Engineering news
Amazon’s digital assistant may soon start recognising different voices. At the moment, neither Alexa nor Google Assistant can interact with more than one person, and they are connected to a single user account at a time.
Amazon hasn’t yet announced when the new feature will be launched, and has declined to comment. But Time reports that since 2015 the company has been working on Voice ID, a way to distinguish between Alexa users’ voices and correctly identify which member of a household is issuing commands.
Echo owners are already able to set up multiple profiles, but to jump between them, they have to say "switch accounts," or do so on the Alexa app.
For its part, Google told The Verge last October that it was working to support multiple accounts. It’s not clear when either effort will come to fruition, though.
Once that happens, “different people would have different preferences and different privileges, such as, for example, the music we listen to and the credit cards we might have entered into the system,” says Alex Rudnicky, computer scientist at Carnegie Mellon University. “There might be other benefits, for example people being able to leave messages for each other with the recipient implicit in the content.”
To get there, Alexa will try to match the voice of a speaker with a sample called a ‘voice print’.
The feature will improve privacy and security of the digital assistants – for instance, in terms of online payments authorisations. “There was a recent leak of data from a web-based teddy bear with children’s messages to their family, which is certainly an issue with any amount of archiving of user's speech,” says Alan Black, a language technologies specialist, also of Carnegie Mellon.
“The companies have to ensure that the data is encoded safely so no one can access it. Often this can be done by with both standard encryption techniques, but also by converting the data to update their machine learning models such that it is impossible to convert if back into understandable speech.”
Privacy concerns
Whether the voice samples are encrypted or not, the system will also have to store a lot of data about the users, “which might not be so great,” adds Rudnicky. Also, he adds, “one clearly useful thing would be for systems to have situational awareness, i.e. of what’s going on in their environment and use this information to improvement understanding of human actions and intents.”
One company that already has such a technology is Nuance Communications, whose systems are used by hundreds of large organisations such as Barclays, HSBC, Citi and others. “Nuance voice biometrics technology can recognise a person by their unique voice, ensuring that every interaction begins with a positive experience by eliminating the need to use PINs, passwords and security questions to log in to devices, call centres, apps, and websites,” says Vlad Sejnoha, the firm’s CTO. “Nuance's voice biometrics is offered as part of our smart home and automotive platforms, and is integrated as part of our wake-up word solutions for these experiences.”
When it comes to the expanding array of IoT and smart home devices, adds Sejnoha, biometric technology makes sense because “it offers a convenient and natural way to personalise and secure access to a device, car or system based on who you are - particularly as many of these devices are shared within a household”.
Voice biometrics allows for a conversational experience that starts by waking up a device so it knows who to respond to, he says, and “from there, engages in a more personalised and secure interaction, whether its accessing apps, music and TV content, home utility and security systems, or other services”.