Trayvon Martin, Vigilante Justice

[WildCat]

"The first step is to evaluate the recording of the unknown voice, checking to make sure the recording has a sufficient amount of speech with which to work and that the quality of the recording is of sufficient clarity in the frequency range required for analysis. The volume of the recorded voice signal must be significantly higher than that of the environmental noise. The greater the number of obscuring events, such as noise, music, and other speakers, the longer the sample of speech must be."

Thomas J. Owen
International Association for Identification, Life Member (New Jersey Division)

But that's the Thomas J. Owen who didn't have a high-profile case to pimp his company over.

 

[Skeptic Ginger]

Here's the most problematic part for me:

Note the part in emphasis, which is also backed up by some of the descriptions and tutorials we've seen elsewhere on this.
He's saying that a large part of the analysis is the subjective judgement of the expert as to whether or not the spectrograms look the same and whether or not the voices sound the same.
That's really troubling to me; whenever evidence is based at least in part on a person's judgement, I want to see blinded testing to assure that the person's judgment is known to be accurate. Do we have any examples of forensic audio experts submitting to blind tests of their judgment abilities?

When I interpret a lab value in light of the patient's entire presentation, you can dismiss it as "subjective judgement" or consider it is the culmination of years of education and experience, aka "expert opinion".

The evidence presented here suggests Owen gave an expert opinion.

You are also misrepresenting the value of the software as it is presented by the company. The benefit of the software is described as wading through a lot of non-matching data quickly so that the matching data can be compared. That's exactly how fingerprint and facial recognition searching software works. Unlike what you see on TV where the computer finds the "match", the computer searches through an algorithm of criteria leaving the final matching for human analysis. So are you saying fingerprint experts get it wrong all the time because the technicians aren't qualified? I believe there is an error rate but it is quite small.

 

[Skeptic Ginger]

That's ok - at least the other expert relies on hard science.
http://edprimeau.com/expert-witness/

"My philosophy is that forensics examination and restoration of audio/video media ideally combines art, as well as science. The methods that an expert uses require attention to detail and scientific principles, complemented by an appreciation for clarity and aesthetics. My techniques are derived from both formal education and application of skills gained by working in many forensics situations."

A lot of people claim medicine is an "art" as well. Some of us perceive it is a "science". Surgery is an art, it takes a certain physical skill. But the common practice of calling some expert skills an "art" is a point of view, not a commentary on the nature of the science behind the expertise.

 

[Professor Yaffle]

I've become a lot more uncertain of even long established identification techniques where expert judgement is required (like fingerprinting) since the Shirley McKie case http://en.wikipedia.org/wiki/Shirley_McKie

But again, it is usually much more simple to rule out a suspect than it is to positively identify one.

Some other fingerprinting errors here http://en.wikipedia.org/wiki/Fingerprint#Instances_of_error

Note that all involve positively identifying someone (incorrectly) rather than incorrectly ruling someone out.

 

[AvalonXQ]

When I interpret a lab value in light of the patient's entire presentation, you can dismiss it as "subjective judgement" or consider it is the culmination of years of education and experience, aka "expert opinion".

This is an important point -- just because a technique requires subjective judgment doesn't mean it's not valid. As you correctly point out, many perfectly valid techniques require expert analysis for interpretation.

But, to the extent that a match or mismatch is judged based on an expert looking at a spectrogram and listening to a recording, I want blind testing to confirm that the expert is usually right.

 

[Unabogie]

Here is one:
http://www.experts.com/Articles/Voi...-Aural-Spectrographic-Method-By-Thomas-J-Owen

ETA - yes, they are all from there. I assume the different titles are just from the CV

I just find it rude when people spam the thread. If he found something of relevance, he should post a single comment and explain why he thinks it's important. Spamming just comes off as childish and if he has a point, no one will notice it.

 

[WildCat]

I've become a lot more uncertain of even long established identification techniques where expert judgement is required (like fingerprinting) since the Shirley McKie case http://en.wikipedia.org/wiki/Shirley_McKie

But again, it is usually much more simple to rule out a suspect than it is to positively identify one.

Some other fingerprinting errors here http://en.wikipedia.org/wiki/Fingerprint#Instances_of_error

Note that all involve positively identifying someone (incorrectly) rather than incorrectly ruling someone out.

As I mentioned earlier ruling a person out is exactly what would be expected if the audio samples used were insufficient, unusable, or otherwise compromised.

And I know of no study done where a 2 second scream meets the criteria for usability when compared to a longer spoken phrase. If you don't throw out the sample and use it anyway (because it gets your company name in all the papers and newscasts and the internet) of course it will be a poor match.

Pre-Trayvon Martin Thomas J. Owen seems to agree that the scream sample shouldn't have been used as a comparison to a normal speaking voice. Odd, isn't it?

 

[WildCat]

I just find it rude when people spam the thread. If he found something of relevance, he should post a single comment and explain why he thinks it's important. Spamming just comes off as childish and if he has a point, no one will notice it.

So you're just going to stamp your feet and complain about form rather than address the substance?

 

[Professor Yaffle]

As I mentioned earlier ruling a person out is exactly what would be expected if the audio samples used were insufficient, unusable, or otherwise compromised.

This wouldn't be true in other methods of identification like DNA, blood type or fingerprinting; what makes you think it is true wrt to voice analysis? Finding the samples to be incompatible is not the same as being unable to make a match.

 

[OmegaWolf]

Really? You've actually read and understood papers in the field? You know what, I really don't believe this to be the case. I read a bunch last night, following citations from the bibliography in

Here are more excerpts from that first paper (2009):

[Page 2]: "In general, phonetic variability represents one adverse factor to accuracy in text-independent speaker recognition. Changes in the acoustic environment and technical factors (transducer, channel), as well as
“within-speaker” variation of the speaker him/herself (state of health, mood, aging) represent other undesirable factors. In general, any variation between two recordings of the same speaker is known as session variability [111, 231]. Session variability is often described as mismatched training and test conditions, and it remains to be the most challenging problem in speaker recognition. "

[Page 16]: "Any variation in different utterances of the same speaker, as characterized by their supervectors – be it due to different handsets, environments, or phonetic content – is harmful."


I encourage everyone to read the entire paper, but certainly screaming versus spoken would qualify as severe "within-speaker" variation and have a corresponding effect on accuracy.

 

[Brainster]

ABC News enhances the video that appeared to show Zimmerman with no injuries to the back of his head. New conclusion: video shows "mark or gash" on the back of Zimmerman's head.

 

[Professor Yaffle]

ABC News enhances the video that appeared to show Zimmerman with no injuries to the back of his head. New conclusion: video shows "mark or gash" on the back of Zimmerman's head.

/me waits for double-blind testing of the specific video enhancement technique used.

 

[Unabogie]

So you're just going to stamp your feet and complain about form rather than address the substance?

Are you trying that hard to miss my point? Spamming is rude and results in people ignoring your posts. So no, I didn't stamp my feet. I just ignored what he wrote until he tries to engage in a way that's conducive to discussion.

 

[AvalonXQ]

* Professor Yaffle;8165771 waits for double-blind testing of the specific video enhancement technique used.

Although your remark is in jest, I'll concur -- I don't trust manipulated or "enhanced" video without some evidence of reliability in the "enhancement", so I'm skeptical of this evidence.

Ironically, if the enhanced video had been bad for Zimmerman instead of good, I'm sure SG and company would see it as another sign that I discard all evidence that hurts Zimmerman.

 

[TheL8Elvis]

ABC News enhances the video that appeared to show Zimmerman with no injuries to the back of his head. New conclusion: video shows "mark or gash" on the back of Zimmerman's head.

Shocking. All of the internet doctors already informed us there was no visible evidence.

 

[totoro]

"The spectrograms of the unknown speaker are then visually compared to the spectrograms of the suspects. Only those speech sounds which are the same are compared."

Thomas J. Owen
Audio/Video Authenticity and Voice Identification, Diplomate

That's if you are using spectrographic analysis. Not at all relevant if you are using biometrics, which do not, in any way, involve looking at visual representations of the samples.

 

[Professor Yaffle]

Although your remark is in jest, I'll concur -- I don't trust manipulated or "enhanced" video without some evidence of reliability in the "enhancement", so I'm skeptical of this evidence.

Ironically, if the enhanced video had been bad for Zimmerman instead of good, I'm sure SG and company would see it as another sign that I discard all evidence that hurts Zimmerman.

It's damn uncomfortable on this fence, isn't it?

 

[totoro]

Here's the most problematic part for me:

Note the part in emphasis, which is also backed up by some of the descriptions and tutorials we've seen elsewhere on this.
He's saying that a large part of the analysis is the subjective judgement of the expert as to whether or not the spectrograms look the same and whether or not the voices sound the same.
That's really troubling to me; whenever evidence is based at least in part on a person's judgement, I want to see blinded testing to assure that the person's judgment is known to be accurate. Do we have any examples of forensic audio experts submitting to blind tests of their judgment abilities?

The current state of the art, at least in the academic research, has nothing to do with what Owens was talking about, at least from following the trails of links from that survey.

None of this has any relevance at all if we are talking about biometrics. One of the questions that has to be answered is whether biometrics works well enough to use as evidence in a court of law. If it does, then Owens is irrelevant. If it doesn't, then what he says makes sense.

 

[totoro]

But that's the Thomas J. Owen who didn't have a high-profile case to pimp his company over.

And who appears not to know or care that biometrics exists. This seems to boil down to which set of methods works, which is the state of the art, and which will be accepted in a court of law.

That's the big question that none of us here can answer with our current knowledge. I'm not sure either of the purported experts being cited on both sides can answer this question satisfactorily, either, which is kind of depressing to me. Certainly nothing in the CV of either of them would indicate that they are capable of understanding the math involved in order to understand how the models work.

 

[WildCat]

This wouldn't be true in other methods of identification like DNA, blood type or fingerprinting; what makes you think it is true wrt to voice analysis? Finding the samples to be incompatible is not the same as being unable to make a match.

It would be obvious if you didn't have DNA to sample. But with sound, any sound will do. We already know that for 2 sound samples to be compared they have to be similar, reciting "Mary had a little lamb" for example. This guy compared a scream for help to a normal speaking voice not saying the word "help" in any way, shape, or form. Of course they won't match! They should never have been compared in the first place, but Owen used them anyway. The same Owen who is on record as saying the comparisons have to be of the same phrase.