Cries For Help: Voice Recognition Analysis In Trayvon Martin Case

This past week has seen tragedies throughout the nation and the world. In Florida, George Zimmerman’s defense may rest on the faint cries for help caught on a 911 recording from a nearby home. Zimmerman claims that he was crying for help after being attacked by 17-year-old Trayvon Martin. Trayvon’s mother has heard the tapes and says she’s sure that the voice crying for help is her son’s. Zimmerman claims that he shot Trayvon in self defense, and under Florida’s Stand Your Ground law, he would be found not guilty. If Zimmerman hopes to use the cries for help as part of his case, the voice will need to be authenticated as his.

Voice authentication is best done with as many potential matches as possible saying the same words. In this case, where the voice could belong to either Trayvon or Zimmerman, identification is limited to matching the voice against Zimmerman and against other random samples. While there is some technology that tries to match voices based on frequencies in the speaker’s voice, this is best accomplished with a large pool of material to work with.

Assume for the moment that the voice on the tape is Zimmerman. Voice authentication is done in two primary ways: an audio forensic specialist listens to voice samples and attempts to find a match or a digital machine compares the biometrics of the voice. Zimmerman would supply a recording of himself, and that would be matched against random samples of other voices saying the same words. The extra voices would be in the mix to help the analyst avoid making a match due to subconscious bias. Zimmerman could also supply a longer sample for voiceprinting, but even in this best case scenario it may be hard to make the match.

Now, assume that the voice was Trayvon’s. Audio forensics are best matched when the same voice says the same word under the same circumstances on a good recording; background voices on a 911 cellphone call are pretty low quality. Even if Trayvon’s family has recordings of him speaking and even if they have recordings of him saying the word help, there’s a good chance that the samples would not sound alike.

The big question is whether audio forensics are admissible in a court of law. Voice matching claims to be as accurate as fingerprinting when under the right circumstances*. Fingerprint expert testimony is admissible in a court of law, but like voice recognition, the standards can greatly differ. Fingerprints samples are often not perfect, like voice samples, and analysts look for points that match. Until recently, the UK required 16 points that match, but the U.S. doesn’t have that high a standard. Factor in that every analyst may see a match differently, as in this Washington Post story from 2004:

Three highly skilled FBI fingerprint experts declared this year that Oregon lawyer Brandon Mayfield’s fingerprint matched a partial print found on a bag in Madrid that contained explosive detonators. U.S. officials called it “absolutely incontrovertible” and a “bingo match.” Mayfield was promptly taken into custody as a material witness. Last week the FBI admitted that it goofed; the print actually belongs to Ouhnane Daoud, an Algerian. … In Mayfield’s case, the FBI located 15 points of similarity, places where the particular ridge characteristics of two prints “matched.” Even the Spanish authorities, though doubtful about the match, identified eight points of similarity. While many American examiners no longer exclusively count points, experts have declared positive fingerprint matches in court after finding even fewer than eight points.

It remains to be seen whether audio forensic experts believe that they can conclusively match Zimmerman’s voice to the recordings. The state and the Department of Justice may conduct their own separate analyses, and they may not come to the same conclusion. The technology has progressed, but it’s far from perfect.

* Link added to article quoting Steve Cain, president of Applied Forensic Technologies International of Chicago and Lake Geneva, Wisconsin, on comparing the accuracy of voice-printing to finger-printing.