How Amazon Alexa Isolates Wake Words

Alexa Wake Word Featured Image

When some people heard that employees of Amazon listen to what is recorded on Alexa, they assumed they can hear everything going on in their homes. This is not true. Alexa cannot record without using a wake word.

The Amazon Echo and other voice assistants that use wake words are only able to record any voice commands made after the wake word. In Amazon’s case, that word is Alexa. But how does the device recognize the word Alexa amid the other voices and background noises in our homes?

Also read: How to Use Amazon Alexa Devices as an Intercom

Three Technologies Working Together

There are three technologies in the Alexa system that, when they work together, enable the device to recognize the wake word. Nothing else will begin the recording. These technologies are the microphones in the machine, a three-second memory buffer, and neural net training.

Microphones

The Amazon Echo has a system of multiple microphones that allows it to pick up the sound of the wake word. These microphones can identify the wake word from among all the other sounds that are in the background. When the device hears the word, it begins recording, and you will see a dark blue ring of light on the top.

Alexa Wake Word Microphones

The microphones also determine the direction the word comes from so they can focus in that direction. When the device determines the direction, a lighter blue color light will follow the sound of your voice.

Limited memory

The memory built into the Echo device only retains three seconds worth of data at a time. Then the data is erased and written over. For those of you old enough to remember cassette tapes, it would be like recording for three seconds and then rewinding to the beginning and recording for three seconds again. Alexa does not transmit data until after it recognizes the wake word.

Neural net training

Without going into mind-blowing technical detail, a neural network is a series of artificial connections designed to work in a way similar to the human brain. Amazon uses this technology to train Alexa to recognize the wake word no matter how high or low the voice or the accent with which the user says it.

Alexa Wake Word Neural Net Training

Every word that Alexa hears runs through several layers of testing. This testing determines if the word is the wake word. After the word passes through several layers of verification, and the Echo device determines that it was indeed the wake word, Alexa begins recording.

Accuracy of wake word recognition

People worry that Alexa will begin recording without actually hearing the wake word. So it is vital that Alexa does not start recording without someone directly speaking to it. Recognizing the word and not getting it confused with other words is a top priority.

The software relies on data that represents every possible way to say the word Alexa. This data helps the device determine when the user is speaking to the device. Amazon also uses other data it receives to ensure it doesn’t activate by hearing the word in casual conversation or background noise.

Companies that use the word Alexa in their commercials send that commercial to Amazon. They upload the sound of the wake word in the advertisement to the neural net. This data teaches Alexa to ignore that sound.

Using the network, Amazon makes sure the device will not start recording when the word Alexa is found in background noise. If someone says, “Alexa,” during a television program, the network understands that it is not likely that a large number of devices all heard the same word, in the same tone, at the exact same time. It will ignore the sound and not begin recording.

Erasing Stored Conversations

If you don’t want Alexa listening for the wake word, the simplest thing to do is to press the mute button on the top of the device. It won’t listen until you turn the microphones back on.

Alexa Wake Word Mute

You can change your settings to keep your data private. On the mobile app, open the settings and then select “Manage How Your Data Improves Alexa,” then toggle the switches to the off position.

You can now even tell Alexa to forget everything you said that day. First, enable the skill by opening the settings, then Review Voice History. Turn the switch on. Then when you say, “Alexa, delete everything I said today,” the server deletes all the data from that day.

Amazon is making improvements to the Alexa system. They have new technology in the microphone array and the neural network to make the system even more accurate. In the meantime, if you are concerned about privacy, take steps to keep your data off the servers.

Subscribe to our newsletter!

Our latest tutorials delivered straight to your inbox

Tracey Rosenberger Avatar

Read next

In 1995, Microsoft shipped a cartoon-house interface called Bob, led by Melinda French, who married Bill Gates while it was in development — it demanded twice the memory of a typical home PC, sold roughly 30,000 copies, and was dead within a year, leaving behind the font Comic Sans and the animated assistant that became Clippy.
The Greenland shark grows about one centimetre a year, does not reach sexual maturity until around age 150, and a specimen carbon-dated by Danish researchers in 2016 was estimated to be at least 272 years old, meaning it was already swimming the North Atlantic when Mozart was composing symphonies.
When Apple shipped iOS 12 in June 2018, a small feature called Screen Time slipped onto every iPhone with a counter nobody had quite prepared for — a tally of pickups — and within a day Tim Cook was telling CNN the number of times he picked up his own phone was simply too many
When NASA lost contact with the IMAGE satellite in 2005, an amateur radio operator in Canada named Scott Tilley picked up its signal in January 2018 while hunting for a classified spy satellite, and the spacecraft turned out to be still spinning, still powered, and still trying to phone home after 13 years of silence.
The original iPhone Steve Jobs unveiled in January 2007 could not record video, could not copy and paste text, could not run a single third-party app, and could only reach the internet over 2G — and Jobs spent ninety minutes on stage at Macworld arguing, one missing feature at a time, that every absence was actually a design decision.
In 1965, Joe Sutter’s Boeing team began shaping the 747 around a future they thought would belong to supersonic jets, lifting the cockpit onto a hump so the nose could open for cargo once the giant subsonic passenger plane had outlived its brief moment
Apple’s original 1984 Macintosh keyboard had no arrow keys, no function keys, and no numeric pad because Steve Jobs wanted users to reach for the mouse first. Then Apple quietly sold the missing keys as an accessory.
When the SS Great Eastern laid the first working transatlantic telegraph cable in 1866, a message that had taken ten days by steamship suddenly crossed the ocean in minutes, and the financial markets of London and New York were forced, within a single trading week, to invent the modern concept of synchronised global price.