At its Alexa event on Thursday, Amazon unveiled the AZ1 Neural Edge processor that it said would significantly improve Alexa’s response time and speech-recognition abilities. Built in collaboration with Taiwanese fabless semiconductor major, MediaTek, the chip is built into the Echo smart speaker, Echo Dot and Echo Show 10 that the company announced last night. It will also be found in more Amazon hardware in the coming days, the company said.
Alongside the new chip, Amazon also announced a slew of new Alexa features, including better on-device neural speech recognition. There’s also a new kid-friendly mode, the ability to detect crying babies and a more natural sounding, conversational voice. The company also announced updates across its entire portfolio of Alexa developer tools and frameworks.
One of the most notable upcoming Alexa features is its ability to ask questions. According to Amazon, its voice assistant will now ask questions to better understand some commands. The questions will initially to limited to smart home actions, like switching lights on/off, before expanding to other domains. The company also announced ‘Alexa Voice Profiles for Kids‘, which automatically recognizes a child’s voice and switches to a kid-friendly mode.
Another notable new feature is ‘Natural Turn Taking’, which will let multiple people join conversations with Alexa without having to use a trigger word every time. To do that, Alexa will apparently use acoustic, linguistic, and even visual cues to detect whether somebody is trying to issue a command.
As for Amazon’s sound detection algorithm, it will now be able to detect crying babies, barking dogs and well, snoring humans, similar to iOS 14’s Sound Recognition feature. Amazon will also soon roll out Alexa Routines that can kick off when Alexa detects any of those sounds. While there’s no exact ETA for that, Amazon says that it will happen before the end of this year.
