Projects / Ideas / Wake-word detector on ESP32-S3

Wake-word detector on ESP32-S3

IntermediateEdge AIIdeaReviewed
In short

How do you build a wake-word detector on ESP32-S3?

A small neural network on an ESP32-S3 listens for a custom wake word and lights an LED, with no internet. You need Seeed XIAO ESP32S3, MAX9814 microphone with AGC, LED 5 mm green and Resistor 220 Ω.

Level
Intermediate
Time
About 240 min
Parts
4
An idea, not a guided build yet. Goal, parts and steps; no wiring, code or tests. Copy it to My Garage to plan it.
Goal

Train and run a keyword model on a microcontroller.

Parts list

How to build it

  1. Record 50 samples of your word and of background noise
  2. Train a model with an online TinyML tool
  3. Export it as a library
  4. Run inference on the mic stream
  5. Tune the threshold to avoid false triggers

Related projects

Common questions

What parts do I need for the wake-word detector on ESP32-S3?

Seeed XIAO ESP32S3, MAX9814 microphone with AGC, LED 5 mm green and Resistor 220 Ω. MAX9814 microphone with AGC: Or the Sense board's mic.

How long does the wake-word detector on ESP32-S3 take to build?

About 240 minutes for a intermediate maker.