Today we’ll be learning about forensic Voice IDs. This is a field that has existed for quite a long time but has never been accepted by any court using even the relatively minimal standard set by Frye in the 1920s. On the other hand, given the types of forensic techniques that are currently allowed in a significant number of jurisdictions, a lawyer would be surprised to learn that it’s still largely inadmissible.
Traditional/Earwitness Voice IDs
It is helpful to begin by looking at how voice identification and authentication is done traditionally during trial. A witness who is familiar with the voice somehow, either by knowing the defendant or being the victim of his acts, listens to the sample or to an exemplar. Then the witness, knowing the defendant’s voice, simply states whether they think it’s the defendant’s voice. This is sufficient to admit and authenticate any voice recording or ID.
Unfortunately, these kinds of IDs aren’t done just by parents, siblings, best friends, or spouses. They are done by victims of crimes very commonly (e.g. blackmail victims, someone who received a death threat over the phone, the victims of chatty rapists). Under these circumstances – being unfamiliar generally with the defendant, having a limited sample of his voice in their memory, and typically hearing that voice distorted – results in an ID which is not only suspicious but has proven to be inaccurate. Studies have shown that IDs under these circumstances result in a correct ID only about 80% of the time. Is that beyond a reasonable doubt? Who knows it’s a purposefully undefined standard. (Note: As with all types of evidence we’ve been discussing you are now allowed to call experts to attack the reliability and credibility of the techniques and witnesses so you could get this information to the jury. We’ll be going more into this when I address eyewitness testimony.) Given the obvious pitfalls of so-called “earwitness” voice IDs, it makes perfect sense that some would seek out a more reliable and scientific method of matching two voice samples.
It has gotten so dramatic that there have been cases where defendants have been required to speak the “criminal words” in front of the jury and earwitness in order for that witness to provide an ID. With enough explanation to the jury, this is perfectly allowable. Eat your heart out Law & Order.
Expert Voice IDs
Voice Spectrographs come to the rescue, sort of. Actually, they completely suck balls. They are poor resolution and there’s no real way to differentiate accurately between similar sounding voices. Regardless lets go through what they do.
The machines are boring and you’ll only see them used by linguistics professors and their graduate students. You speak into a microphone; it goes through a filter; then data are spit out on a recording device. The recording device is a laptop or a cool drum and ink recorder that looks like what comes out of seismographs.
The issues with the technique are, again, many. First, the spectrographs, even modern ones, are awful at producing a good resolution data set. Second, the entire ID is entirely in the head of a voice spectrogram examiner, who is following no real standards. The whole standards debate is so ridiculous that the forensic association split over whether they should have real, stringent standards, or effectively “none” and the “none” group took all the members and the original association dissolved. Finally, there’s just no significant scientific data supporting that intravoice deviation is greater than intervoice deviation. That is to say, there’s no data to support that the difference between my voice and your voice will be greater than the difference between two samples I give on two different days.
There have been several studies about the efficacy of the method – the best-case scenario is that it is useful in only ~40% of attempts. It can result in a false positive rate of 6%. The National Academy of the Sciences just outright said that there is no basis to accept it as a reliable forensic technique. Mind you, that was in the mid 70’s. The FBI dabbled in it during the 80’s and early 90’s and even they concluded it has no forensic merit but might sometimes be useful in investigations. Currently there is no reliable technique or standard for voice IDs through spectrographic analysis. Admissibility is generally disfavored but there have been a few cases that admitted it. The Alaska Supreme Court laughably stated that there was no scientific consensus on whether it could be used reliably, and in fact, its reliability was heavily disputed. Then they admitted it anyways. Whoops!
While this kind of ID is generally inadmissible at trial, it might be admissible for search warrants and arrest warrants – things that need only meet a probable cause standard.