Amazon Improves Alexa Wake Word Accuracy To Reduce Unwanted AI Interruptions

Amazon wants to make sure that its digital assistant only springs to life when it is actually summoned, versus misinterpreting a word or phrase for "Alexa." To make that possible, Amazon in introducing a new feature called Cloud-Based Wake Word Verification. Just as it sounds, the new feature is intended to improve the accuracy of its wake word detection scheme to reduce or eliminate false wakes caused by similar sounding words.

"With this update, the wake word engine (WWE) on the device handles the initial detection of 'Alexa', and then a secondary cloud-based check verifies the utterance. If a false wake word is detected, the verification process directs the device to close the audio stream and turn off the LED indicator," Amazon explains in its developer blog.

Amazon Echo

As it stands, some third-party devices only do a single check to see if someone said "Alexa." Amazon's new Cloud-Based Wake Word Verification allows developers to program their devices to upload the audio to Amazon's servers for further analysis to make to improve upon the detection of the wake word. Granted, there are not many words we can think of that can be confused with "Alexa," but there are certain phrases that might trip up devices. This might also be helpful for bi-lingual homes.

This is not something that Echo devices owners have had to worry about. That's because Amazon's own hardware already uses a similar system to properly detect when a wake word is muttered. However, all Alexa products (first-party and third-party) are susceptible to unwanted wake-ups that can result from TV commercials, which was a problem for Google's competing Google Home smart speaker during the Super Bowl. It also became an issue when Burger King thought it was a good idea to air an ad that intentionally tried to wake Google Home devices.

One of the first third-party devices to use Amazon's new cloud-based verification is the Ecobee4, a smart thermostat that can listen and respond to voice commands.