Trayvon Martin, Vigilante Justice

[Natural Born Skeptic]

Right.

And just what do you expect to learn about current technology from 40+ yr old textbook?

Edited by Tricky: 

Edited for civility.

Here is a book on Google that is someone dated (1979) that is nevertheless interesting simply because it gives insight into how the issue has progressed back and forth over the years *up to that time*.

p.s. I even added asterisks for you to make it easier for you to follow, no need to thank me.

 

[Dancing David]

Are worried he might be shot by some stranger who thought he was "suspicious", or that he might shoot someone and get convicted on the basis of speaker recognition software?

My son is 17, he wears hoodies and walks slowly when talking to his girlfriend.

Since there don't seem to be any good grounds for Martin being suspicious (other than the obvious elephant in the room, which for some reason we aren't talking about), I don't understand this. Any random person could have been targeted by Zimmerman (if we claim he wasn't racially motivated). If he was racially motivated, then things are uglier.

I believe that age is a factor is some people suspicions.

 

[totoro]

For those who are truly interested, the paper published here, by a real scientist (as opposed to a guy with a degree in history that took some side courses) who's life is devoted to the study of sound and speech issues and who has been published and cited repeatedly in scientific journals is quite good if you are willing to spend $35.

If you don't want to spend $35 you can find numerous papers he's written on everything from how our vocal chords developed over time to this paper titled Person Authentication by Voice, a Need for Caution.

If you are a pirate, you can get it for free, of course. It was an interesting read.

Note that the guy who did the tests on the scream doesn't present himself as a pioneer in the development of biometrics. He's a practitioner in the field. At this point, the field seems to have clearly ended up in the ee/cs machine learning domain. The guy isn't Andrew Ng, but the world wouldn't be a better place if people like Ng spent their time analyzing evidence for court cases.

This paper that someone else unearthed has a really nice summary of the state of the field:

http://cs.joensuu.fi/pages/tkinnu/webpage/pdf/speaker_recognition_overview.pdf

 

[Dancing David]

While I did not lump you in with the other three, you also are failing to consider the opinion was given by what appears to be one of the top forensic audio analysts in the country. Why are you implying you have legitimate challenges to his assessment that you don't believe he made just because the analysis is not spelled out for you in the news article? Would you be making the same kind of challenge if the top psychiatrist in the country was giving a forensic opinion or would you be wording it differently?


How about, "Owen may have to show [xyz] when he gets to court (if he were called to testify) but I think it's safe to assume he knows that"?


The audacity of the challenges here sound like people consider themselves to be equally qualified in this area. The newspaper probably couldn't have found a more qualified expert in the field than Owen. He's clearly recognized by his peers.

And the JREF is what it is , I have questions about the technique, they can be resolved easily. I know enough about data transmission and audio compression to have questions, they can be answered easily in testing.

 

[Dancing David]

I was under the impression the expert conclusion was it is not GZ yelling for help on the tape. Where is the level of doubt? Some forum expert wannabe's opinions they aren't satisfied?

None of the naysayers in this discussion are arguing in their fields of expertise, they are arguing about something they have a passing knowledge of.

And there are questions that need answers. If this is to be presented as evidence, they will be answered.

These are the same standards I apply to all claims, eliminate confounding factors, when that is satisfied then I will be. It may be they have been addressed already, so far we do not know.

Digital compression involves sampling, the type of phone matters a great deal as does the data compression on the audio storage, and the means of transmission. When the technique has error bars for the given equipment and situation, then we will know.

ETA: I do want to say that I do not believe that this possible evidence is fraudulent or woo in any way. I just have questions that can be resolved very easily and would likely be raised in a court of law. I do not feel that Owens is a shyster, until I have proof of that. I believe that he does have his tools and that they are likely valid within certain parameters.

My instructors do discuss however the possible errors of biometrics, especially vocal recognition.

 

[totoro]

Edited by Tricky: 

Edited for response to modded post.

I didn't say we would learn anything about current technology from it.

I skimmed it. It was antique stuff. Who cares? Have your read:
http://cs.joensuu.fi/pages/tkinnu/webpage/pdf/speaker_recognition_overview.pdf?

If not, why not? It actually has a summary of where the field is now. Reading the book you posted won't actually help you understand where the field is now. That being the case, why would you read that book first?

This link that you posted:

http://www.afcp-parole.org/doc/AFCP_SpLC_HotTopicsEurospeech03_final.pdf

actually talks about the need for similar environments, etc. This is interesting, because the later papers are, to a great extent, about how to make models that automatically (or semi-automatically, depending on the model) adjust for known changes during training in order to be able to deal with unknown changes in later test data. And certainly later systems have been demonstrated to be able to detect a given person's voice even when the person is trying to disguise it.

 

[Dancing David]

OK, has anyone taken a close look at Thomas Owen's educational background? Want to know what his B.A. is in? History! Other than that, he's taken some courses from SunAudCon which is one of those organizations that offers courses that last a few days and then you get a certificate. Then in 1985 he took some type of a "voice identification course". i.e. he probably sat through a course for a few days and took a test. Then from there he turned himself into an "expert". BAHAHAHAH.

Some of you guys have no idea how easy it is to become an "expert".

Tell me this. How many climate scientists do you know who's degree is in History? Want to keep telling us being skeptical about this guy is like doubting climate scientists?

http://www.owlinvestigations.com/about.html

Now that doesn't matter as much as experience, my instructor this semester has a BS in Agribusiness and a MS in Agricultural Economics, yet he managed to become a high powered networking admin for a very large university.

 

[Natural Born Skeptic]

I skimmed it. It was antique stuff. Who cares? Have your read:
http://cs.joensuu.fi/pages/tkinnu/webpage/pdf/speaker_recognition_overview.pdf?

If not, why not? It actually has a summary of where the field is now. Reading the book you posted won't actually help you understand where the field is now. That being the case, why would you read that book first?

How do you know what book I read first? I've read numerous papers, am reading more, and have posted several links. The fact that you and Ginger keep harping on someone posting something that I merely found interesting and even stated its limitations indicates to me you mainly want to argue. Why would you even want to argue against me posting that?

 

[Natural Born Skeptic]

Now that doesn't matter as much as experience, my instructor this semester has a BS in Agribusiness and a MS in Agricultural Economics, yet he managed to become a high powered networking admin for a very large university.

I actually don't have a degree in the subject matter I'm an "expert" in either, and agree that experience is important too. However there are certain areas where I want people to be both credentialed and have experience. For instance, a medical doctor. Or a scientists developing drugs. Similarly, I want a person who's testimony is potentially putting people in jail, and determining guilt and innocence, especially if they are supposedly the best in their field, to be credentialed and have experience.

All the more so when dealing with the adage that extraordinary claims require extraordinarily proof. Well in this case, I find the claim that someone can identify that a voice heard screaming in the background of a 911 call is not the voice in another 911 call to be quite extraordinary.

Lastly, I'll gladly put aside that demand for credentials if said expert/s could demonstrate their accuracy with double blind tests. Absent that, I have to conclude that:
1. I don't believe the claims here but am not 100% certain I am correct.
2. I'm willing to change my mind if these guys could prove their accuracy. As I said earlier, I'd be willing to put up a considerable amount of money that they can't (using similar circumstances).

 

[curi0us]

Sorry for jumping in without having followed the full discussion but I am electronic/computer music geek and have a lot of hobbyist knowledge in the audio software realm.

For the audio analysis are the "experts" just comparing the publicly available 911 call that Zimmermen made to the overheard screams that could be heard on a different 911 call later that night?

 

[totoro]

How do you know what book I read first? I've read numerous papers, am reading more, and have posted several links. The fact that you and Ginger keep harping on someone posting something that I merely found interesting and even stated its limitations indicates to me you mainly want to argue. Why would you even want to argue against me posting that?

So you have read the survey? Cool, that probably makes you the only other person on this thread who has.

I guess the thing about that book is that it's really only of historical interest. The newer stuff is pretty complicated and would take years to get completely up to speed on, even if you already knew a lot about machine learning and/or signal processing. I'll probably read some more of the newer papers that are close to what I do for work, irrespective of this argument, so I owe you for bringing that survey into the discussion .

As far as this discussion goes, that older book just doesn't seem that relevant to me, especially when there are much more interesting things to read in the same field (that all seem to have been written in the last 15 years or so), and the software we are discussing is written using the newer techniques (or at least is advertised to do so).

 

[Natural Born Skeptic]

For the audio analysis are the "experts" just comparing the publicly available 911 call that Zimmermen made to the overheard screams that could be heard on a different 911 call later that night?

Yes.

 

[totoro]

I actually don't have a degree in the subject matter I'm an "expert" in either, and agree that experience is important too. However there are certain areas where I want people to be both credentialed and have experience. For instance, a medical doctor. Or a scientists developing drugs. Similarly, I want a person who's testimony is potentially putting people in jail, and determining guilt and innocence, especially if they are supposedly the best in their field, to be credentialed and have experience.

All the more so when dealing with the adage that extraordinary claims require extraordinarily proof. Well in this case, I find the claim that someone can identify that a voice heard screaming in the background of a 911 call is not the voice in another 911 call to be quite extraordinary.

Lastly, I'll gladly put aside that demand for credentials if said expert/s could demonstrate their accuracy with double blind tests. Absent that, I have to conclude that:
1. I don't believe the claims here but am not 100% certain I am correct.
2. I'm willing to change my mind if these guys could prove their accuracy. As I said earlier, I'd be willing to put up a considerable amount of money that they can't (using similar circumstances).

The thing is that that field seems to have changed a lot. The current state of the art is machine learning, which didn't even really exist when these guys went to school.

They're basically practitioners, who know how to use the best software available, but from what I've seen of their listed papers, neither of them is writing new machine learning based speaker recognition software.

 

[Natural Born Skeptic]

So you have read the survey? Cool, that probably makes you the only other person on this thread who has.

I guess the thing about that book is that it's really only of historical interest. The newer stuff is pretty complicated and would take years to get completely up to speed on, even if you already knew a lot about machine learning and/or signal processing. I'll probably read some more of the newer papers that are close to what I do for work, irrespective of this argument, so I owe you for bringing that survey into the discussion .

As far as this discussion goes, that older book just doesn't seem that relevant to me, especially when there are much more interesting things to read in the same field (that all seem to have been written in the last 15 years or so), and the software we are discussing is written using the newer techniques (or at least is advertised to do so).

Fair enough .

Like I said early on, part of the reason for my extreme skepticism here is the field I work in and having worked with a lot of audio engineers over the years, and seeing how much woo is involved and how incredibly subjective it is. I've also been called as an expert witness (not for voice identification and not for a criminal trial) and was rather amazed by the entire process. Being an "expert" is BIG MONEY, so I know these guys have a huge financial motivation. That does not mean in and of itself that they are not to be believed, but seeing how easy it is to find an expert to say what you want increases ones skepticism which is why I place a lot more value in what scientists say than someone who makes their living testifying about this.

Anyone who says "they are not testifying and therefore have no financial interest" is incredibly naive. These guys will now do the press rounds, that will be added to their credentials, and they will get more lawyers calling them because of this. This press means hundreds of thousands of dollars in new business for them, it's a dream come true for an expert witness to get this type of press.

 

[totoro]

The stand you ground statute means that if someone is attacked in public, they do not need to flee before using deadly force, when they meet force with force. Before, someone was required to run away or avoid the confrontation first, and this law removed that requirement.

There is already a stipulation in the FL SS that says that as the aggressor, you can use deadly force when meeting force with force, but ONLY if you try to break off the altercation before hand. Meaning, the original aggressor would HAVE to, by law, try to run away, or back away from the fight, before they could use deadly force.

So, IF Zimmerman was the aggressor, he would have to meet force with force, and also try to escape or stop the confrontation before using force.

He has no right to stand his ground IF he is the aggressor.

IF he was not, then he would have to have been in fear of his life and meet force with force.

Hope that helps. If you want the actual statutes, let me know, and I will post links.

Thanks, that was informative.

 

[curi0us]

Yes.

Thank you.

Unless that expert software has proven itself in double blind tests then declaring anything conclusive at this point seems premature. Instead of comparing Zimmermen's calm 911 call with a distant screaming voice someone should get Zimmermen to do an audio reenactment in the same spot as he was in when the shooting happened and then record it onto the 911 system via the same 911 caller's phone and location. Then there would be two recordings with virtually identical set ups and a significantly more informed comparison could be made. I am no expert, but what I just described seems like Audio Forensics 101 to me.

 

[totoro]

Fair enough .

Like I said early on, part of the reason for my extreme skepticism here is the field I work in and having worked with a lot of audio engineers over the years, and seeing how much woo is involved and how incredibly subjective it is. I've also been called as an expert witness (not for voice identification and not for a criminal trial) and was rather amazed by the entire process. Being an "expert" is BIG MONEY, so I know these guys have a huge financial motivation. That does not mean in and of itself that they are not to be believed, but seeing how easy it is to find an expert to say what you want increases ones skepticism which is why I place a lot more value in what scientists say than someone who makes their living testifying about this.

Anyone who says "they are not testifying and therefore have no financial interest" is incredibly naive. These guys will now do the press rounds, that will be added to their credentials, and they will get more lawyers calling them because of this. This press means hundreds of thousands of dollars in new business for them, it's a dream come true for an expert witness to get this type of press.

Yeah, I can see being a bit suspicious. And neither of the guys is exactly a professor of machine learning at Stanford or CMU. But they are most likely pretty good at using whatever the current best software is.

The current research honestly does look pretty legit, and even has its own Nist/DARPA bake off. If it comes down to a trial, there will be guys on both sides, so hopefully whatever technology gets used, any misapplications will be pointed out by the other side.

The idea that some people have on this thread that we should preemptively declare them wrong on the basis of no knowledge seems wrong headed to me, though. This is the kind of thing that should be contested by another expert at trial.

 

[crimresearch]

I doubt seriously if any judge would make the rookie mistake of conflating the entire discipline with every new development within it.

The Frye standard doesn't depend on what was accepted in 1974, it depends on whether or not Easy Voice Biometrics software (or whatever makes it unique) is widely accepted in the field today.

The discipline can be rock solid, the *old* methods can be reliable, and the new process can be an amazing breakthrough... but if it is still 'cutting edge', it might never get into a Florida court, no matter how good it is.

If it never makes it into evidence, then it will fall into the category of 'correct but not useful'... if it does make it in, then the game of dueling experts may take place.

Looks dramatic, but still too early to know for a fact.

 

[KatieG]

Originally Posted by triforcharity View Post
The stand you ground statute means that if someone is attacked in public, they do not need to flee before using deadly force, when they meet force with force. Before, someone was required to run away or avoid the confrontation first, and this law removed that requirement.

There is already a stipulation in the FL SS that says that as the aggressor, you can use deadly force when meeting force with force, but ONLY if you try to break off the altercation before hand. Meaning, the original aggressor would HAVE to, by law, try to run away, or back away from the fight, before they could use deadly force.

So, IF Zimmerman was the aggressor, he would have to meet force with force, and also try to escape or stop the confrontation before using force.

He has no right to stand his ground IF he is the aggressor.

IF he was not, then he would have to have been in fear of his life and meet force with force.

Hope that helps. If you want the actual statutes, let me know, and I will post links.

Seconded, very informative. Most of your posts have been explanations of details of the FL law. It isn't very cheerful but it's accurate.

 

[lane99]

Claims that don't pass the smell test

...Unless that expert software has proven itself in double blind tests...

They've been called for here, but I doubt any which would replicate what was analyzed in this specific event even exist.

Notwithstanding that, though, apparently the analyst/sales rep making the claim of "scientific certainty" has a very glitzy looking CV. And glossy business cards printed on premium stock.