Merged Continuation Part 2: Discussion of the George Zimmerman case

[TheL8Elvis]

My brain was not multitasking properly. Yes, her decision was overturned by the DCA, but with strict limitations.

The rulings she made on the matters before the court last wee were made based on case law.

Two entirely different things and I didn't separate them out. My apologies.

No apologies necessary.

 

[ABC10]

What do folk know about the recording of Martin's voice? I am deaf and can't hear any of it. What do Zimmerman and Martin's voice sound like? Can the FBI do more analysis with now having a Martin voice print? What's with Owen and the snipes at Don West. Owen even made Don laugh which is amazing. Isn't the FBI the Gold Standard?

 

[Unabogie]

One thing that none of the experts heard, prosecution or defense, was "you're gonna die tonight, ************".

In George's telling, just after he "felt" that Trayvon had seen his gun, the victim said "you're gonna die tonight, ************", and that's when George grabbed his gun and shot him.

Yet that moment is clearly covered by the 911 tape at issue, and though all of the witnesses can hear "help me", none of them hear this phrase from Trayvon. One of them does hear "I'm begging you" coming from Trayvon, but again, no one hears the shocking declaration of intent to murder that George claimed.

Especially since even laymen can hear both of their voices, shouldn't this be audible as well if actually happened?

 

[phonebook]

Especially since even laymen can hear both of their voices, shouldn't this be audible as well if actually happened?

Who are these laymen you're referring to? As a laymen, I've never been under the impression that two voices were distinctly heard, especially so if I was supposed to distinguish specific words attributed to the respective speakers/screamers.

There would obviously have to be a certain decibel threshold to have been exceeded for that to have been picked up by the 911 call.

 

[boooeee]

Yet that moment is clearly covered by the 911 tape at issue, and though all of the witnesses can hear "help me", none of them hear this phrase from Trayvon. One of them does hear "I'm begging you" coming from Trayvon, but again, no one hears the shocking declaration of intent to murder that George claimed.

Can you clarify which witness are you referring to that claims they heard Trayvon say "I'm begging you"? The only one claiming that is Alan Reich, as far as I am aware. And Alan Reich hears lots of things.

 

[Unabogie]

Can you clarify which witness are you referring to that claims they heard Trayvon say "I'm begging you"? The only one claiming that is Alan Reich, as far as I am aware. And Alan Reich hears lots of things.

How is that different from what I said, and how does it answer the question of why you can hear words, but not the words that the defendant claimed?

 

[Unabogie]

Very interesting. Apparently the the U.S Naval Criminal Investigative Service is buying a copy of Easy Voice Biometric. Apparently at least one branch of the Federal Government is impressed by it even if the FBI doesn't like it.

https://www.fbo.gov/index?s=opportu...5309bfa7d6424f9b3d2dd&tab=core&tabmode=list&=

Actual description of contract here:

http://www.neco.navy.mil/synopsis_file/N632851310126_N6328513SU10126_SSJ.pdf

The proposed acquisition is to purchase Easy Voice Biometrics software that is used for voice
identification and analysis. The Easy Voice Biometrics product is designed for Law Enforcement
Agencies to perform the following tasks:
1. Facilitating voice expert identification analysis in the performance of multi -target
forensic audio investigation by eliminating imposters and ranging the top-in-the list
speakers according to the biometric traits likelihood probability.
2. Express attribution of the investigated speakers voices by the proximity degree.
3. The product is designed to assist in two ways:- For multi-target voice analysis cases to
arrange the speakers by match probability.
4. 1 vs 1 comparison
5. Easy Voice Biometrics uses a total variability method; the following steps are applied
for comparison using this method: Speaker's voice model calculation.
6. On that stage, there are MFCC (Mel Fourier Cepstrum Coefficients, cepstral
coefficient by Mel scale) speeches features calculated- the features, describing spectral
representation of speech signal in different moments. Density of identification features
distribution is modeled by using of GMM. Using the specifically adapted factor
analysis, the feature of GMM-model are represented as an 1-vector of small dimension
in the space of "total variability", which includes subspaces of" own channels" and
"own voices", which are used in JF A (Joint Factor Analysis Distinctive feature of such
GMM-model representation is its high self-descriptiveness and small data size.
7. Making the decision about the identification, on stage ofl-vectors comparison the
SVM-classifier is used, which provides high speed and good quality of identification.
By the resulting SVM distance, there are FR/F A errors calculated)

 

[Mister Earl]

Yeah, but that's an appeal to authority fallacy. Some people of equal authority levels also bought magic dowsing rods and magic healing silver suspension fluid that turns you into a smurf. Their approval has no bearing on whether or not it works.

 

[johnny karate]

If you are following the case, you know the judge had not ruled on the issue when I made the post, and in fact, will not rule on these issues until after the trial. That's why I used the hilited words to express my interpretation of the evidence that was presented yesterday.

Do you really need a judge to tell you if someone is lying or not, or can you make that determination on your own ?

Yeah, my bad. I keep forgetting to apply that double standard.

Shellie Zimmerman and Mark O'Mara are not liars because I can't prove to your satisfaction they are lying.

Bernie de la Rionda is a liar because you say so.

Of course, I can point to statements made by both Shellie Zimmerman and O'Mara that are demonstrably untrue, but that doesn't quite meet your threshold of evidence.

But with de la Rionda, apparently that threshold is lowered to a gut feeling.

 

[Unabogie]

Yeah, but that's an appeal to authority fallacy. Some people of equal authority levels also bought magic dowsing rods and magic healing silver suspension fluid that turns you into a smurf. Their approval has no bearing on whether or not it works.

But this goes directly to the defense expert. He claims he does not consider it possible to make a call based on the shortened amount of clean audio available. But some people go further and claim that the entire field of software voice identification is nothing but a scam. That may be true or false, since none of us have gotten a chance to use the software ourselves, but it's a data point in its favor that the Navy crime people have tested it and found it effective, hence the Navy is buying it and says it works for:

Facilitating voice expert identification analysis in the performance of multi -target
forensic audio investigation by eliminating imposters and ranging the top-in-the list
speakers according to the biometric traits likelihood probability.

And that's exactly what Tom Owen did here. He took voice samples of Martin and Zimmerman and tested them against the sample audio and ranged them according to probability. That's no an appeal to authority any more than citing that the FBI didn't approve the software is an appeal to authority.

 

[Mister Earl]

Hrm! Ok, gripe withdrawn. I had missed the context of it.

 

[phonebook]

I was under the impression that it wasn't the software itself that was the problem, it was that the sample that was used with the software didn't even meet the software's minimum requirements for analysis.

 

[Unabogie]

I was under the impression that it wasn't the software itself that was the problem, it was that the sample that was used with the software didn't even meet the software's minimum requirements for analysis.

The criticism of this software has been morphing from one to the other. I have seen no evidence that the software is "woo", and indeed the Navy has found it effective and is purchasing it for their use. It may be true that it requires a certain length of audio to work, or conversely that looped audio may work just fine. This seems to me to be a ripe area for cross examination, but doesn't prove that software voice identification doesn't work at all. That should go to weight, not admissibility. If you look at what Nakasone does, he finds markers in the audio spectrum one by one, using manual notation. All this software does is replicate that work using code. It assigns probabilities to the results based on a Gaussian algorithm. I'd liken it to the way Netflix finds movies I like and Netflix is usually spot on. If it thinks I'll hate some movie and give it a "1" and thinks I'll love some other movie and give it a "5", it's using the same sort of algorithm we see here. Easy Voice is perhaps also using some sort of Monte Carlo simulation to run subtle alterations to the audio peaks and running that against the samples thousands of times and then averaging out the results.

I'd love to get ahold of a copy and test it out, though. It looks like a fun toy.

 

[Cylinder]

I was under the impression that it wasn't the software itself that was the problem, it was that the sample that was used with the software didn't even meet the software's minimum requirements for analysis.

There are two different issues in play here. The FBI asserts that biometric speaker recognition is not robust enough to use as evidence at trial under the best of circumstances but rather should be confined to use as an investigative tool.

Even if you don't adopt this view, there's no methodology - aural, spectral or biometric - that has proven useful in speaker recognition with three seconds of speech under ideal situations. There's no methodology that has proven useful in scream-to-scream or scream-to-speech speaker (screamer?) recognition. There's no methodology that's proven useful in distant voice speaker recognition.

One of Dr. French's students is undertaking an analysis study of distress calls from the UK equivalent of 911 recordings to try and understand the "terror" component of speech by comparing it to normal speech exemplars. According to French, the variability of a screaming voice and normal speech is as great and the variability between random speakers.

 

[Unabogie]

There are two different issues in play here. The FBI asserts that biometric speaker recognition is not robust enough to use as evidence at trial under the best of circumstances but rather should be confined to use as an investigative tool.

Even if you don't adopt this view, there's no methodology - aural, spectral or biometric - that has proven useful in speaker recognition with three seconds of speech under ideal situations. There's no methodology that has proven useful in scream-to-scream or scream-to-speech speaker (screamer?) recognition. There's no methodology that's proven useful in distant voice speaker recognition.

One of Dr. French's students is undertaking an analysis study of distress calls from the UK equivalent of 911 recordings to try and understand the "terror" component of speech by comparing it to normal speech exemplars. According to French, the variability of a screaming voice and normal speech is as great and the variability between random speakers.

But we have other audio to listen to. For instance, if we have two voices from the victim and the killer happening together, and one of them is screaming, we can try and match the other voice to the killer. If that matches, then we can deduce that the screamer was the victim. So even though the scream may not be 100% provable, if the other voices are, that can tell us who was screaming. Especially if the scream is found to be more likely Trayvon while a spoken phrase that happened before or after the scream can be shown to be George, then that merely corroborates the finding of the scream analysis. None of this shows that the science behind this is "woo". And until you hear the entire state's results and subject it to cross, you won't be able to assert that they are using this to corroborate other evidence or using this on its own. That hasn't come up in the Frye hearing and won't, since a Frye hearing is only to determine if a technology is "novel" and clearly this isn't novel in any way.

 

[BenBurch]

Free demo of the software here; http://www.easyvoicebiometrics.com/media/demos/EVBdemo.zip

 

[Skeptic Tank]

And that's exactly what Tom Owen did here. He took voice samples of Martin and Zimmerman and tested them against the sample audio and ranged them according to probability. That's no an appeal to authority any more than citing that the FBI didn't approve the software is an appeal to authority.

Do you have anything to back up that Tom Owen ever compared an exemplar from Trayvon to the screams? I may have missed it, but I don't remember him ever attesting that he'd received such, let alone run it through his software. I don't recall him saying so in his testimony in the Frye Hearing, and I don't recall him saying so in the media at any time. The only time I remember him referring to an exemplar from Trayvon was expressing a desire for one, a need for one, and a plan to receive it from Trayvon's family. That was the last thing I heard from him on it and that was over a year ago. Please correct me if I've missed something.

A few points I'd like to make just generally on these matters also:

• Appeal to authority shouldn't be mixed up entirely with appeal to expertise, appeal to expertise is what expert testimony is all about and when you have people with several PhDs and decades of work in this field, dismissing things like Reich's EVP out of hand as pattently absurd... vs. senile sounding Reich and history degree wedding photographer Owen... I have no problem appealing to expertise. Their testimony doesn't belong at trial.
• I have no reason to doubt that EasyVoice Biometrics can be effective comparing spoken voice to spoken voice. Comparing distant scream to spoken voice is clearly outside of it's capabilities or the capabilities of this entire science currently, leaving the only conclusion as to why someone would attempt it to be what I've said before... a desire to increase one's profile and bank account.
• If Nelson isn't able to figure out that she has quackery and woo on her hands from people like Reich and Owen attempting to worm into her courtroom after hearing from people who work for/with the FBI and NSA and helped develop this entire field, who are prepared to outright say it's absurd on it's face... then she doesn't deserve to be a judge, as two reversals from the DCA in one day tends to point to also.
• If there is a "I'm begging you" present in the background, the logical conclusion based on injuries and GZ's recollection of appealing to John NOT to just merely go call cops, but rather to come help him, is that it is Zimmerman screaming it. By no means conceding this utterance is present, but I do hear the part people THINK is that utterance and I don't dismiss it entirely. Again, it would be Zimmerman based on other evidence, if it's present at all.
• "These shall be" is nonsensical. I hear the part where it's claimed to be present, and it is in the foreground, not coming from outside where the parties would be. It's either a TV, or some sort of signal overlap/interference. Again, not conceding this is what is actually heard, but I can at least hear the portion Reich thinks this is. Regardless of WHAT it is being said, it is VERY clearly not coming from outside the building like the screams.
• "You're gonna die tonight" would have been something, like "shut the **** up" which was spoken solely for Zimmerman's ears by Trayvon, if it happened. Whereas Zimmerman's screams of pain and appeals for mercy were screamed specifically with neighbors and responding officers hearing it in mind. Given the recording conditions, nobody should expect these utterances to be audible. I'm surprised you'd suggest that they would be, Unabogie...
• People have said "why doesn't the defense have any experts saying it's George?" implying that they can't find anyone to say it, and that this is really damning of their case. Well, as we just saw... ABC provided samples of both Zimmerman AND Trayvon (something I'm only aware of H&H also having had, anyone have evidence Reich, Owen, or Primeau ever had a Trayvon sample? Any of them issue findings about that?) and ABC's expert attested that it was significantly more likely to be Zimmerman. Intriguing that the one other group (H&H) who we know definitely had Trayvon exemplars, also happened to reach much more nuanced conclusions than Owen, Reich, and Primeau - whom to my knowledge have not ever directly expressed possession or use of such Trayvon exemplars. I'm not going to be Mr. Hypocrit here and say I accept the science's ability to do this when it comes to the conclusion I want, whoever this expert is (Gibson I think?) for ABC, he doesn't belong at trial saying that either. It's as I said a few days back, the defense is being ETHICAL by bringing in experts who say the truth, which is that this isn't within the science's capabilities, rather than just bringing in tit-for-tat dueling experts who say it was Zimmerman. Which the ABC situation proves THEY COULD HAVE DONE. Contrary to what some here have said.

 

[johnny karate]

From the other thread:

It's as I said a few days back, the defense is being ETHICAL by bringing in experts who say the truth, which is that this isn't within the science's capabilities, rather than just bringing in tit-for-tat dueling experts who say it was Zimmerman. Which the ABC situation proves THEY COULD HAVE DONE. Contrary to what some here have said.

No, they couldn't have:

However, he adds, so much of the howling and pleading overheard on that 911 tape is muffled or obscured, that only two seconds of the tape are useable. Therefore, he says, that there can be no definitive identification of "the screamer."

If the defense had an expert willing to testify that the screams were Zimmerman's, they put him or her on the stand.

But they don't, so they can't.

And I find it utterly hilarious that it's the prosecution accused of being unethical, when so far we've seen from the defense:

• A deception to hide Zimmerman's finances from the court to get a lower bond

• A release of irrelevant evidence damaging to the victim's reputation on O'Mara's website

• A blatant misrepresentation of evidence in open court that, once again, damages the reputation of the victim

It would be refreshing if one of Zimmerman's defenders would actually provide some kind of evidence that the prosecution is being unethical instead of just repeatedly asserting it.

Because that song and dance - which has been ongoing since the probable cause affidavit was ignorantly labeled unethical - is getting tiresome.

Meanwhile, the defense files fraudulent documents with the court, and proclaims admitted falsehoods in court, and Zimmerman's defenders don't bat an eye.

 

[Unabogie]

Free demo of the software here; http://www.easyvoicebiometrics.com/media/demos/EVBdemo.zip

Oh cool! Gonna try it out tomorrow!

 

[johnny karate]

From the other thread:

... then she doesn't deserve to be a judge, as two reversals from the DCA in one day tends to point to also.

You apparently don't have a very high opinion of Judge Nelson.

Do you disagree with all her rulings, or just the ones that don't favor Zimmerman?