Forensic Audio

christian-joudrey-65080 16x9.png

This episode was written & produced by Miellyn Fitzwater Barrows.

You might not realize it, but audio can be just as crucial to solving a case than video or eyewitness testimony. But understanding how to interpret audio evidence, or having the ability to enhance it to a point of intelligibility, requires highly-specialized training and expensive software. Meet Kent Gibson, one of the country's leading forensic audio experts, who's done audio analysis for the FBI, the Secret Service, and the Department of Homeland Security, among many others, and get his take on audio evidence from the missing Malaysian Airlines flight, the Trayvon Martin case, and Mel Gibson's infamous domestic dispute.

Music featured in this episode

We All Speak in Poems by Alaskan Tapes
Washedway by Evolv
In Which Way by Aeuria
Light + Breath by Dustin Lau
Where I Last Saw You by Dustin Lau
Timeline by Blake Ewing
Carved In Mayhem by Luke Atencio
Islands by Blake Ewing
Brilliant by One Hundred Years

Twenty Thousand Hertz is produced out of the studios of Defacto Sound and hosted by Dallas Taylor.

Follow the show on Twitter & Facebook. Our website is 20k.org.

Consider supporting the show at donate.20k.org.

Sign up for Audible here.

Sign up for Squarespace and use the promo code “20k”

View Transcript ▶︎

[Music start]

From Defacto Sound, you're listening to Twenty Thousand Hertz... The stories behind the world's most recognizable and interesting sounds. I'm Dallas Taylor. This is the story of how audio can be a key to solving a crime.

[Malaysia Airlines Flight 370 clip]

Forensic audio is the use of audio evidence in the investigation of a crime or for use in a court of law...and the clip you just heard was from Malaysia Airlines Flight 370. The one that went missing in March 2014 somewhere near the southern Indian Ocean.

Investigators found some debris in January 2017 that they have identified as being part of the plane, but to this day, no one really knows what happened.

When these final recording between the pilots of the missing Malaysian jet and their air traffic controllers were released, and appeared to have been edited, NBC news turned to Kent Gibson for analysis.

Kent: I'm a forensic audio and video examiner. I’ve been in the business about 30 years. I do authentication, which is, is this recording edited or inauthentic in any way?

He determined that the audio had been edited in at least three places. Listen carefully to the following clips for words that have been cut off.

[Malaysia Airlines Flight 370 clip]

He was also convinced that this section has been recorded through a speaker or microphone.

[Malaysia Airlines Flight 370 clip]

The tape may have come from multiple sources. Listen to it back to back with another section of the tape…

[Malaysia Airlines Flight 370 clip]

And now the part that sounds like it was recorded through a speaker...

[Malaysia Airlines Flight 370 clip]

It’s entirely possible that this is either a case of the Malaysian authorities editing the tape because parts of it contained information that the government didn’t want shared with the world, or that the tape was poorly put together from multiple recordings of the same thing.

Regardless, maybe this evidence, along with other findings, will someday help add to the bigger picture of the case.

Kent does a lot of authentication cases like this one. Some of this work relies on tech and some on good old-fashioned listening.

Kent: There are ways to find out and there are a lot of software ways, there are a lot of critical listening for non sequiturs and split words.

There is some new issues with authentication in that there are some techniques and some editing systems where you can't really detect an edit. These days you can't really say, "I guarantee you there are no edits here." All you can say is, "I don't hear any edits there." If you can say, "There is an edit right here," then it's not authentic. Then you're clear on that issue. The opposite is not the case.

If you ever hear a half of word, it's pretty clear that's an edit because it's really hard for a human voice to make half a word.

Another important service Kent provides is enhancement.

Kent: Enhancement is probably 50% of what most forensic audio people do. Making things understandable, that are not understandable.

Like this recording of a bribery sting with a television blaring in the background.

[bribery clip]

He managed to get a copy of the television broadcast on its own and using that clean audio, he could then eliminate it from the recording.

[bribery clip cleaned up]

How exactly does he do that?

Kent: By now I have a whole array of software that treats, it filters, it's limiters, it's different kinds of filters that are very specific, and that try to remove background noise. You sample a recording and it learns the background noise and then you output it at the end and the output is theoretically the voice without the background noise.

Here are two of those clips back to back…

[bribery clips]

But this doesn’t work for everything so sometimes Kent and others who do this kind of work have to manage expectations.

Kent: Especially with digital distortion and digital corruption, my famous line is not every file can be enhanced.

But plenty of recordings can be enhanced...or with a little bit of pre-planning, investigators can set up their own recordings to catch conversations between people they’re holding in jail.

Kent: Well sometimes the CI, who's the confidential informant wears a wire, but in this case there was a grate, an air conditioning grate, in the cell. They put the recorder behind the grate and then put the two guys in there.

They talked to each other, and basically they were claiming they didn't even know each other and we got them saying, "Where'd you hide the gun?" Basically it was clear that they were both in cahoots.

A quick aside here about recording conversations, in California it’s legal to record someone without their knowledge it if could provide evidence of a crime.

Unless they are somewhere where it’s posted that they are being recorded, like in a jail or a prison. This law varies from state to state.

Back to Kent.

Kent: I've done a lot of those, basically you can whisper for only about 20 minutes. Then after you're whispering for 20 minutes, you start talking with tonality in it. No matter how hard you try, you start speaking in something that can be understandable.

Kent worked with one local jurisdiction to take this idea a step further in their holding room.

Kent: Every jail has a room where people are admitted. They're arrested and they're put in this holding room while other things happen. I had them install this fan in that room, which was very noisy but only in the low frequencies. It was like a rumble, rumble, rumble. So we could record people in there, and they hear this noise in the air, so they don't think they can be heard, but we would record the noise then take out the rumble and hear exactly what they were saying.

That take preparation, sometimes you can do that when there's a fan in there anyway, but we were pretty sharp and thought to install things to make it easier for me to do my work.

We don’t have the exact same hear that Kent does, but we are able to reduce a consistent noise, so we tried this in the studio.

Person 1 Exmaple: Here I am talking with a loud fan in the background. As you can tell, it’s a little hard to hear, but when we process out the sound of the fan you can now hear me a little better.

Now, let’s hear that again, with the sound of the fan reduced.

Person 1 Example: Here I am talking with a loud fan in the background. As you can tell, it’s a little hard to hear, but when we process out the sound of the fan you can now hear me a little better.

This can be done fairly easily if the sound is consistent because the software can identify just those frequencies and reduce them. It’s much more difficult to reduce sounds that change frequency like a car engine [car engine SFX] or other people talking [people talking SFX].

Part of what makes it difficult to remove voices, the individual and unpredictable movement of the sound, is what makes it possible to identify them. Everyone uses their voices in different ways and that, coupled with the actual vibrations of their vocal cords, can be really unique.

Kent: I do a lot of voice identification, which is this voice is the same as that voice. The person on this recording is the same as the one over there.

Kent used this analysis during the Trayvon Martin case in 2012.

Kent: I wasn't the head guy but I was one of them. It was voice identification and basically my findings were different from the one that they had.

Find out how and go behind the scenes of another famous case...next.

[music out]


[music in]

Kent Gibson analyzes audio for criminal and civil cases and when he was consulted by the media about a 911 call in the Trayvon Martin case, his opinion was different than the one presented by another expert in court.

And, a quick warning here. We’ll be playing a portion of that 911 call in a moment. It could be disturbing to some listeners. There’s not any gunshots, but there is a lot of yelling in the background. If you might be sensitive or triggered in any way, I suggest skipping ahead to the fifteen minute mark in the show.

Kent: I get a lot of calls from news and news will come in and say, "Well there's this case and is that true?" I will often give them an opinion, but it's not necessarily for the legal counsel for that particular person.

The man charged with Martin’s murder, George Zimmerman, claimed self defense and one piece of evidence was a 911 call placed by a concerned neighbor.

There was shouting in the background, but there was a debate over whether it was Martin or Zimmerman.

Kent: It used to be that it was done by spectrograms and formants. Basically if you record a voice and look at it on a spectrogram they're these bars, horizontal bars, in the spectrogram and they're called formants, and they represent the vocal folds, your voice will have a specific characteristic of its formants that no other voice will have in theory.

It used to be we had to have the exemplars, which is a recording of the person in question saying the same words. If the guy said a certain thing, you had to get a recording of the exemplar saying the exact same words, but now software's able to separate the vowels so we can do it with any words just using vowels.

That means that now they can use any sample of the suspect’s voice to see if it’s a match. This new technology is pretty amazing, but it still isn’t perfect.

Kent: You have to have 11 seconds at least of the test and then you get an exemplar, which is a recording of the person in question, and then you run it through this very expensive and highfalutin software that gives you a return on whether or not the program thinks it's the same person, and it's always in a percentage. In forensics and most things like that, it's never 100%.

It's often a much lower percentage and you have to know what the benchmarks are to be able to say, "This is this, or this is not that." It was a news source who wanted me to opine about whether or not the voice that was recorded was Trayvon's voice.

Kent: Some other forensic person had opined that it was his voice and my opinion was that there wasn't enough of the recording to make that determination. It wasn't that it was contrary, it was just it's not enough information.

Who was telling paints dramatically different pictures of what happened. If it was Zimmerman yelling, was it because he was in fear for his life? Or was it Martin yelling because he was being threatened by the strange man who had been following him with a gun?

All we know for sure is that Trayvon Martin was shot dead by George Zimmerman and Zimmerman was acquitted of the murder after he claimed self defense.

Sometimes it’s more difficult to get at the truth...and others it’s a little easier.

Kent: I end up doing one that was for Osanka Grigorieva, who was the girlfriend of Mel Gibson.

As we've heard about Mr. Gibson, he was not good at controlling his anger. He was calling her to tell her to watch out that he was going to do her in. He would call and threaten her with a lot of different things.

She had recorded some of these phone calls and presented them along with voice messages as evidence of Mel Gibson’s physical and verbal abuse in criminal court.

Kent: I was hired on that case to show that A, it was Mel Gibson's voice and B, it was not edited. So that they did not make an inauthentic file putting together the threats.

I can’t play this example for you, even if it weren’t exclusively owned by Radar Online, the language is very offensive.

Kent: If we prove that it was him, and that it was not edited, then clearly he did it.

Once my testimony came out, he gave up. He said, "Okay. You can’t refute that."

Mel Gibson plead guilty to misdemeanor battery. He didn’t serve any jail time.

Kent: I get a lot of cases from famous celebrities and famous politicians where it's very classified and you can't tell who it was. The first time you leak something like that is going to be your last job. It's like being an accountant, you can't mess up.

Forensic audio is an imperfect science, but a necessary one. And people like Kent take that job very seriously as they lend their expertise and software to court cases around the world.

Twenty Thousand Hertz is presented by Defacto Sound, a sound team dedicated to making the world sound better. Whether it’s a commercial, television show, web video, trailer, video game, documentary, VR, or even physical products, Defacto makes it sound insanely cool. Check out prior work and get in touch at Defacto Sound dot com.

This episode was produced by Miellyn Fitzwater Barrows. ..and me, Dallas Taylor. With help from Sam Schneble. It was mixed by Jai Berger.

Thanks to Kent Gibson. Check out his website at forensicaudio dot org to find more incredible examples.

Our soundtrack is from our friends at Musicbed. Who offer a highly curated catalog from great indie artists and composers. Like what you hear? Listen to all of the songs from our show and even license them for your own projects at music.20k.org.

As soon as this podcast is over, go connect with us on Twitter or Facebook! If you have a few minutes, say hello. Also, while you’re at it, tell a friend about the show!

If you have any show suggestions, partnership opportunities, or if you’d like to advertise on the show, reach out through our website at 20k.org.

You can find all of these links on our website or in the show description.

Thanks for listening.


Recent Episodes