Home Is Where the Assistant Is
Tristam's DIY ESP32-S3 smart speaker offers secure, customizable, and offline voice control of your entire home automation system.
It used to be the case that if you wanted to take advantage of a voice assistant, you would have to trust a remote, cloud-based service with access to always-listening microphones positioned throughout your home. That is a huge ask, especially considering the steady stream of security breaches that we read about in the news. But thanks to the technological advances of the past few years, voice assistants no longer require the backing of massive data centers. We can even run very capable voice assistants right in our own homes — no internet connection required.
Not only is the technology available, but it has also become very simple and inexpensive to build a local voice assistant. One of the simplest systems you are going to come across anywhere was recently built and described by Tristam. It consists of an ESP32-S3 development board-powered smart speaker with a local conversation agent running onsite, with Home Assistant integration, that responds to user requests.
The voice assistant contains an ESP32-S3-DevKitC-1-N8R2 development board, a microphone, an amplifier, and a speaker. These components are installed in a well-designed 3D-printed case so that they will look nice on a desk or nightstand in your home. The boards were flashed with ESPHome to make them easily accessible to a Home Assistant automation network, and they run microWakeWord. This is a TensorFlow-based wake word detection training framework that enables tiny hardware platforms to recognize wake words like “Alexa” or “Hey, Google.”
Once the wake word is detected, the device captures the speaker’s voice and transmits it to a conversation agent running on a local machine under the control of Home Assistant. This can be either a Home Assistant-native conversation agent, or an LLM-based one. And because it is integrated with Home Assistant, it can control all of your other smart devices, like lights and media players.
If the thought of sending your private conversations to the cloud has kept you away from using voice assistants or home automation technologies, be sure to check out Tristam’s project write-up. It is a very simple, secure, and completely customizable way to jump into this area.