William Steed writes:
On CSI/NCIS/SVU and the other acronym cop shows, you can compare a voice sample with the voice of a criminal. The sound waves match up perfectly, and the geeky, yet still somehow spunky technician says “The voiceprints match. We’ve got our perp!” Easy as that.
Well, no. Not really. Most people know these days that you can’t just zoom in on a blurry picture from a dodgy security camera and apply some nifty programming to it so that it clarifies and you can see the criminal in the reflection of someone’s sunglasses. But do they realise that the same thing applies to voice-matching?
How does it really work, then? Unfortunately for the scriptwriters, realism is not kind to plot development in a good murder mystery. Good forensic speaker identification involves a lot of boring fingerwork and database matching. In order to tell if a recording of a suspect is likely to be the same person as the recordings of a criminal, what has to happen is a lot of comparison, among other things. The aim is to find out what can be compared between the recordings. That is, what are the features that are the same? Do they both have the same odd speech impediment? Or do they both have similar vowel pronunciation? But that’s not all. They have to be able to say how many other people in the population it could be. How many other people have that same speech impediment? Or those same vowels?
What’s more, they can’t conclusively say whether the recordings are the same people or not. That wouldn’t be acceptable in court. What they say is how likely it is to find those features in the recording if they are the same person, compared to how likely it is that you’d get the same results from a random person off the street. The ideas behind all this aren’t so well understood yet by lawyers and judges, let alone jurors. Forensic voice analysis is only a young discipline – only some 15 years old. Once the process is better understood by the general public, there will be more chance to use it well in the legal system.
It all seems rather vague, right? But it’s pretty much the same process used on DNA evidence. Ask a biologist about the realism of how cop shows handle the processing of DNA evidence sometime.
Will Steed is a geeky yet still somehow spunky phonetician.