The Law is Not Science: The Validity of Voice Identification

Peter Ladefoged - oldfogey@ucla.edu
Phonetics Lab, Linguistics, UCLA
Los Angeles, CA 90095-1543

Popular version of paper 3pAA3
Presented Wednesday Afternoon, November 12, 2003
146th ASA Meeting, Austin, TX

Courts often need expert opinion as to whether two recordings were made by the same speaker or not. But the evidence in a case is not like data in an experiment. Scientists are used to gathering data in controlled situations and then drawing inferences from it. But every court case is like an experiment in which there are only one or two observations, made under unique circumstances. As a result, phoneticians, experts in the science of speech, do not know how to respond when they are asked to give evidence on speaker identification. They generally agree that it is possible for a trained observer to offer a valid professional opinion on the degree of similarity between two voices, each on a 15 minute studio recording. But they usually find it impossible to judge the degree of similarity between two voices on noisy three second bomb threats. Court cases can be like either of these situations or anywhere in between. There is no one-way of considering the data that can be applied in all cases. Forensic phoneticians are like medical doctors giving prognoses. They make many tests that provide useful clues, but their opinions are inevitably based on their own experience of similar situations rather than on rigorous scientific procedures. Like doctors, their opinions should certainly be noted. They have evidential value. But they are not established scientific truth.

In many court cases one voice has been recorded by an unknown speaker on a telephone answering machine and the other somehow obtained from a suspect. This is not always the case, however. On one occasion I was asked to offer an opinion as to whether Mr. L. Ron Hubbard, the founder of the Church of Scientology, was dead, as some members of his church submitted, or alive but in seclusion, as others maintained. (He was at the time being sought by the IRS in connection with possible tax evasion). The evidence in the case included the studio recordings of two Christmas messages he had sent to his followers, one from the current year and one from the previous year. Both referred to events of the time, and the earlier message was acknowledged to be by him. If the later message was made by the same voice it was evidence that he was alive at that time. Each message was about 15 minutes long, and contained many similar words and phrases. After careful listening and acoustic analysis it was clear to me that they were highly likely to have been made by the same speaker. (It turned out that Mr. Hubbard was in fact alive at the time. His death was reported some time later.)

When an expert gives an opinion in court everyone wants to know how reliable it is. There is no easy answer to this question; it depends on the circumstances. We will consider a few of the points that have to be taken into account.

1. Were the physical conditions similar? Both recordings may have been made over telephone lines, but these lines may have had different characteristics. In addition some home recording devices preserve much more of the characteristics of the speaker than others. Many identification questions concern 911 calls, which are recorded on a system that saves one whole day at a time at a comparatively low fidelity. I have one case in which a police officer allegedly got out of his squad car and made a 911 call from a pay phone reporting the presence of a man with a gun. This 911 call was his justification for later making an arrest. The identification of his voice on the 911 call was complicated by the low fidelity of the recording system.
2. Were both the speakers talking in the same style? People committing crimes may be more tense or agitated. When suspects provide samples of their voices they are often in quite a different mood. It is hard to compare two voices when one is scaring the life out of someone and the other is leaving an innocuous telephone message about feeding the dog, with no words or phrases in common with the threatening message.
3. Was one speaker trying for a disguise? Fortunately for us, most criminals are not very smart and forget that they are making a recording that might be used to identify them. When they do try to disguise their voices they seldom try to do so by imitating another person's voice so that an innocent man might become charged with an offence. In over 100 cases of comparing voices I have never come across this happening. A guilty person might go free because of a successful disguise, but it is most unlikely that a recording of one person's disguised voice will be sufficiently like that of another person's voice to be the cause of that innocent person being convicted. I might add that I have never encountered anyone trying to defend themselves by saying "That's not my voice, it's someone imitating me."
4. Was there any background noise? Studio recordings are rare in voice identification cases (though, living in Los Angeles, the entertainment capital of the world, I have been involved in a few cases in addition to the L. Ron Howard matter). The main problem in comparing most recordings of voices is the amount of noise on the recording. The speaker may not be talking directly into the phone (in one case I was asked to identify a murderer muttering while walking away from the dangling phone on which the victim had made a 911 call), or there may be noise from the busy street outside, or shouting bystanders. Recordings made by undercover police engaged in drug busts are often full of extraneous noises. They are seldom high enough quality to make an identification of the drug dealer (who usually makes only monosyllabic remarks).
5. How long was the recording? A five second bomb threat on a noisy phone line almost certainly does not have enough distinct information about the speaker. But even on a very short but clear recording there might be features that have evidential value. If the speaker had a very unusual voice even a word or two might be indicative of a possible suspect.
6. Does the suspect have family or friends with very similar voices? I have made recordings of my mother, my father and myself having a conversation in which even I, making a careful analysis, cannot say whether my father or I made a given remark. Our voices were almost identical. I also know of two cases in which a suspect was charged with making a telephone call, but in one case her sister confessed to making the call, and in the other a college roommate admitted that he was responsible. Very often in court cases the expert has no knowledge as to whether there are other people who might have made the recording and can only report that the two voices are very similar and highly likely to be the same individual or two close relatives or friends.

A scientist giving an opinion on whether two recordings are likely to have been made by the same speaker would like to do so by reference to recordings that were of the same quality and length, and were made by similar speakers in similar circumstances, saying similar things. But it would be almost impossible to set up a valid experiment that can take all these (and many other) variables into account. Lawyers often try to assist juries by asking for numbers indicating the degree of certainty of an identification. When I am asked how likely is it that I have made an error, I can only answer that I have never had a case exactly like this one, so I do not know. It would take enormous resources to do all the experimentation required to quantify the reliability of an opinion that any two recordings that might be presented in a court case were likely to have been made by the same speaker. We certainly cannot say that voice identification by experts should always be allowed or never be allowed. Sometimes expert opinion has evidential value, and sometimes it does not. Judges often have to make rulings about whether a given recording is acceptable as evidence, unfortunately there is no scientific data on which they can base their rulings because every case is different. We just have to rely on the court's good common sense. It's not science, but it's the law.