Author Topic: Voice ID experts  (Read 31590 times)

0 Members and 1 Guest are viewing this topic.

Offline MJW

  • Hero Member
  • *****
  • Posts: 1304
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #120 on: June 05, 2013, 01:35:54 PM »
Somehow O'Mara should make a request that the hearing be based on Daubert.  It might be too late to file a motion.  I suppose he can do it as the hearing starts tomorrow.  As Hornsby told me, the state would probably have to agree.  But them not agreeing would be quite telling.

That seems like a good idea. O'Mara predicts jury selection will take two or three weeks, so by the time the trial begins, Daubert may apply.

Offline MJW

  • Hero Member
  • *****
  • Posts: 1304
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #121 on: June 05, 2013, 01:43:20 PM »
I disagree with Hornsby that the voice evidence meets the Frye test but not the Daubert test. I think it meets neither.

Offline nomatter_nevermind

  • Hero Member
  • *****
  • Posts: 5447
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #122 on: June 05, 2013, 05:00:52 PM »
Direct link to Primeau article on Zimmerman case.

I'm amused that in the first article on the page, Primeau complains repeatedly that 'In our country today, we are guilty until proven innocent, the opposite of what our United States Constitution promises.' Then, in the Martin/Zimmerman article, this:

Quote
Remember, Zimmerman ignored the police when they told him not to follow Trayvon.

That has not been proven. I believe it myself, but that's another matter.

It's also tendentiously worded. The dispatcher is an employee of the police department, but he's not a police officer, and so not 'the police' for the purpose of giving citizens directives.

Offline nomatter_nevermind

  • Hero Member
  • *****
  • Posts: 5447
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #123 on: June 05, 2013, 06:17:17 PM »
He now takes into account Zimmerman's examplar, and appears to back off the certainty of his initial conclusions.

On Lawrence O'Donnell, 4/2/12, Primeau said his confidence was 'in the nineties'.

Transcript

Video

It's at 3:53 on the video. O'Donnell starts talking to Primeau at 3:00.

I don't see where Primeau backs off that in the recent article or video. He seems to me be talking more about what other experts will say, than his own opinions.

Offline RickyJim

  • Hero Member
  • *****
  • Posts: 1576
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #124 on: June 05, 2013, 08:16:45 PM »
Has the overall coverage of the Martin/Zimmerman on MSNBC become any more objective over the past year?  I haven't seen much progress in that direction at the Huffington Post.  It makes me uncomfortable that my politics is so close to the people who I think are dead wrong about this case.  I suppose it bothers TalkLeft too.

Offline Philly

  • Jr. Member
  • **
  • Posts: 37
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #125 on: June 05, 2013, 08:38:08 PM »
Quote
I don't see where Primeau backs off that in the recent article or video. He seems to me be talking more about what other experts will say, than his own opinions.

You may be right that he's speaking from the point of view of what other experts might say.  But given what he was saying in 2012, the new article seems a big shift.

As folk have noted, Ed's 2012 articles and interviews asserted that he was all all but certain (95%) that Zimmerman was not the one screaming. 

“I believe that’s Trayvon Martin in the background, without a doubt,” Primeau says, stressing that the tone of the voice is a giveaway. “That’s a young man screaming.”

April 4, 2013: http://edprimeau.com/category/trayvon-martin/

This particular article is colored by his assumptions, he states, "Think about it, why would a guy with a loaded gun who was stalking someone suddenly turn into a victim? That’s what the Zimmerman team is claiming. Whatever Trayvon did was in self-defense."  He already had a conclusion in mind before comparing Zimmerman's speaking voice with the 911 call screaming.

The more recent article is a huge contrast in that unlike Alan Reich, Ed no longer puts forward such strong conclusions, but rather dangles it as an open, unresolved question: "The question still remains; who screamed for help the night of February 26, 2012."  He emphasizes the difficulty comparing the exemplars with the screaming, noting that the screams in the original recording are more elongated and intense.  He ends the youtube video suggesting steps that could be taken to help establish "reasonable certainty."

Offline nomatter_nevermind

  • Hero Member
  • *****
  • Posts: 5447
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #126 on: June 06, 2013, 02:17:59 AM »
Has the overall coverage of the Martin/Zimmerman on MSNBC become any more objective over the past year?

I don't recall seeing any MSNBC coverage for months.

I think interest in the case all over the national media declined sharply after Zimmerman was arrested. For many people I think the national story was the outrage over Zimmerman not being arrested. Once that was done, for them it became another local crime story with a racial angle.

Offline jjr495

  • Full Member
  • ***
  • Posts: 118
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #127 on: June 12, 2013, 09:07:09 PM »
I know that Owen has been discredited, but I do not feel it is inappropriate to pile on further given the gravity of the situation in which he has put himself. I got a hold of the Easy Voice Biometrics manual from SpeechPro.

One page of the manual says the following:

False Rejection, FR is a probability of error when two voice samples being compared are considered different.
The larger FR value is, the higher is the probability that two voice samples are of the same speaker.
False Acceptance, FA is a probability of error when two voice samples being compared are considered of the
same speaker. The smaller FA value is, the higher is the probability that two voice samples are of the same
speaker.
As a result of comparing two voice samples (one is a voice standard and another is indefinite), the FR and FA
values are calculated.
Likelihood ratio, LR is a measure of similarity between two voice samples being compared; it is calculated as
FR/FA probability ratio. The lager ratio value is, the higher is the probability that two voice samples are of the
same speaker.
Comparing all these three values, the system then calculates generalized similarity value for biometric
characteristics of the target and source speakers.


Here is a screen shot of the EVB main window from SpeechPro.

After a comparison of two audio samples, EVB reports FR,FA, and the generalized similarty value referenced above. This is the Matching percent seen in the EVB screen. It is not the likelihood ratio. Unfortunately, how SpeechPro calculates this matching percent is proprietarty information, but it is on a scale from 0-100%. I believe Dr. Nakasone references the likelihood ratio in his testimony, and Dr. Doddington uses the log of the likelihood ratio in the NIST evaluations.

Also, you should know that the relative strength of the match is not colored, grey, yellow, or green for poor,soft, medium, and strong matches, repectively.

Owen does not know that the key value in everything he is doing is not the likelihood ratio. If you have followed this far, watch this interview with Soleded OBrien (video at the bottom of the page). It is Chris Farley meets Spinal Tap. He has no idea what any of the numbers mean, so he keeps avoiding her questions. He continues to call the Matching percent the likelihood ratio. This is the software that he sells and on which he teaches courses?

It gets much worse....

Offline RickyJim

  • Hero Member
  • *****
  • Posts: 1576
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #128 on: June 13, 2013, 08:17:36 AM »
Does the manual go into how the pair of voice samples must be chosen so that the resulting probabilities are meaningful?   Do they mention pairs where one is screams and the other is speech?

Offline MJW

  • Hero Member
  • *****
  • Posts: 1304
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #129 on: June 13, 2013, 11:43:37 AM »
On June 7, 2013, the state filed a Notice of Preliminary Objection regarding the Frye hearing. The state says the defense shsouldn't be allowed to question witnesses about what they believe was said, since that's outside the realm of Frye hearings.

I think the defense might respond that it's necessary for voice ID to know what the expert witness believed the words were, since the formants depend on it. Comparing two words with different vowel sounds is far from standard methodology. They won't match for the same speaker or for a different speaker.

Offline RickyJim

  • Hero Member
  • *****
  • Posts: 1576
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #130 on: June 13, 2013, 11:59:47 AM »
I think Mantei's motion has to do exclusively with Reich's testimony.  I find it difficult to understand what he is getting at.  I thought the defense was complaining that amplifying a section and listening to it over and over isn't a scientific method, and thus ruled out by Frye,  unless different people doing it will get the same results.  Mantei seems to be saying that what Reich is doing is no different than what is done every day with a recording where some sections are inaudible.  I doubt that is true.

Offline MJW

  • Hero Member
  • *****
  • Posts: 1304
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #131 on: June 13, 2013, 12:25:36 PM »
I think Mantei's motion has to do exclusively with Reich's testimony.

The case it cites, McCoy v. State, 853 So. 2d 396 (Fla. 2003),  and a case McCoy cites, Holland v. State, 773 So. 2d 1065 (Fla. 2000), seem to completely undermine the admissibility of Reich's testimony about what was said.

Offline MJW

  • Hero Member
  • *****
  • Posts: 1304
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #132 on: June 13, 2013, 06:16:11 PM »
I think Mantei's motion has to do exclusively with Reich's testimony.

I think it more likely refers to Owen's manually entered interpretation of what was said, as mentioned in cboldt's comment on the June 7th hearing thread.

Offline cboldt

  • Hero Member
  • *****
  • Posts: 1262
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #133 on: June 13, 2013, 06:41:37 PM »
I think it more likely refers to Owen's manually entered interpretation of what was said, as mentioned in cboldt's comment on the June 7th hearing thread.

The effect of the notice would only affect the Q&A during hearing anyway.  Watch for the "word ID is an audibility issue" argument to resurface when the state and defense argue their respective sides after evidence has been heard.

If the objection is supposed to salvage Reich's hearing of words, it fails.  Burden is on the proponent to establish that the methods are not new and novel, are accepted by the scientific community.  Reich's methodology is scientifically opaque - there is no science there (word ID via amplification and trained ear and hundreds of hours of listening), at least not written in his report.

Mantei's argument is essentially that inaudible/non-transcribable "words" are an out of bounds subject in the Frye hearing.  That Reich's testimony passes the Frye test because Reich is transcribing inaudible "words."  IOW, if it's inaudible, it passes the Frye test.  Well, okay then, Nelson can be the last word on the transcription.

Transcripts under McCoy have (inaudible) notations right in them; and the judge has to find/agree that the transcription doesn't make representations it isn't entitled to make, before a jury is allowed to see the transcript.

Offline nomatter_nevermind

  • Hero Member
  • *****
  • Posts: 5447
  • Rate Post +0/-0
Re: Voice ID experts
« Reply #134 on: June 14, 2013, 01:55:24 AM »
If the objection is supposed to salvage Reich's hearing of words

It doesn't read that way to me. To me it sounds like they are conceding that there is no relevant expertise in word identification, and implying that they won't be offering anyone to testify as an expert on that.

 

Site Meter
click
tracking