identification.184 They are presumably borrowing from the eyewitness identification context in which “cross-racial” identification has been classified as a matter of particular concern,185 but it would be better to refer to, and focus on, unfamiliar accents and languages. This was perhaps best illustrated by a 1998 study that found that 179. Schiller & Köster, supra note 178 at 178-79 (English-speaking listeners with some knowledge of German better than English-only listeners in identifying a voice speaking German); Köster & Schiller, supra note 178 at 24-25 (“[i]n all cases (English, Spanish, Chinese), subjects with a knowledge of the target language (German) were able to identify a German speaker better than subjects without any knowledge of the target language”); Kirk PH Sullivan & Frank Schlichting, “Speaker Discrimination in a Foreign Language: First Language Environment, Second Language Learners” (2000) 7:1 Forensic Linguistics 95 at 98-99, 105-07 (some knowledge of Swedish improved the accuracy of native English speakers when identifying a voice speaking Swedish, although the improvement did not increase as knowledge of Swedish increased).180. Köster & Schiller, supra note 178 at 23 (Chinese and Spanish listeners with some knowledge of German both significantly less accurate than native German listeners when identifying a voice speaking German). But see Schiller & Köster, supra note 178 at 180 (English-speaking listeners with some knowledge of German as accurate as native German-speaking listeners in identifying a voice speaking German). Regretfully, Sullivan and Schlichting’s study did not compare its results with native Swedish-speaking listeners. See Sullivan & Schlicting, supra note 179.181. The studies were generally imprecise as to the amount of knowledge the test subjects had of the foreign language.182. See e.g. Stevenage, Clarke & McNeill, supra note 27. 183. Identification accuracy is affected not just by repeatedly hearing the sounds of a foreign accent or language but also by understanding the content of the words spoken. See e.g. Köster & Schiller, supra note 178 at 25.184. See e.g. Pinch, supra note 56 at para 75. The court mentioned “the dangers and potential prejudice of cross-racial voice identification evidence.” See also Masters, supra note 48 at para 47.185. See e.g. R v Richards (2004), 70 OR (3d) 737 at para 32, 186 CCC (3d) 333 (CA).SHERRIN, EARWITNESS EVIDENCE 851British and American listeners, Caucasian and non-Caucasian alike, were all significantly better at identifying voices of various races from their home country than from a foreign country.186 Lack of familiarity can coincide with dissimilarity of race but is really a function of differences in geography, culture, education, and socioeconomic status.187I. DISTINCTIVENESSCourts quite frequently refer to the distinctiveness of the perpetrator’s voice (or lack thereof ) as a basis for assessing the reliability of a voice identification. This is not a reference to the perpetrator’s manner of speaking, but rather to some inherent characteristic of the voice, such as pitch, nasality, or hoarseness.188 Several courts, for example, have asked whether there is “some peculiarity or distinctiveness to the [accused’s] voice that would make it more readily identifiable.”189 A finding that there is enhances reliability.190 A finding that there is not sometimes diminishes reliability.191Two studies have specifically tried to examine voice distinctiveness. The first found that distinctive voices were more accurately identified than non-distinctive voices in some experimental conditions but less accurately identified in other 186. Nathan Daniel Doty, “The Influence of Nationality on the Accuracy of Face and Voice Recognition” (1998) 111:2 Am J Psychol 191 at 196, 202-204. Voices from the United States, England, Belize, and France were used. They were predominantly Caucasian voices, but a substantial minority was non-Caucasian. Listeners were Caucasian, African-American, African-English, Hispanic, and Middle Eastern.187. See e.g. Yarmey, “Psychology of Speaker Identification,” supra note 9 at 114. Yarmey noted that “accents differ as a function of factors other than simply race/ethnicity. Circumstances such as socioeconomic factors, education, historical and political groupings over time, and geographical regions of various sizes may play a role in affecting accents.”188. Courts do sometimes refer to the distinctiveness of the perpetrator’s manner of speaking (for example, repeated use of a particular phrase), but I do not examine that factor here due to the paucity of available research on the topic.189. See e.g. Saddleback, supra note 43 at para 25; JE, supra note 43 at para 29; Whalen, supra note 84 at para 47. All these cases are citing language mentioned in Williams. See Williams, supra note 4 at para 13.190. See e.g. Pinch, supra note 56 at para 87.191. See e.g. Clouthier, supra note 42. The court discounted an earwitness’s evidence that the robber had an Ottawa Valley accent on the basis that “this robbery occurred in a place where such an accent is commonplace” (ibid at para 20).(2015) 52 OSGOODE HALL LAW JOURNAL852conditions, with a substantial number of errors being made.192 The second study found a more consistent (positive) impact of distinctiveness193 but only in experimental conditions where the target was absent from the lineup.194 These two studies leave one in a state of uncertainty regarding the impact of voice distinctiveness. Perhaps more importantly, they really tell us nothing about what specific (allegedly distinctive) voice characteristics facilitate identification. The voices labelled distinctive in the studies were judged to be so by the experimenters and by the people assisting them, but no indication was given as to the characteristic(s) of the voices that made them distinctive or even whether they were all distinctive in the same way.195 In the first study, in fact, a distinctive voice was simply defined as “highly striking and not likely to be confused with other voices.”196 The two studies, therefore, really only show that some voices are distinctive enough to allow, to some extent, for easier identification and that a group of people is, to some extent, capable of spotting such voices. That is of little assistance to a court charged with assessing the weight to be given to an identification made by a single witness based in whole or in part on that witness’s personal belief that the voice is distinctive in some particular or undefined way. It seems obviously true that some voices will be easier to identify than others, but there are real dangers in relying on distinctiveness as a basis for assessing the reliability of an identification. If the particular distinctive characteristics of a voice are not specified by the earwitness, the evidence that the witness perceives it to be distinctive is of no assistance on the issue of reliability. Indeed, it raises the danger that a court will determine the reliability of a voice identification by determining credibility—i.e., by reference to the fact that the earwitness sincerely asserts that the voice was distinctive and thus identifiable. One cannot forget that a voice might seem distinctive to a witness only because it is unfamiliar to 192. Orchard & Yarmey, supra note 35 at 256. Distinctive voices were properly rejected more commonly than non-distinctive voices in a target-absent lineup when speech was whispered both at exposure and at test. Distinctive voices were less commonly identified than non-distinctive voices in a target-present lineup, both when speech was whispered at exposure and normal at test and when speech was normal at exposure and at test.193. The study technically examined the effects of typicality rather than distinctiveness per se, but I am including the study here on the assumption that an atypical voice is a distinctive voice.194. JW Mullenix et al, “Typicality Effects on Memory for Voice: Implications for Earwitness Testimony” (2011) 25:1 Appl Cognit Psychol 29 at 31-32. Test subjects were more likely to mistake one typical voice for another typical voice than to mistake an atypical voice for another atypical voice, although they were about equally accurate in correctly recognizing both atypical and typical voices.195. Orchard & Yarmey, supra note 35 at 251-52; Mullenix et al, supra note 194 at 30.196. Orchard & Yarmey, supra note 35 at 252.SHERRIN, EARWITNESS EVIDENCE 853the witness, even though there are in fact many similar voices. Someone hearing a voice with a particular accent, for example, may not realize that the accent is common to a whole group of people. Indeed, in some cases, a witness may perceive a voice to be distinctive because the witness is not aware of the fact that he or she cannot distinguish it from other voices (including the accused’s voice).It is only when a court is told what it is that allegedly distinguishes a voice that the court has any chance of assessing whether the voice truly is more readily identified.197 Unfortunately, even then a court is not in a great position. Earwitnesses are unlikely to identify many vocal characteristics, even when asked.198 I know of no research that has sought to quantify the incidence within a population group of any particular vocal characteristic. And very little research has been undertaken to determine what specific voice characteristics actually facilitate (or impair) identification. Some research suggests that high- and low-pitched voices may be more easily identifiable,199 but the research is neither clear nor consistent.200 There is also the occasional hint in the empirical research that nasality and creakiness do not facilitate identification, but the evidence is 197. See Whalen, supra note 84 at para 18.198. See Yarmey, “Psychology of Speaker Identification,” supra note 9 at 111-12. Yarmey noted that test subjects “describe only four or five voice characteristics of the target in spite of being repeatedly prompted to remember additional characteristics … . Most witnesses tend to describe specific characteristics, that is, pitch, enunciation, and tone of voice.”199. Some courts have accepted testimony that a voice was distinctive at least in part because it was high-pitched. See e.g. Saddleback, supra note 43 at para 33; Sanghera, supra note 68 at para 145.200. Foulkes & Barron, supra note 61 at 194 (among a small number of very familiar listeners, the voices with the highest differences from the group average pitch were on the whole more successfully identified than the voices with pitch closer to the group average); John W Mullenix et al, “Earwitness Memory: Distortions for Voice Pitch and Speaking Rate” (2010) 24:4 Appl Cognit Psychol 513 at 517, 521 (results showed, on average, fewer errors for low-pitched voices, although no statistical analysis was done to see whether the difference was significant, and there were no apparent differences for high-pitched voices); Yarmey, “Earwitness Identification,” supra note 33 at 69 (test subjects significantly more accurate in a showup the higher they rated the pitch of the target’s voice); Steven E Stern et al, “Distortions in the Memory of the Pitch of Speech” (2007) 54:2 Experimental Psychol 148 at 151, 153, 156 (in two of three experiments, error rates lower for low- and high-pitched voices versus moderate-pitched voices, but no statistical analysis was done to see whether any of the differences were significant, and in the third experiment the error rate was highest for high-pitched voices).(2015) 52 OSGOODE HALL LAW JOURNAL854very far from conclusive.201 In the end, a judge can usually do little more than guess whether a particular voice characteristic is either intrinsically helpful or sufficiently unusual to facilitate identification. Unfortunately, experience shows that this can lead judges into error. Cases exist where a judge has deemed a voice to be distinctive at least in part because it was accented202 even though studies show that accent generally impairs accurate identification.203Tempting as it is to use voice distinctiveness as a basis for assessing the reliability of an identification, judges should be wary of doing so. Simply labelling a voice as distinctive is of no help, and we do not know enough about what specific characteristics actually make a voice more readily identifiable. Judges would be wiser to focus on other factors.IV. CONCLUSION AND RECOMMENDATIONSMore frequently than one might imagine, criminal courts in Canada are faced with the task of assessing the reliability of a voice identification made by an earwitness. Empirical research shows that voice identifications can sometimes be accurate, but they can also be highly unreliable, even more so (on average) than eyewitness identifications. Indeed, the research results might even lead to the conclusion that identifications of unfamiliar voices should only rarely be given much weight. At the very least, a very cautious approach should be adopted by the courts. Even identifications of familiar voices can frequently be mistaken.Canadian criminal courts have sometimes displayed impressive sensitivity to the frailties of earwitness testimony, but, not surprisingly, they have not always got it quite right. Empirical research into earwitness evidence offers the most reliable source of information regarding the factors that affect accuracy, yet it is extremely rare to find any reference to the research in the case law.204 This is not a criticism of the courts. It is always difficult for scientific information regarding human behaviour and acuity to reach the courts, and to date, there has been little in the legal literature to bridge that divide. Hopefully, this article will offer some assistance. A high degree of familiarity with the accused’s voice, extended exposure 201. Yarmey, “Earwitness Identification,” supra note 33 at 69 (no correlation found between accuracy and the degree to which a test subject rated a voice as nasal); Foulkes & Barron, supra note 61 at 190 (relative to a small number of other voices, two voices that “displayed a markedly creaky phonation quality” not as well identified by close friends).202. Pabla, supra note 43 at para 26; R v Meyer, 2012 ONCJ 791 at paras 32-33 (available on QL).203. See Part III(H), above.204. For an exception, see Bubar, supra note 127 at para 86.SHERRIN, EARWITNESS EVIDENCE 855to the perpetrator’s voice at the time of the crime, and deliberate attentiveness to its characteristics are all factors that will generally enhance the likelihood that a witness will make a reliable identification. The witness’s confidence in his or her accuracy, the suggestion that the voice was distinctive, and the fact that it was heard over the telephone are probably of no significance. If the perpetrator spoke in a foreign language or with an accent or if the perpetrator disguised his or her voice or spoke with anger or emotionality, it is less likely that an earwitness will make an accurate identification. There are no definitive clues, however. All the factors and all the details must be considered.Going forward, a more cautious approach and a more nuanced understanding of the factors impacting reliability should lessen the dangers associated with earwitness evidence. But the potential weakness of such evidence is so real that we would probably be unwise to leave matters there.205 Further reforms are likely necessary in order to safeguard the reliability of the criminal justice process. I offer some preliminary thoughts to conclude.Some have suggested that expert testimony could both inform courts about the factors affecting the reliability of voice identifications and correct any mistaken beliefs that judges or juries might hold.206 There is some empirical support for the suggestion that expert testimony would be of assistance,207 but I wonder whether 205. One mock jury experiment that looked at some of the factors affecting earwitness reliability found that jury instructions had no effect on the percentage of guilty verdicts across good and poor listening conditions. See Cindy E Laub, Can Earwitness Limitations be Overcome by the Court System? Strategies to Help Mock Jurors Appreciate the Limitations of Earwitness Testimony (PhD Thesis, University of Nebraska, 2010) [unpublished] at 41. See also Gary Edmond, Kristy Martire & Mehera San Roque, “‘Mere Guesswork’: Cross-Lingual Voice Comparisons and the Jury” (2011) 33:3 Sydney L Rev 395 at 421-23. The authors argue that, at least in cross-lingual voice identification cases, jury instructions and warnings are likely to be “vague imprecations,” the primary benefit of which is “to make a trial formally fair” (ibid at 423). I am a little more optimistic, especially for judge-alone trials.206. Cindy E Laub, Lindsey E Wylie & Brian H Bornstein, “Can the Courts Tell an Ear from an Eye? Legal Approaches to Voice Identification Evidence” (2013) 37 Law & Psychol Rev 119 at 152-54; Eeva K Komulainen, “Subjective Voice Identification: The Literal Meaning of ‘Talking Yourself Behind Bars’” (1988) 26:3 Alta L Rev 521 at 544.207. Laub, supra note 205 at 41.(2015) 52 OSGOODE HALL LAW JOURNAL856it would be of much value most of the time. Its admissibility is questionable,208 it is expensive for both the litigants and the trial process, and the available experts are few and far between. How often will the necessary testimony be available, affordable, and admissible in the average criminal case?Another option would be to exclude from evidence voice identifications of dubious reliability. Counsel have occasionally sought to have courts do so in the past.209 However, their success rate has been very low,210 and the idea of excluding evidence based on reliability concerns runs counter to the general preference in the law for such matters to go to weight.211 The courts would probably only end up excluding the most glaringly weak identifications, much as they have done in the eyewitness identification context.212The most fruitful option would be to encourage the police to administer pre-trial voice lineups,213 at least in cases where earwitness testimony will play any significant role and the witness is not a very familiar listener. The police have occasionally conducted voice lineups in the past,214 and the courts have very occasionally discounted earwitness evidence that was not subjected to a lineup test.215 Lineups have long been accepted as a valuable tool in the eyewitness context.208. See Badgerow, (2010), supra note 78 at paras 6-9. In this case, proposed expert evidence on “the inherent frailties in ear witness identification” and “the methods used by the police … when interviewing potential voice identification witnesses” was deemed unnecessary and inadmissible (ibid at para 6). Expert testimony on the frailties of eyewitness evidence has frequently been excluded. See e.g. R v Frimpong, 2013 ONCA 243, 305 OAC 93; R v Woodard, 2009 MBCA 42, 240 Man R (2d) 24. However, Komulainen reports that an expert was allowed to testify at the preliminary hearing in the unreported early 1980s Alberta case of R v Laberge. See Komulainen, supra note 206 at 544.209. See e.g. Badgerow, (2010), supra note 78.210. Thus far, courts have only been willing to exclude voice identification evidence on the rare occasion when the witness had no real exposure to the suspect’s voice before or after the crime. See e.g. Portillo, supra note 4 at paras 40-42.211. See R v Buric (1996), 28 OR (3d) 737, 106 CCC (3d) 97 (CA), aff’d [1997] 1 SCR 535, 32 OR (3d) 320.212. For the most part, only in-dock identifications have been excluded. See e.g. R v Holmes (2002), 62 OR (3d) 146 at paras 39-41, 169 CCC (3d) 344 (CA); R v Tebo (2003), 175 CCC (3d) 116 at 124-25, 172 OAC 148 (CA).213. This could be done through court rulings stressing the desirability of voice lineups or through internal police policies directing officers to perform lineups in earwitness as well as eyewitness cases.214. See e.g. Nichols, supra note 171; Komulainen, supra note 206 at 532. Komulainen reported in 1988 that more than 150 voice lineups had been held across Canada.215. See Whalen, supra note 84 at para 16.SHERRIN, EARWITNESS EVIDENCE 857The police could borrow from the extensive study of eyewitness lineups for guidelines on how to construct and administer earwitness lineups.216 Several recommendations of eyewitness researchers would seem obviously transferable.217 The lineup should be double-blind, meaning that neither the witness nor the lineup administrator should know which voice sample belongs to the suspect. The entire lineup identification process should be accurately recorded, preferably on videotape. The witness should be warned that the perpetrator might not be in the lineup, that it is equally important to clear the innocent as to identify the guilty, and that one of the available options is to select no one. If the witness makes a selection, the witness’s level of confidence should be ascertained and noted.218Earwitness researchers have also offered additional recommendations specific to the voice identification context.219 The lineup voices should be tape recorded and be a minimum of sixty seconds (or two hundred words) in duration in order to offer a reasonable representation of the speaker’s phonetic and idiosyncratic speech characteristics. To avoid deliberate distortion by a guilty suspect, the lineup should not include words or phrases known to have been spoken by the perpetrator during the crime.220 However, the lineup should ideally include content using the same tone of speech as used by the perpetrator during the commission of the crime. The witness should be allowed to listen to the tapes in a quiet environment, preferably with headphones. The witness should be questioned about his or her hearing ability. A particular challenge for voice lineups may be finding appropriate foils. The foils should match any descriptions of the perpetrator’s voice given by witnesses and should broadly match the suspect’s voice in terms of accent, pitch, and speech 216. For a good overview of this research, see Gary L Wells, “Eyewitness Identification: Systemic Reforms” (2006) 2006:2 Wis L Rev 615.217. See Yarmey, “Psychology of Speaker Identification,” supra note 9 at 126.218. Much attention has been paid in the eyewitness context to the choice between simultaneous and sequential lineups. In the earwitness context, the choice does not arise; all voice lineups are sequential. It has been recommended, however, that the witness not be told how many voices he or she will hear and that the witness only be allowed to go through the lineup once. See APA Broeders & AG van Amelsvoort, “A Practical Approach to Forensic Earwitness Identification: Constructing a Voice Line-Up” (2001) 47 Probs Forensic Sci 237 at 239.219. See e.g. Yarmey, “Psychology of Speaker Identification,” supra note 9 at 124-28; Harry Hollien, “On Earwitness Lineups” (2012) 4:1 Investigative Sci J 1; APA Broeders & AG van Amelsvoort, “Lineup Construction for Forensic Earwitness Identification: A Practical Approach” in John J Ohala et al, eds, Proceedings of the XIVth International Congress of Phonetic Sciences (Berkeley: Regents of the University of California, 1999) 1373.220. Yarmey found that using phrases spoken during the crime did not improve identification accuracy. Yarmey, “Earwitness Descriptions,” supra note 28 at 119-20.(2015) 52 OSGOODE HALL LAW JOURNAL858rate as well as the suspect in terms of age, gender, ethnicity, and, if possible, socioeconomic background. Finding such individuals may not be easy and may even require the assistance of linguistics experts.221 Five or six foils are sufficient. They should not be allowed to hear the suspect’s voice prior to tape recording in order to prevent them from mimicking or differentiating their voices from the suspect’s, consciously or not.Another challenge for voice lineups may be to obtain an adequate voice sample from the suspect. The police can request a sample for purposes of holding a lineup, but a suspect has the right to remain silent and may not speak, or may speak unnaturally, knowing that the recording will be used for an identification procedure. An alternative source might be a portion of a recorded police interview, but that comes with potential dangers.222 The words spoken may signal to the listener that the speaker is under suspicion, and foils uttering the same words may find it difficult to replicate the stress and emotionality felt and conveyed by the suspect; one study, using samples from a real police interview, found that it was easy for random listeners to pick out the suspect in the lineup.223 Ideally, samples of spontaneous, casual speech would be used.224 Police who have intercepted the suspect’s communications during the investigation of the offence would probably have easiest access to such samples. Police in other cases could try to conduct a relaxed, non-accusatorial interview at the police station that is clearly separated 221. In the United Kingdom, the Home Office has suggested that a “suitable source of such material may be other police recorded interview tapes from unconnected cases.” See UK, Home Office, Advice on the Use of Voice Identification Parades (Correspondence) (London: Home Office, 2003), online: .222. A guilty suspect, aware that voice identification will be critical, might refuse to speak during a recorded interview, but it is possible that the police could surreptitiously record the interview. The courts are divided over whether a suspect has to be informed that he or she is being taped. See e.g. R v Hanuse, 185 AR 384, [1996] AJ No 615 (QL) (QB) (not informing undermines voluntariness); R v Stoddart (2005), 72 WCB (2d) 365, [2005] OJ No 2207 (QL) at para 45 (Sup Ct) (suspects must be informed they are being videotaped); R v Dickens (1999), 242 AR 148, 41 WCB (2d) 495 (QB) (not informing does not undermine voluntariness); R v MOJP, 2001 ABPC 170, 301 AR 362 (not informing does not undermine voluntariness even for a young person).223. Ann Stuart Laubstein, “Problems of Voice Line-ups” (1997) 4:2 Forensic Linguistics 262. Yarmey recommends against using a speech sample from a police interview. See Yarmey, “Psychology of Speaker Identification,” supra note 9 at 127.224. See Yarmey, “Psychology of Speaker Identification,” ibid at 124-25. Yarmey and Hollien also recommend that the voice lineup include samples of speech read from a selected text. See ibid at 125; Hollien, supra note 219 at 8. That would seem impossible unless the suspect cooperates.SHERRIN, EARWITNESS EVIDENCE 859from any interrogation. More realistically, they may have to obtain judicial authorization to wiretap the accused following arrest.225 I do not minimize the challenges involved in creating voice lineups, and I do not suggest that change can happen overnight.226 I also acknowledge that, to date, most courts have been largely indifferent to the lack of a lineup.227 But the general frailty of voice identification for all but the most familiar listeners is so apparent that we would be foolish not to try to seek out the information that lineups can provide. Until we do, perhaps the best advice for a court assessing proof of identity is to focus as much as possible on evidence in the case other than earwitness testimony.225. Komulainen suggests consent wiretaps were used in the three cases she discusses. See Komulainen, supra note 206 at 536.226. Komulainen estimated that preparation of a voice lineup takes approximately forty to sixty hours. See ibid at 532. That estimate may be premised on the extensive involvement of a linguistics expert. Hopefully, the need for their participation would be minimized by the development of detailed and scientifically-informed guidelines.227. See e.g. R v Hammerstrom, 2014 BCSC 94 at paras 24-25 (available on QL) (dismissing the argument that the lack of a voice lineup must contribute to a reasonable doubt); Bubar, supra note 127 at para 78 (not commenting on the defence submission that “no audio line-up was done”); Anderson, supra note 103. Conversely, Pinch indicates that one factor to consider is whether a “voice identification parade” was conducted. See Pinch, supra note 56 at para 75.(2015) 52 OSGOODE HALL LAW JOURNAL860V. APPENDIX A: AMERICAN WRONGFUL CONVICTION CASES BASED PARTLY ON VOICE IDENTIFICATION EVIDENCEName of accused Source(s) indicating use of voice identification evidenceSource(s) indicating accused’s innocenceChristopher Coleman People v Coleman, 996 NE 2d 617 (Ill Sup Court 2013).People v Coleman, 996 NE 2d 617 (Ill Sup Court 2013); Radley Balko, “This week in innocence,” The Washington Post (14 March 2014) online: .Sedrick Courtney The Innocence Project, “Sedrick Courtney,” online: .Bill Braun, “Wrongly imprisoned Tulsa man declared innocent, eligible to seek compensa-tion from state,” Tulsa World (24 July 2013) online: .Lesly Jean State of North Carolina v Lesly Jean, 311 SE 2d 266 (NC Sup Ct 1984).Estes Thompson, “Easley Pardons Ex-Marine Falsely Accused of Rape,” The Charlotte Observer (10 February 2001) 2B.Kerry Kotler Michael Slackman, “Parting Shots Rape Indictment Dismissed, Kotler Raps DA Delay,” Newsday (15 December 1992) 3.Michael Slackman, “Parting Shots Rape Indictment Dismissed, Kotler Raps DA Delay,” Newsday (15 December 1992) 3.Alprentiss Nash Maurice Possley, “Alprentiss Nash” The National Registry of Exonerations, online: University of Michigan Law School .Maudlyne Ihejirika, “State’s attorney vacates conviction in 1995 murder,” Chicago Sun-Times (31 August 2012) 10.SHERRIN, EARWITNESS EVIDENCE 861Name of accused Source(s) indicating use of voice identification evidenceSource(s) indicating accused’s innocenceKirk Odom Maurice Possley, “Kirk Odom,” The National Registry of Exonera-tions, online: University of Michigan Law School .Spencer Hsu, “After DNA retesting, Kirk Odom exonerated,” The Washington Post (14 July 2012) B06.David Sheppard The Innocence Project, “David Sheppard” online: .The Innocence Project, “David Sheppard” online: .Steven Snipes Sarah Ovaska, “After 9 years, a pardon to prize: Falsely accused man wins 1 of only 4 issued by Easley since 2001,” The Charlotte Observer (3 February 2007) 1B.Sarah Ovaska, “After 9 years, a pardon to prize: Falsely accused man wins 1 of only 4 issued by Easley since 2001,” The Charlotte Observer (3 February 2007) 1B.Keith Turner Bruce Tomaso, “Wrongfully convicted, he’s trying to live right: Cleared by DNA test, Dallas man still weighed down by rape he didn’t commit,” The Dallas Morning News (12 April 2007), online: .Bruce Tomaso, “Wrongfully convicted, he’s trying to live right: Cleared by DNA test, Dallas man still weighed down by rape he didn’t commit,” The Dallas Morning News (12 April 2007), online: .Eduardo Velasquez, aka Angel HernandezThe Innocence Project, “Eduardo Velasquez” online: .“Briefing,” Providence Journal Bulletin (17 August 2001) 1 (WL Can).James Douglas Waller Jeff Carlton, “Man exonerated of rape in Texas,” The Oakland Tribune (20 January 2007).Jennifer Emily, “Dallas man pardoned in wrongful rape conviction: DNA exonerated 50-year-old who was jailed for 11 years,” The Dallas Morning News (10 March 2007) 20A.(2015) 52 OSGOODE HALL LAW JOURNAL862Name of accused Source(s) indicating use of voice identification evidenceSource(s) indicating accused’s innocenceCalvin Washington Maurice Possley, “Calvin E Washington” The National Registry of Exonerations, online: University of Michigan Law School .“Pardon given to inmate cleared by DNA evidence,” Houston Chronicle (11 October 2001) A32.Cathy Watkins Maurice Possley, “Cathy Watkins” The National Registry of Exonerations, online: University of Michigan Law School .Daniel Beekman, “17-year jail nightmare finally ends,” New York Daily News (14 December 2012) 20.Andrew Whitley Jill King Greenwood, “DNA reversal leaves case at square one,” Pittsburgh Tribune Review (2 October 2006).Jill King Greenwood, “DNA reversal leaves case at square one,” Pittsburgh Tribune Review (2 October 2006).Michael Williams Maurice Possley, “Jailed 24 years, freed by DNA: Innocence Project key to exoneration in Louisiana rape case,” Chicago Tribune (7 March 2005) 1.Maurice Possley, “Jailed 24 years, freed by DNA: Innocence Project key to exoneration in Louisiana rape case,” Chicago Tribune (7 March 2005) 1.Calvin Willis Innocence Project, “Calvin Willis” online: University of Michigan Law School .Henry Weinstein, “DNA Frees Man Jailed for 22 Years,” Los Angeles Times (20 September 2003) 11.Collin Woodley Robin Topping, “Deliverance Day: After 9 years in jail on wrongful conviction, LIer freed” Newsday (20 March 1998) A05.Robin Topping, “Deliverance Day: After 9 years in jail on wrongful conviction, LIer freed” Newsday (20 March 1998) A05.

Understanding earwitness testimony

Abstract

Extracted data

Understanding earwitness testimony

Abstract

Extracted data

Related items

Related items