Amazon is debuting its new Conversation Mode, which lets you talk to the Alexa assistant without having to repeat the wake word, for the Echo Show 10. Amazon's Alexa digital assistant can already combine multiple requests into one, such as "add milk, sausage, and butter to my groceries list," without having to repeat yourself for each item.
But of course, you still have to repeat the "Alexa" keyword every time. There are few other instances of a digital assistant responding to commands without saying the wake word. With Google Assistant, you can say "Stop" when a timer is going off without saying "OK Google, stop." And with Siri on Apple Watch, you can enable a feature that lets you immediately start talking to Siri without having to say "Hey Siri," but that only works on the initial command.
But none of those are quite as impressive as what Amazon is announcing today. The company announced via a press release that its enabling Conversation Mode on the Echo Show 10. Users can opt-in to the feature by saying "Alexa, join the conversation." Once that's enabled, anyone in the room can start talking to Alexa without saying the wake word. According to Amazon Science, the new feature uses a combination of visual device directedness detection (CVDD) and audio-based device voice activity detection (DVAD). In regards to CVDD, Amazon trained a deep neural network model to estimate the orientation of your head. Of course, in the real world, a way humans know they're being spoken to is by making eye contact and facing in a similar direction. "This approach reduced the false-rejection rate (FRR) for visual device directedness detection by almost 80% relative to the [ standard perspective-n-point] approach,” Amazon says. DVAD is pretty straightforward. Amazon processes the audio signal and determines if it's being directed towards the device and uses machine learning to determine whether or not to respond.
More Natural Than Ever Before
Amazon says it's doing all of this with the user's privacy in mind. As mentioned before, users will need to invite Alexa to join the conversation. Likewise, users are able to say, "Leave the conversation," or press the mic/camera off button to disable Conversation Mode. Amazon also says that the feature will automatically turn off if it doesn't detect a request "after a short period of time." Users will know when Conversation Mode is on as there will be a solid blue border around the screen. A light blue bar at the bottom indicates that requests are being sent to the cloud. Amazon reiterates that no images or videos will be sent to the cloud and that only audio clips that are directed at Alexa will be sent up to the cloud. Of course, users always have the option to delete voice recordings like before.
Unless there's a digital assistant out there that processes all voice commands on-device without an active connection to the internet, privacy will always be an issue. No matter how secure or encrypted the connection is, there's always a data breach risk. However, the new Conversation Mode from Amazon is certainly cool. Digital assistants have come a long way, however, having to repeat the wake word has always felt unnatural. You don't repeat someone's name every time in a real-life conversation, do you?
Source: Amazon, Amazon Science