Listen very carefully (Image: Gilles Coulon/Tendance Floue)
NOW your phone knows you better than ever. The latest version of Apple’s mobile operating system learns what your voice sounds like, and can identify you when you speak to Siri, ignoring other voices that try to butt in.
Siri, the intelligent personal assistant, is not the only one who knows your voice. As learning software improves, voice-identification systems have started to creep into everyday life, from smartphones to police stations to bank call centres. More are probably on the way. In a paper published at the end of September, researchers at Google unveiled an artificial neural network that could verify the identity of a speaker saying “OK Google” with an error rate of 2 per cent.
Voice is a “physiological phenomenon” shaped by your physical characteristics and the languages you speak, says Roger Moore at the University of Sheffield in the UK. A passphrase such as “Hey Siri” or “OK Google” is a powerful way to verify that you are who you say you are, he adds.
“My voice is different from your voice, which is different from your mother’s voice, which is different from someone on the far side of the world,” Moore says. “The latest machine-learning techniques can tease apart the tiny differences.”
For machines, recognising individual voices is different from understanding what they are saying. The recognition software has been fuelled by massive sets of vocal data built into a huge model of how people speak. This allows measurements of how much a person’s voice deviates from that of the overall population, which is the key to verifying a person’s identity. Changes to someone’s voice due to sickness or stress can throw off the software.
The technology is already being used in criminal investigations. Last year, when journalist James Foley was beheaded, apparently by ISIS, police used it to compare the killer’s voice with that of a list of possible suspects. And the banks JP Morgan and Wells Fargo have reportedly started using voice biometrics to figure out whether people calling their helplines are scam artists.
Your voice doesn’t just give away who you are, but what you’re like and what you’re doing, says Rita Singh at Carnegie Mellon University in Pittsburgh, Pennsylvania. “Your speech is like your fingerprints or your DNA.”
“Your voice doesn’t just give away who you are, but what you’re like and what you’re doing”
Singh is figuring out how to build profiles of a stranger from audio recordings. A voiceprint gives insight into the speaker’s height and weight, their demographic background, and even what their environment is like. She is working with doctors in Massachusetts and Ohio to detect a person’s likely diseases or psychological state through voice analysis.
Having devices in the home that recognise voices does raise security concerns, especially if they understand what you’re saying. Speech and voice algorithms often aren’t embedded in the device itself; instead, what you say is sent to a server somewhere else for analysis, and then ported back quickly. For example, Samsung fell into hot water this year with the revelation that its smart TVs could record private conversations.
“There are privacy concerns everywhere,” says Singh. “There is no device out there that ensures privacy.”
This article appeared in print under the headline “OK computer, who am I?”
No comments:
Post a Comment