The big headline for the past several days has been that an audio expert hired by The Orlando Sentinel, Thomas J. Owen, had concluded that the person screaming for help on the 911 tape when a neighbor called in was not George Zimmerman because voice biometric analysis of the scream could only find a 48% match, and a much higher match (I’ve seen reference to 60% and 90% in different reports) would be needed.
The analysis was done comparing the Zimmerman 911 phone call voice with the voice on the separate 911 call from the neighbor.
Like the original ABC video tape supposedly showing Zimmerman uninjured, Owen’s analysis has caused a firestorm of pontification that Zimmerman’s self-defense argument was a lie and that Zimmerman executed Martin.
Tom Maguire has done an exellent job running down the various uncertainties with this type of voice analysis, as has Jeralyn at Talk Left, a criminal defense attorney who has used Owen as an expert in the past. The point of their numerous posts is not that Owen is wrong, but that caution must be used given the nature of voice biometric analysis which is not universally accepted and which may be subject to challenge in court as not sufficiently reliable as to be admitted in evidence.
Equally important, as Tom points out again today, such analysis normally is done using numerous similar exemplars from the subject, so that a subject saying “help” could be compare to the person on the tape saying “help.” In this case so far, however, there were no such exemplars, as an exemplar of Zimmerman’s voice screaming help was not taken so as to be compared to the 911 recording.
I’ll add just one piece of information to this mix not to prejudge the outcome either of the case or Owen’s analysis, but again to urge media caution.
In Red Ant, L.L.C v Sony Music Entertainment, Inc. , et al. (Docket No. 97 Civ. 4917), a 1997 case in federal court in the Southern District of New York (Manhattan), Owen was hired as an expert witness by the defendants to identify a voice on a tape. The court ordered the witness, Melvin Britt, to provide Owen with voice exemplars so that Owen could compare Britt speaking certain words with the voice on the tape speaking the same words.
Owen believed that Britt deliberately was not giving accurate voice exemplars so as to stymie Owen’s analysis, and a motion for sanctions was made.
The court rendered a Decision and Order on the motion for sanctions on May 15, 1998. I obtained a copy of the Order via Westlaw. Unfortunately, there is no publicly available copy that I could find (not even on the PACER system, which is not surprising since old docket documents often have not been scanned, although the docket sheet is on the system). Since Westlaw copyrights the way in which it presents cases (but it cannot copyright the text of the case) I can’t post the Westlaw copy. For those of you with access to Westlaw, the cite is 1998 WL 249195 (S.D.N.Y.).
In the Order, the Judge laid out the procedures Owen demanded so that he could perform a proper analysis, including the requirement of at least six similar exemplars. Here’s the pertinent portion (emphasis mine):
EATON, Magistrate J.
A. Defendants’ motion for sanctions against Mr. Britt for willful failure to give useable voice exemplars
Pursuant to my March 24 order, Melvin C. Britt, II (also known as Melieck Britt) appeared in Los Angeles on April 1st and gave voice exemplars over the telephone to Thomas J. Owen, an expert in voice identification retained by the defendants.
2. The session wound up having five parts, and they were all recorded. In the first part, Mr. Britt gave exemplars for about 32 minutes, and did not come close to doing what my order required—“repeat the word in the same manner ( i.e., same emphasis and same melody pattern and same rate of speech) as requested by the expert.” In the second part, the attorneys spoke with me by telephone and I told Mr. Britt that (a) his exemplars were coming nowhere near the melody pattern, (b) I did not think it was very difficult to do what the expert was asking and (c) he should concentrate on trying to mimic the expert’s melody pattern and speed. In the third part, lasting about 5 minutes, Mr. Britt again gave exemplars, on two phrases. In the fourth part, the attorneys again spoke with me, and I said that (a) he was not even close on the second phrase, (b) he gave 3 or 4 good exemplars on the first phrase (in the course of more than 30 repetitions) and (c) I was surprised that Mr. Owen claimed that none of them were useable. I told Mr. Britt that I had heard about his speech impediment (I would not call it a stutter, but a mild tendency to repeat a word occasionally) but that I did not think it was all that hard for him to mimic Mr. Owen. (It seems to be undisputed that Mr. Britt is a singing coach.) Once again, I directed him to (a) listen very closely to what Mr. Owen was saying, and (b) deliberately try to copy his pitch and speed for another half hour. In the fifth and final part of the session, lasting about 34 minutes, Mr. Britt’s performance was substantially improved for the last 26 minutes.
3. On April 21 the defendants moved for sanctions and also for an order directing Mr. Britt to give exemplars in a second session in my presence. Mr. Owen’s April 20 declaration says:
… virtually each and every time I asked Mr. Britt to repeat a particular phrase, Mr. Britt would either speak in an obvious wooden monotone, repeat declarative statements in a question form, alter the beat or pitch of the phrases, vary the speed of the words, change the word coupling used by me or over-articulate particular linguistically significant words.
I have listened to the tape of the session, and Mr. Owen’s statement is quite accurate as to the first 32 minutes. But Mr. Britt’s performance was substantially improved for the last 26 minutes (from 6:56 to 7:22 p.m. New York time).
4. Mr. Owen further writes:
In order to complete a proper voice comparison, I regularly require six useable samples of each of the phrases I use for comparison purposes. Here, however, given Mr. Britt’s obfuscation, I do not have six useable samples of even one such phrase, despite the fact that I attempted to obtain 600 exemplars from Mr. Britt for approximately 18 phrases.
But for my present purposes (deciding whether Mr. Britt willfully disobeyed my orders) I must rely primarily on my non-expert view as to whether he provided exemplars that a reasonable person would consider to be useable. Mr. Britt’s performance was certainly not as natural as his speaking voice at his video deposition. But, in my non-expert view, Mr. Britt provided at least one useable sample in 14 of the 17 phrases tested in the last 26 minutes, namely:
(1) “If it don’t drop in March, they, they breach the contract.”
Approx. 7 repetitions, 2 useable (# 3 and # 6, approx.)
(2) “They have money.”
Approx. 3 repetitions, 2 useable (# 2 and # 3, approx.)
(3) “These are white folks, they have money.”
Approx. 4 repetitions, 2 useable (the last two)
(4) “Troy didn’t write nothing but …”
Approx. 10 repetitions, 1 useable (# 8, approx.)
(5) “… two lyrics on the first verse.”
Approx. 5 repetitions, 2 useable (the first two)
(6) “I came up with the hook.”
Approx. 18 repetitions (10 + 8 later on), 5 useable (# 3, # 6, # 9, # 10 and # 17, approx.)
(7) “So Troy can player-hate nothing.”
Approx. 9 repetitions (5 + 4 later on), 0 useable
(8) “I came up with the hook of Erotik City.”
Approx. 17 repetitions (9 + 8 later on), 3 useable (# 6, # 8 and # 15, approx.)
(9) “He’s not even hooked up to a major yet.”
Approx. 15 repetitions (9 + 6 later on), 4 useable (# 4, # 5, # 13 and # 15, approx.)
(10) “You can’t fulfill none of your obligations …”
Approx 10 repetitions, 0 useable
(11) “… until you get hooked up with a major.”
Approx 5 repetitions, 2 useable (the last two)
(12) “Trying to get you rich.”
Approx. 10 repetitions, 1 useable (# 6, approx.) [Note: 3 or 4 were useable when this phrase was tested in an earlier part of the session.]
(13) “I got ways you can get outta that shit, bro.”
Approx 10 repetitions, 2 useable (# 4 and # 10, approx.)
(14) “That’s the easiest part, that’s the ea-siest part.”
Approx. 19 repetitions, 4 useable (# 8, # 9, # 17 and # 19, approx.)
(15) “I guarantee that this shit is on, man.”
Approx. 8 repetitions, 5 useable (the last five)
(16) “It’s not Red Ant.”
Approx. 20 repetitions, 6 useable (# 5, # 13, # 14, # 15, # 16 and # 19, approx.)
(17) “Troy didn’t write nothing but ah, ah, ah two lyrics on the first verse.”
Approx. 10 repetitions, 0 useable, but see phrases (4) and (5) above.
5. To summarize, in my non-expert view, Mr. Britt provided 4 useable samples on three phrases (including “Trying to get you rich”), plus 5 useable samples on two phrases, plus 6 useable samples on one phrase. In his earlier declaration (dated March 13) Mr. Owen said at ¶ 23 that the International Association for Identification generally recommends that the text be read or recited at least 3 times, and the American Board of Recorded Evidence generally recommends 3 to 6 repetitions. I realize that an expert may need more repetitions when the subject performs poorly. Nevertheless, weighing all of the circumstances, I decline to order sanctions and I decline to order Mr. Britt to give more voice exemplars at this time. I do note, however, that plaintiff’s other arguments seem to me to be unfounded. Moreover, I disagree with the statement of plaintiff’s expert that “it is all but impossible for an individual to disguise his voice or to speak in a consistently unnatural manner for the amount of time Mr. Britt spent in the voice exemplar session.”
6. Mr. Owen will have to make do with the April 1st exemplars, plus any other samples of Mr. Britt’s voice which may be collected. I note that I was given only 5 minutes of excerpts of the videotaped deposition of Mr. Britt, and even those brief excepts have captured him speaking several of the words which appear in the crucial portions of the disputed tape.
This 1998 case does not prove that Owen is wrong in his current 911 tape analysis. Perhaps there are new techniques developed, but based on the information in Tom’s posts, the industry still seems to adhere to the standard practice of comparing numerous voice exemplars to the subject recording.
Caution is urged. The media in particular needs to think and investigate and put new drips of information in context rather than jumping to conclusions.