How a Smart Speaker Hears Its Wake Word
A tiny always-on model listens for one sound pattern. Only after it matches does the bigger system wake up.
The robot repeats this loop many times a second. Tap a step to pause on it.
A smart speaker does not run its full assistant all the time. A very small model, using little power, listens only for the pattern of its wake word.
When that small model is confident it heard the pattern, it wakes the larger speech system that understands the rest of your sentence.
The device can stay on all day on little power and respond hands-free.
It would need a button press, or it would have to run the heavy speech model constantly, which drains power.
It sometimes wakes on similar-sounding words, or misses the wake word in a noisy room.
Why use a very small model for the wake word?
Go deeper in the Atlas
How Robots Sense
- Ultrasonic Sensor
- Computer Vision
- How a Smart Speaker Hears Its Wake Word
- Gyroscope and IMU
- How Self-Checkout Cameras Recognise Fruit
- How LiDAR Works
- SLAM
- Sensor Fusion
- SLAM Loop Closure
- Weeding Robots