June 7th Hearing

[cboldt]

West elicits that Owen has performed analysis other than from EZ Voice.  The defense doesn't have any of this, but Owen did send some spectrograms to Mantei.  Owen says he also sent a PowerPoint, and West says the defense doesn't have that, either.

West wants to know about the standards applied in aural spectrographic analysis.  Owen says the standards are available at his website, but what is there is not current.  Owen says there is a 12 step methodology that he teaches.  From memory, this is where Owen is self-disqualifying, as he does not adhere to the standards that he developed and published.  How much speech do you need under you standards?  24 words, plus clarity of recording.  Owen admits he does not have a sample that size, so he cannot make a conclusive/positive finding; any finding is couched in probability.

He also says that he does not have good evidence to work with.  His conclusion is simply that it is not probable the voice is Zimmerman.  West wonders if the probability is quantified.  Owen obfuscates.

West asks if Owen compared screams with normal voice side by side.  Owen says yes, West asks for a description of the method.  West asks Owen if he is aware of any published methods for doing this sort of comparison.  Owen goes on to explain how he does a comparison, without referring to any published study.  Owen finally says he is not aware of any studies.  West gives Owen an opportunity to raise any other points, not brought up in questioning, that relate to owen's aural/spectrographic analysis of the 911 call.  Owen has none.

West on to criteria other than number of words, that appear in the standards.  Same language, signal to noise.  West asks if "stress" on part of speaker is part of the standard; or if stress contributes to ability to reach a conclusion.  Owen says emotional state of speaker is part.  Other factors? Recording method, distance between speaker and recorder.  West asks what are the standards, and where, if anywhere, did you deviate.  Owen obfuscates.

West asks the minimum probability of error to warrant the label "probable."  Owen says FBI standard is less than 1% error rate.  Owen has tested his system, and it meets the 1% error rate.  This ducks the question, but sets the stage for West to ask if Owen is 99% sure of his conclusion.

[cboldt]

Owen has only testified once (out of thousands of times not disqualified) using EZ Voice.  West asks if Owen was challenged in a Frye hearing in the case in Connecticut.  That was the Sheila Davalou case.  He asks Owen if there was an admissibility hearing, outside of the jury earshot, in that case.  Owen obfuscates.  Owen says that he doesn't think there was an official admissibility hearing in that case.  Owen did testify in that case, but Owen does not recall any challenge to the Frye or Daubert threshhold.

West is done.  Redirect examination.  Mantei rehabilitates Owen as far as inability to program/understand EZ Voice.  The example is microscope, which Owen knows how to use, but not how to make.  Same with Skype.  Goes on to rehabilitate as to uncertainty or deviation from preferred samples (not) precluding analysis.  Mantei asks about the pitch change being an alteration of the evidence, Owen obfuscates (he could have answered this).  Mantei points out that the pitch change was performed on a known Zimmerman voice (from the recreation).  Mantei goes on to looping, and says that Owen said that looping was not new or novel - West objects, saying that Owen testified that it was new or novel.  Nelson allows the line of inquiry to proceed.  Owen says he's looped before.  What is new here is EZ Voice demands a sample length.  Mantei raises the point that deviation from a standard doesn't render the opinion invalid (West never said that, he asked the witness to quantify the effect of deviation).  Mantei says basically this, that the deviation affects the certainty of conclusion.  Owen says that positive conclusions are generally not obtainable in this field.  Mantei is done.

West on re-cross.  Focuses on standards.  May analysts do whatever they want and it is okay?  No, says Owen.  Anything less that 10 words precludes reaching a conclusion (that might be useful for defense), but the standards are more in the nature of guidelines.  West says that this analysis includes substantial deviations - Owen says he did not substantially deviate, he just had less than 20 words.  Owen says he had 10 words.  Owen wants a definition of "phonetically balanced."  West says "recognizable words."  West asks how much money have you been paid?  Zero.  That is nuts.  He worked for free for Orlando Sentinel!  West makes a remark about sale of software.  He's being paid for his testimony today.

The court is done with this witness.  Court is in recess until 2 p.m.

[RickyJim]

Apparently they haven't asked Owen to compare the screams with the now available Trayvon Martin speech.  I wonder why.

[cboldt]

Owen said the program would not try to make an id on less than 16 seconds. So he doubled his eight second compilation. I would think there is a reason that the software creators had a reason for the 16 second limit. Why not take 1 second and repeat it 16 times.

The software creators are touting a certain degree of reliability, and in order to get that, some parameters regarding the sample have to be set.  Owen should have been able to explain the principles behind this (given a 140 Hz nominal voice pitch, 16 seconds gives a fixed number of "cycles" to work with - about 2200, more for harmonics and semitones) and approximate the resulting loss in reliability as the sample size is reduced.

To take an extreme example, if I ask you the probability of tossing heads or tails (and you don't know if the coin is "weighted" or "loaded" to use dice parlance), and give you one toss, you will say either 100% or 0%, depending on the result of your sample toss.  It takes more samples in order to have confidence in your conclusion.

[Redbrow]

Apparently they haven't asked Owen to compare the screams with the now available Trayvon Martin speech.  I wonder why.

Because Martin's voice is noticeably lower and deeper than Zimmerman's and even less probable to be the screamer on the 911 tapes by this charlatans ezbake software analysis.

The fact that a sample of Trayvon's voice has been kept away from scrutiny for this long speaks volumes.

[jjr495]

The software creators are touting a certain degree of reliability, and in order to get that, some parameters regarding the sample have to be set.  Owen should have been able to explain the principles behind this (given a 140 Hz nominal voice pitch, 16 seconds gives a fixed number of "cycles" to work with - about 2200, more for harmonics and semitones) and approximate the resulting loss in reliability as the sample size is reduced.

To take an extreme example, if I ask you the probability of tossing heads or tails (and you don't know if the coin is "weighted" or "loaded" to use dice parlance), and give you one toss, you will say either 100% or 0%, depending on the result of your sample toss.  It takes more samples in order to have confidence in your conclusion.

Sorry I was being sarcastic. My wife has the same reaction to me all the time. So do my students.
I am a professor and teach and use  statistics, probability, and spectral analysis in my work on turbulence.
Pitch is the fundamental frequency produced by the vocal chords. The formants are secondary, tertiary, etc, resonances in the throat. When you change the pitch of your voice the formants dont necessarily scale up. So by increasing the frequency Owen is not scaling the formant changes correctly. He should never have used this technique.
Your coin analogy is correct. One needs to build a sampling distribution to assign probability. There is no sampling distribution for screams because nobody has done the research. Further, Nakasone, testified that nobody could do this research at the present time.

[ding7777]

Owen compares too-short audio clip to a partial fingerprint, says could be used to rule out people.

If you only had 1 or 2 ridges and "looped" them to get a big enough partial, then, imo, 100% of the population could be ruled out

[cboldt]

Sorry I was being sarcastic. My wife has the same reaction to me all the time. So do my students.
I am a professor and teach and use  statistics, probability, and spectral analysis in my work on turbulence.
Pitch is the fundamental frequency produced by the vocal chords. The formants are secondary, tertiary, etc, resonances in the throat. When you change the pitch of your voice the formants dont necessarily scale up. So by increasing the frequency Owen is not scaling the formant changes correctly. He should never have used this technique.
Your coin analogy is correct. One needs to build a sampling distribution to assign probability. There is no sampling distribution for screams because nobody has done the research. Further, Nakasone, testified that nobody could do this research at the present time.

West was trying to get to the questions about scaling up the pitch vs. the spectrum if the human scales up the pitch.  Owen claimed to have undertaken a study, published in 1985, that relates to this subject.

My point, which I now know you understand, given your background, is that maybe 4 seconds is enough, depending on what is independently known.  The software is for dummies who don't understand the physics and math, so "forces" them to provide a good sample.

I'd have to revisit Nakasone's testimony, but my take was not that the research couldn't be done.  The research has been done, and there is no way to scientifically analyze screams and correlate them with normal voice examples.

[annoyedbeyond]

What happened re the materials Owen was going to email in? I had to leave before they came back from that recess.

[RickyJim]

Another "expert"  that tested both Martin and Zimmerman against the screams.  Notice who the reporter is. 

[jjr495]

The research has been done, and there is no way to scientifically analyze screams and correlate them with normal voice examples.

The research has not been done. Nakasone described the most recent studies in which people are made to talk over noise. This is called Lombard speech. There are very big changes that occur to spectral signatures in Lombard speech. Nakasone noted that Lombard is the outer limit of what software algorithms had been modified to sample.
In order to train software to id mortal screams a database of mortal screams is necessary. Someone did collect roller coaster screams. Of course the other part of the equation is the exemplar. Could the software ever be trained to take normal or perhaps shouts to identify screams. Not likely.
Anybody claiming they can do it today is "disturbing".

[annoyedbeyond]

the EZTalk software people have a vid on their site where they tried to fool the software by including other random noise. I didn't bother to watch.

[Evil Chinchilla]

Because Martin's voice is noticeably lower and deeper than Zimmerman's and even less probable to be the screamer on the 911 tapes by this charlatans ezbake software analysis.

The fact that a sample of Trayvon's voice has been kept away from scrutiny for this long speaks volumes.

I can't get the ABC website entry at the link posted earlier in the thread to open when I go there.

I can see they have something posted there, but my ancient computer won't open it.

Can you hear TM's voice in that audio clip (or whatever ABC's linked there)?

[annoyedbeyond]

I can't get the ABC website entry at the link posted earlier in the thread to open when I go there.

I can see they have something posted there, but my ancient computer won't open it.

Can you hear TM's voice in that audio clip (or whatever ABC's linked there)?

Yes. It's deep--not Barry White deep but deeper than GZ's and also a little...sharp. Have you tried a different browser? I couldn't get it to open in FireFox but it opened in Safari.

[RickyJim]

Trayvon says "Ooooh, he coming for that". and "We need a behind the scenes.".  They are very clear.