Home Assistant local voice hardware: what to buy before adding a cloud assistant

Give Assist a voice before you give it another subscription. Try a command in the Home Assistant app first; once that works, a small microphone-and-speaker device gives you a dedicated place to speak. The useful purchase is an endpoint that fits how much building you want to do.

Hardware and setup facts come from the makers and Home Assistant docs, checked November 1, 2026.

Pick your kind of voice project

Choose M5Stack Atom VoiceS3R for an ESPHome project. Its ESP32-S3, microphone, speaker and maker-maintained Assist configuration give you a compact device to flash and configure.

Choose THIRDREALITY Voice/Music Assistant Dev Edition for a preloaded Linux endpoint. You get dual microphones, a speaker and voice/audio software in one unit. Add it to Home Assistant and choose where speech processing runs.

Choose M5Stack ATOM Voice C008-C for a tiny first build. This is the hardware called ATOM Echo in Home Assistant’s tutorial. Follow that device’s installer rather than choosing a similarly named board.

Already want the official Home Assistant voice hardware? The Voice Preview Edition deserves a look through its official product page. It is Home Assistant’s recommended voice device, with dual microphones, an XMOS audio processor and a switch that physically cuts microphone power.

Compare the voice hardware

Pick	Maker hardware	Home Assistant route	What you set up
M5Stack ATOM Voice, C008-C	ESP32-PICO-D4, 4 MB flash, microphone and integrated speaker	Official ATOM Echo tutorial / ESPHome	Flash the matching firmware; configure host openWakeWord and your Assist pipeline
M5Stack Atom VoiceS3R, C126-VOICE	ESP32-S3, 8 MB flash + 8 MB PSRAM, MEMS microphone and 1W speaker	Maker’s VoiceS3R Assist firmware / ESPHome	Flash, add the discovered device, assign its area and pipeline
THIRDREALITY Voice/Music Assistant Dev Edition	Linux platform, dual digital microphones, 3W speaker, 2.4 GHz Wi-Fi	Preloaded software implementing the ESPHome native API	Add the speaker and choose host speech processing or the firmware’s English-only on-speaker mode

M5Stack C008-C hardware · VoiceS3R hardware · THIRDREALITY hardware

The endpoint hears you and plays the answer. Your selected pipeline turns speech into text, handles the command and produces the spoken reply. Buying a local-capable speaker does not select those services for you. How Assist fits together

VoiceS3R: the ESPHome pick

The Atom VoiceS3R is the pick when the firmware is part of the fun. M5Stack documents a complete Assist configuration with wake-word detection, voice input and announcements. Use its Home Assistant setup guide, which calls for ESPHome 2026.1.0 or later and links the precompiled installer.

Choose the exact VoiceS3R kit with its integrated audio hardware. Some firmware filenames still use “EchoS3R”; the current maker guide calls the device VoiceS3R. After flashing and connecting it, add the discovered ESPHome device, assign an area and select your Assist pipeline.

Have a USB data cable ready for flashing and a suitable USB power source for everyday use. Use your existing Home Assistant host with the local pipeline below.

ATOM Voice: start small

The C008-C ATOM Voice is the tiny alternative for following Home Assistant’s established ATOM Echo tutorial. M5Stack’s current page identifies the same C008-C hardware under the Voice name and links a Home Assistant tutorial. Use the official ATOM Echo setup page in desktop Chrome or another supported Chromium browser, with a USB-C data cable.

For the C008-C wake-word route, install openWakeWord and add its Wyoming integration. In the assistant’s menu, select Add streaming wake word, then choose openwakeword and ok nabu. This path streams audio to the Home Assistant host for wake-word recognition. Wake-word setup

Flash the Home Assistant voice firmware for this route. M5Stack also documents other cloud-connected voice demos; those are separate projects. Keep your first build focused on getting one named light to respond through Assist. C008-C maker documentation

THIRDREALITY: a preloaded Linux endpoint

The THIRDREALITY Dev Edition combines voice capture and audio playback. Its hardware page documents dual digital microphones and a 3W speaker. Choose it when you want the assembled endpoint and room to explore its Linux software.

Update using the maker firmware guide, then follow its Companion-app setup, ESPHome discovery and pipeline selection. It also offers an English-only “3R Speaker” mode for on-speaker speech-to-text and text-to-speech after an initial model download. For that mode, enable Allow the device to perform Home Assistant actions in the speaker’s ESPHome Configure dialog. Home Assistant still handles intents and actions; the speaker remains connected to it.

Start with the host pipeline below. For developer images, disable network ADB and replace default logins using the maker’s debugging notes. Voice commands work independently of the optional Music Assistant audio project.

Choose what processes your speech

Route	Good fit	What to select
Focused local control	Lights and other supported home-control phrases	Speech-to-Phrase, Home Assistant conversation agent, Piper
Open-ended local speech	Broader transcription, with a host sized for the workload	Whisper, your chosen conversation agent, Piper
Optional cloud speech	Offload speech processing	Home Assistant Cloud’s speech services

Start with Speech-to-Phrase for supported home-control commands. It handles a defined set of phrases; open-ended shopping-list entries and similar requests need a different route. Choose Whisper when broader transcription matters and your host can carry the work. Piper supplies local text-to-speech. Check language support in the local Assist guide.

For focused commands, Speech-to-Phrase plus Piper suit a Pi 5 NVMe host: Home Assistant’s local Assist documentation describes fast Speech-to-Phrase on a Pi 4 and Piper optimized for that hardware.

For a fully local first setup, select the Home Assistant conversation agent too. You do not need an LLM for ordinary device control. The Voice Preview Edition guide explains the focused-local versus full-local choices and recommends Intel N100-class hardware or equivalent for its Whisper Base route. Step up to a mini PC for Whisper.

Set it up, then try one light

1. On Home Assistant OS, build the pipeline. Install and start Speech-to-Phrase or Whisper, plus Piper, using the Home Assistant local setup instructions. Add the discovered Wyoming services under Settings → Devices & services.
2. Add an assistant. Open Settings → Voice assistants → Add assistant. Choose your language, the Home Assistant conversation agent, your local speech-to-text service and Piper.
3. Add the endpoint. Follow the matching M5Stack installer or THIRDREALITY’s Companion-app setup, then add its discovered ESPHome device and select that assistant.
4. Expose a light. Under Settings → Voice assistants → Expose, enable one light you want to control. Give it a clear name and area. Entity exposure instructions
5. Speak, then inspect. Try turning that light on and off using its name. Repeat from the phone’s Assist interface to separate command configuration from microphone or speaker issues.

Start with one useful job: turn the desk light on, turn it off, then ask about a named temperature sensor you already have. Keep device names distinct so Assist can find the right target. Our starter kit and room-sensor guide give you practical devices to name and expose.

Fix the layer that failed

The phone works but the endpoint does not: check its selected assistant, area and Wi-Fi connection. Assist understands the words but cannot find the light: check its name and exposure. A supported command works but a free-form request fails: check whether you selected Speech-to-Phrase. You get no spoken reply in the host-based setup above: check the pipeline’s Piper selection and the endpoint’s playback settings.

Choose VoiceS3R for the ESPHome build, THIRDREALITY for a preloaded Linux voice/audio endpoint, or C008-C ATOM Voice for the small tutorial project. Start with one room’s controls, then build out from the Home Assistant hub.