Art manager

Automatic speaker recognition technology outperforms human listeners

A key issue in a number of court cases is whether a speaker on an audio recording is a particular known speaker, for example, whether a speaker on a recording of an intercepted telephone call is the defendant.

In most English-speaking countries, expert testimony is only admissible in court if it is likely to help the judge or jury reach a decision. If the identification of the speaker by the judge or jury was as accurate or more accurate than the forensic voice comparison of a forensic scientist, then the forensic voice comparison testimony would not be admissible. .

In a research article “Speaker identification in courtroom contexts – Part I”, recently published in the journal International Forensic Sciencesan international multidisciplinary team of researchers has reported the first set of results from an extensive study that compares the accuracy of speaker identification by individual listeners (such as judges or jury members) with the accuracy of a medical system -forensic voice comparison based on state-of-the-art automatic speaker recognition technology, using recordings that reflect the conditions of a real case.

The interviewed speaker’s recording was of a telephone call with office background noise, and the known speaker’s recording was of a police interview conducted in an echo room with voice system background noise. ventilation.

The forensic voice comparison system performed better than all 226 listeners that were tested.

The research team was made up of forensic data scientists, lawyers, experimental psychologists and phoneticians, based in the UK, Australia and Chile.

Corresponding author Dr Geoffrey Stewart Morrison, Director of the Forensic Data Science Laboratory at Aston University, said: “A few years ago when I was giving evidence in a court case, a lawyer asked me why the judge couldn’t just listen to the tapes and make a decision. Wouldn’t the judge do better than the forensic voice-comparison system I had been using? It was the spark that [led] it is up to us to carry out this research. I expected our forensic voice comparison system to perform better than most listeners, but I was surprised when it actually performed better than all of them. I’m glad we now have such a clear answer to the question posed by counsel.

Dr Kristy A Martire, Contributing Author from the School of Psychology at the University of New South Wales, said: ‘Past experiences where we have been able to recognize familiar speakers, such as family members or friends, can lead us to believe that we are better at identifying unknown voices than we really are. This study shows that regardless of a listener’s ability to recognize familiar speakers, their ability to identify unfamiliar speakers is unlikely to outperform a forensic voice comparison system.

Contributing author Professor Gary Edmond, University of New South Wales School of Law, said: “Unmistakable scientific findings show that the identification of unknown speakers by listeners is surprisingly difficult and far more error-prone than the judges and others appreciated. We must not encourage or allow non-experts, including judges and jurors, to engage in unduly error-prone speaker identification. Instead, we should seek the services of real experts: specialized forensic scientists who use empirically validated and proven forensic voice comparison systems.

– This press release was originally published on the University of Aston website