Become a fan of Slashdot on Facebook

 



Forgot your password?
typodupeerror
Crime Your Rights Online

Forensic Experts Say Screams Were Not Zimmerman's 1046

Posted by samzenpus
from the hearing-is-believing dept.
Hugh Pickens writes "As the Trayvon Martin controversy splinters into a debate about self-defense, a central question remains: Who was heard crying for help on a 911 call in the moments before the teen was shot? Now the Orlando Sentinel reports that Tom Owen, a leading expert in the field of forensic voice identification sought to answer that question by analyzing the recordings. His result: It was not George Zimmerman who called for help. Owen, forensic consultant for Owen Forensic Services LLC and chair emeritus for the American Board of Recorded Evidence, used voice identification software to rule out Zimmerman. Another expert contacted by the Sentinel, utilizing different techniques, came to the same conclusion. Owen used software called Easy Voice Biometrics to compare Zimmerman's voice to the 911 call screams. 'I took all of the screams and put those together, and cut out everything else,' says Owen. The software compared that audio to Zimmerman's voice and returned a 48 percent match. Owen says to reach a positive match with audio of this quality, he'd expect higher than 90 percent. Owen cannot confirm the voice as Trayvon's, because he didn't have a sample of the teen's voice to compare however 'you can say with reasonable scientific certainty that it's not Zimmerman.'"
This discussion has been archived. No new comments can be posted.

Forensic Experts Say Screams Were Not Zimmerman's

Comments Filter:
  • by AdmiralXyz (1378985) on Sunday April 01, 2012 @01:28PM (#39542165)
    Zimmerman's claim of being badly beaten up before he shot the kid doesn't hold up either: there's some footage taken at the police station [go.com] the night of his arrest, and he looks totally unharmed. No cuts to the back of the head, no broken nose, nothing. Guy's story has more holes than... eh, I'm on my eighteenth hour without sleep due to a project and can't come up with an apt metaphor, but something with a lot of holes.
    • by Anonymous Coward on Sunday April 01, 2012 @01:31PM (#39542193)

      More holes than Trayvon?

    • by tomhath (637240) on Sunday April 01, 2012 @03:17PM (#39542981)
      Note that the video was taken well *after* EMTs had treated him at the scene. I wouldn't expect video of that quality to show anything at that point. While it would be interesting to hear what the EMTs saw, if they say he appeared to have been beaten you won't believe them.
      • by Mitreya (579078) <mitreya@gmail.LAPLACEcom minus math_god> on Sunday April 01, 2012 @04:10PM (#39543313)

        While it would be interesting to hear what the EMTs saw, if they say he appeared to have been beaten you won't believe them.

        I would totally believe the EMTs if they testified to that under oath in court. Of course I wouldn't really believe Zimmerman's claims, since he obviously has a bit of a bias
        Problem is, he's not had to defend himself in court, to bring all that exonerating evidence. Police just took his word for it.

      • by frank249 (100528) on Sunday April 01, 2012 @04:58PM (#39543677)

        If the EMTs treated Zimmerman, then they would have put a dressing over any break in the skin to prevent infection. The back of his head does not even have a band aid on it. If he had a broken nose there would have been blood on his shirt and jacket and you would likely see nose plugs to stop the bleeding. There also would have been swelling. No evidence of this either.

  • by Jazari (2006634) on Sunday April 01, 2012 @01:38PM (#39542249)
    Has this software been subjected to repeated double-blind testing? Otherwise, there's no point in reporting this.
    • by Lillesvin (797939) on Monday April 02, 2012 @08:42AM (#39548137) Homepage

      I'm not familiar with this particular piece of software, but I am however very familiar with the subject of forensic phonetics. The usual MO isn't like that of Hollywood, where the software does it all for you and simply gives you a percentage chance that it's X. (Judging from the website of Easy Voice Biometrics, however, it seems that that is what they're going for, which is an incredibly bad idea.) Usually the forensic phonetician will use software like Praat [praat.org] for working with the speech samples, but all conclusions will be his/her own.

      Oh, and I really have a problem with the word Biometrics in the name of the software, because the human voice isn't very dependent on biometric factors --- it is extremely easy to manipulate. I can't imagine any program being capable of reliably identifying and eliminating voice disguises as they claim EVB does. (Source: Working on my MA thesis in forensic phonetics.)

  • by Stormy Dragon (800799) on Sunday April 01, 2012 @01:39PM (#39542255) Homepage

    While I'm personally of the opinion that Zimmerman is guilty of manslaughter, I've also seen too many cases of "forensic science" of dubious validity being allowed in trials (for example Steven Hayne in Mississippi help convict dozens of people on the base of 'bite mark identification' techniques that are widely considered fraudulent). Can anyone point to any independent blind trials to demonstrate 1) that the metrics used by this program actually are invariant for a particular individual and 2) are sufficiently unique that they can be used to reliably distinguish two individuals?

    • by gstrickler (920733) on Sunday April 01, 2012 @01:53PM (#39542345)

      Yes, but there are so many holes in his story that it's beyond belief. There is no question of who shot Trayvon, Zimmerman has admitted that. He's claiming self-defense, and his story doesn't hold up to any scrutiny. He should be (should have already been) arrested and charged, and let the legal system handle it, not the media and public outcry. If they don't have enough evidence, then the grand jury can refuse to indict. If they do, then he goes to trial or makes a plea.

      • by sjames (1099) on Sunday April 01, 2012 @02:29PM (#39542619) Homepage

        That's exactly it. We know for a fact that he began from a position of safety (since he was on the phone w/ 911). Stand your ground does not mean charge into danger. We know the victim had no weapon. That makes the threat to life or grievous bodily harm seem unlikely. It really does seem like more than enough to justify deeper investigation, but instead the DA has already delivered a poor excuse for no charges which generally indicates there will be none..

    • by Dan East (318230) on Sunday April 01, 2012 @02:17PM (#39542525) Homepage Journal

      I think there's a significant problem in the way this software is being used in this case. It is one thing to say that, based on samples of an individual's voice, that there is X probability that some other voice sample could be from the same person. However, there isn't any way the software can do the opposite and prove a negative. For example, you can take hours of recordings of my voice from phone conversations and train the software. Now, if I make my voice sound like Donald Duck, or if I sing in falsetto, which are two types of sounds that would not be in the sampling of my normal voice, what probability would the software return that the silly voice matches my normal voice? It would be very low, at is should be.

      In this case we have a sample of someone's NOT NORMAL VOICE. They are screaming and yelling. What voice samples of Zimmerman's were used to train the software? Was it recordings of him screaming and yelling? If not then of course the software will return a low probability. Actually, I'm surprised it returned as high of a probability as it did.

      Now, as a side note, I think the news is really trying to inflame this whole deal. I think Zimmerman is likely in the wrong here, however, the media is running rampant with speculation. A good example are the video clips of Zimmerman at the police station, and them (news media) saying it doesn't look like his nose is broken, or that there doesn't appear to be blood all over him. That sort of thing is completely impossible to determine from the crappy, low resolution security video. Again, yes, if there was that much blood and it could be seen in the video, then a positive could be "proven", but just because it doesn't show up in the video doesn't mean he has significant injuries or a broken nose, etc.

      • by Anonymous Coward on Sunday April 01, 2012 @02:30PM (#39542629)

        This reminds me of the infamous plane identification software. Its accuracy got up to 100%. They were so happy that they managed to do it. Put it out in the field and there were all sorts of false positives and false negatives. Turns out that the software got so good at guessing because with the test photos, all the good planes were shot against bright blue sky and the bad planes were shot against dark sky. The software made all its predictions based on the sky color.

      • by gstrickler (920733) on Sunday April 01, 2012 @02:32PM (#39542643)

        Voice analysis is a well established and studied technique. Unless you're particularly practiced and adept at disguising your voice, there are certain characteristics introduced by the physics of your vocal cords, throat, mouth, and nasal passages that are pretty consistent and identifiable, even with most people doing "impersonations" that sound like some other person. So, yes, it's reasonable to rule out that it was Zimmerman, in fact, it's more reliable at excluding people than at matching them. As Mr Owen points out, this recording has low background noise levels, and it's got a notable amount of Zimmerman's speech recorded in addition to the screams for help, so there is no issue introduced by aging, colds, etc. It's a nearly ideal case for matching.

        • by YesIAmAScript (886271) on Sunday April 01, 2012 @02:45PM (#39542737)

          'there are certain characteristics introduced by the physics of your vocal cords, throat, mouth, and nasal passages that are pretty consistent and identifiable'

          Yeah, but you have no way of knowing this software uses those.

          We actually come back to an RMS argument. This software is a black box. How do you know what it is measuring?

        • by buybuydandavis (644487) on Sunday April 01, 2012 @03:33PM (#39543093)

          Nice to see someone actually talking about the content of the article, instead of arguing about their conflicting certainties over fact.

          But I disagree with your conclusions. Yes, theoretically you can do speaker identification. It is a very hard problem, however. In this case, we're talking about someone screaming in the distance in the background of a telephone call. Unless he did a lot of training with screamed samples, we should be skeptical of his results in conditions off his training set.

          In particular, since there are really only two possible speakers, I'd at least wait until the same analysis is performed on the young man and the resulting match rate compared before jumping to conclusions. The fact that he didn't do that makes me think this is a guy with a company looking for some publicity, and not someone diligently trying to find the truth.

    • by girlintraining (1395911) on Sunday April 01, 2012 @02:40PM (#39542695)

      d 2) are sufficiently unique that they can be used to reliably distinguish two individuals?

      Short answer: No.

      Long answer:
      Consider this: There are any number of comedians that are capable of mimicking the vocal characteristics of famous people to the point that unless you are directly observing the person, you would have no way of saying that say, Bill Clinton, wasn't standing in front of you. Consider as well the wide dynamic range of the human voice -- From Opera to pop music singer, we have the ability to make a stunningly wide range of sounds; and reliably reproduce them. Now consider that this recording was taken of a telephone call. First, telephone calls are stripped of most frequencies to focus only on those used during normal conversation. Some women are nearly unintelligible on a phone without deepening their voice because their natural range falls outside what the phone will transmit. Now consider that this signal was further degraded because it was a cell phone: Which use very small and notoriously distortion-prone microphones to capture audio. Now I'll throw in my personal and subjective experience here; I've worked in a call center and have done QA on a number of calls as well as taken a number myself; I would say roughly 15% of women can be misidentified as men, and vice versa. That is how much of the signal is lost or distorted. When people can't even positively identify the gender of a caller to better than 85%, well... food for thought.

      The idea of differentiating from two people with similar vocal characteristics over the phone is patently absurd. It may be sufficient to identify a person from, say, 30 other people that can be identified using, say, a security camera... but it is nowhere near as forensically sound as fingerprints. That is to say, by itself, it's worthless -- I wouldn't count on it to establish more than reasonable suspicion, let alone beyond reasonable doubt.

    • by mosb1000 (710161) <mosb1000@mac.com> on Sunday April 01, 2012 @04:05PM (#39543267)

      It's always bothered me that forensic science doesn't involve science. It would be easy enough to reproduce the events using the same phone and recording equipment and voice samples with a few different people to establish a baseline for what we "should" expect. But instead they simply can't be bothered with scientific controls. We are just supposed to take their word for it that it should be a "90%" match. I'm this context we don't even know what 90% means.

  • by Anonymous Coward on Sunday April 01, 2012 @01:52PM (#39542335)

    I'd be very, very leery of trying to do any sort of "voice print analysis" on the basis of recorded cellphone audio. There's a lot of coding artifacts. After all, the goal is to allow people to communicate, not to convey voice identity.

    You can run voice through a fairly low bit rate LPC coder and it's quite intelligible on the other end, and actually "sounds" like the speaker, but if you look at the spectrogram, it's totally different. Your ear hears the dominant formants in the vowels, and you recognize speakers by that. LPC basically encodes the vocal tract as a 8-10 term filter plus a buzz excitation source. "voice print id" depends more on fine structure, which is lost in the encoding/decoding. It would be like trying to identify a paper document that was watermarked by looking at a photocopy. The watermark may or may not come through, but the intelligibility of the document is the same either way.

    • by gstrickler (920733) on Sunday April 01, 2012 @02:18PM (#39542533)

      Cell phone audio is generally higher than 8kbps, even on GSM [wikipedia.org] where 12k-13kbps is typical. CDMA (including 3G networks) is as high or higher. That's plenty for very reliable voiceprint analysis. And in this case, they have all the samples of Zimmerman talking before the screaming encoded using the same method, making the comparison even easier.

    • by mc6809e (214243) on Sunday April 01, 2012 @02:50PM (#39542779)

      I'd be very, very leery of trying to do any sort of "voice print analysis" on the basis of recorded cellphone audio. There's a lot of coding artifacts. After all, the goal is to allow people to communicate, not to convey voice identity.

      You can run voice through a fairly low bit rate LPC coder and it's quite intelligible on the other end, and actually "sounds" like the speaker, but if you look at the spectrogram, it's totally different. Your ear hears the dominant formants in the vowels, and you recognize speakers by that. LPC basically encodes the vocal tract as a 8-10 term filter plus a buzz excitation source. "voice print id" depends more on fine structure, which is lost in the encoding/decoding. It would be like trying to identify a paper document that was watermarked by looking at a photocopy. The watermark may or may not come through, but the intelligibility of the document is the same either way.

      Absolutely agree. I've actually written audio segmentation code for an alarm system that first obtained audio, not straight from an ADC, but only after the audio had been compressed. Compression absolutely destroys important details.

      I'm also skeptical of the software's ability to take calm, slow speech, and from that determine what a scream OUGHT to sound like. Comparing samples generated by a person speaking under similar conditions and looking for similarities isn't that hard. But the science of of voice analysis isn't yet to the point where we can take a sample of audio and from that build a virtual larynx, throat, mouth, and controlling nerve fibers so that we can extrapolate from that what a scream from that might sound like.

      This test might have been legitimate had the software been used to compare the scream on the 911 recording with another scream from Zimmerman, but that's not what was done.

  • by peter303 (12292) on Sunday April 01, 2012 @01:57PM (#39542367)
    "Trial by social media". "Trial by the self-righteous" Glad we have court systems to sort out all these rumors.
  • by Mitreya (579078) <mitreya@gmail.LAPLACEcom minus math_god> on Sunday April 01, 2012 @02:06PM (#39542435)
    Even if Zimmerman was the one screaming
    He followed and confronted someone for no good reason, even after having been explicitly told (by 911 operator) to stay away
    Even if Zimmeriman screamed through the whole process, the killing of Trayvon is not justified.
    • by Kneo24 (688412) on Sunday April 01, 2012 @02:23PM (#39542573) Homepage
      A thousand times this. You can't claim self defense when you purposely stalk someone and then confront them out in public. You are purposely potentially putting yourself in harms way. I wish more people would understand this.
    • by N0Man74 (1620447) on Sunday April 01, 2012 @02:49PM (#39542769)

      I agree... I don't doubt that there was a physical confrontation that escalated into a shooting. There seems to be evidence to support that part of the story. The identity of the screamer is a bit of a moot point, considering that there wouldn't have been a confrontation in the first place if the over-zealous, paranoid, and armed Zimmerman had not pursued the kid and created the confrontation (against the instructions given to him by the 9-1-1 dispatcher).

      I do wonder how this story would have turned out if Zimmerman had been the one killed in this. Would it have been dismissed as self-defense under the Stand Your Ground law (since Zimmerman was pursuing him in a way that could easily be perceived as threatening), or would he have been arrested at the scene? If he had been arrested, would there have been a public outcry? What if he had been released?

  • by the eric conspiracy (20178) on Sunday April 01, 2012 @03:05PM (#39542883)

    After the Richard Jewel case I don't get into rush to judgement any more.

    However the initial response of the police to let Zimmerman walk without a real investigation deserves a whole lot of criticism. Now we are getting the investigation so hopefully the facts will become evident.

  • by l0ungeb0y (442022) on Sunday April 01, 2012 @05:02PM (#39543701) Homepage Journal

    There's so many opinions flying around here about this, yet I honestly can't pass judgment on any of this information.
    All I do know is that the existing evidence is more than sufficient to merit formal charges and an arrest warrant.

    If the DA has not put in for a warrant for Zimmerman as a suspect for murder by end of business on Monday, then I think the DOJ should pursue a more vigorous investigation into racism and misconduct within the DA's office and the Police Department.

    Frankly, Zimmerman's story has stunk from the get go, the actions of the police have stunk and the more information comes out, the more the entire thing stinks. If we don't see the DA take swift and immediate action on Monday, we'll know where their motives truly lie.

    • by Howitzer86 (964585) on Sunday April 01, 2012 @11:12PM (#39545857)
      You claim to not know enough information to pass judgement, and then you pass judgement, declaring racism. Nice doubletalk there. Just goes to show how insidious the anti-discrimination laws are. If he's broken the law, judge him with the law, but don't go tacking on bullshit like that just to prosecute your man anyway.
  • Hypothetical (Score:5, Insightful)

    by metacell (523607) on Monday April 02, 2012 @04:24AM (#39546973)

    Let's assume a large black man follows a 17-year-old white boy into an alley, claiming the boy looks suspicious. There, the black man shoots the white boy. The black man emerges with minor wounds to his head and claims that he, for unknown reasons, was attacked by the white boy, who, it turns out, had done nothing wrong and wasn't carrying anything suspicious. The prosecutor decides to not prosecute because they have no proof the black man didn't act in self-defence.

    Would people be so quick to jump to the black man's defence?

"No job too big; no fee too big!" -- Dr. Peter Venkman, "Ghost-busters"

Working...