Google Assistant Can Better Tell Apart Voices After Enhancing Set Up Process
Google Assistant will now be able to identify a speaker more accurately than ever before. Google has augmented the Voice Match feature of the AI by adjusting and expanding the setup process to require full sentences from a user.
Google Knows You
Voice Match creates a kind of audio log-in for Google Assistant that enables the AI to distinguish between different members of a household. The voice assistant can then essentially operate multiple accounts out of a central household account and personalize its responses to music requests, calendar reminders, and other details specific to that person. Google Assistant has included the Voice Match feature for a few years, but the update makes voice profile creation more elaborate.
“Now, when you set up Voice Match, Google Assistant will prompt you to say full phrases instead of just the hotword ‘Hey Google,'” the company explained in a blog post. “For example, during Voice Match set up, the Assistant will ask you to say “Hey Google, play my workout playlist” so it can better identify who is engaging with significantly higher accuracy.”
By using multiple, longer phrases as a baseline, Google Assistant will have a more robust database to draw on in identifying someone when they speak. The change in the setup process doesn’t alter how it works, just how well it can pick out a speaker based on their voice. Additionally, Google Assistant can remember up to six voice profiles per device, two more than it could before, although it’s not clear if that’s because of the improved accuracy.
In the same announcement, Google confirmed that it is adding a “hotword sensitivity” control to Google Assistant, as discovered in the software earlier this week. The slider aims to reduce accidental awakenings of the voice assistant by adjusting how easily alerted the AI is to potential wake words. As around two-thirds of voice assistant users have accidentally awakened a voice assistant over a month, the control could be very popular.
The issue of privacy ties together both updates. Most people don’t like the idea of being overheard, even by accident, and limiting the times when they mistakenly turn on the voice assistant might ease people’s suspicions about how their recordings are used. The same goes for the voice ID accuracy update. Deepfake voice fraud is not at all common right now, but fear of someone stealing vocal identification is linked to slower adoption of the concept. Biometric security and technology startup ID R&D recently conducted a survey that found two of three adults in the United States are concerned that someone could mimic them well enough to illegally access accounts linked to their voice. Even small improvements in vocal security hotword sensitivity and enhanced Voice Match setups might make people feel better about signing up to use a voice assistant, a goal shared by Google and its rivals.