Home Assistant Voice: Local Voice Assistants Explained (Wyoming, Voice PE) UK

If you’ve followed our Home Assistant beginner’s guide and started building out automations, local voice control is the natural next step - asking your home to do things without Alexa or Google listening on a cloud server somewhere. Home Assistant Voice has matured quickly over the past couple of years, and it’s now genuinely usable rather than an enthusiast science project.

What “Local Voice” Actually Means

Standard voice assistants (Alexa, Google Assistant) send your voice recording to the manufacturer’s cloud for processing, then send a response back. Home Assistant’s voice stack - built on an open-source project called Wyoming - runs speech-to-text and intent recognition entirely on your own hardware, so nothing leaves your network unless you specifically choose a cloud-based model for more complex queries.

The Hardware: Home Assistant Voice PE

Nabu Casa (the company behind Home Assistant) sells a purpose-built device called the Home Assistant Voice PE - a small speaker/microphone puck that acts as a “satellite,” listening for a wake word and streaming audio to your Home Assistant instance for processing. It’s the officially supported way to add voice control without piecing together a DIY microphone setup, and it’s a sensible first purchase if you want to try this without troubleshooting hardware compatibility. At roughly £50-60 per unit, adding one per commonly-used room is a modest cost compared to a whole-house Alexa/Echo rollout, and there’s no ongoing subscription required for the core local voice features.

Check Home Assistant Voice PE on Amazon UK →Affiliate link

What It Can and Can’t Do (Yet)

Capability	Status
Controlling lights, plugs, switches, climate	Works reliably, fully local
Simple sensor queries (“what’s the temperature in the bedroom”)	Works reliably, fully local
Free-form conversation / general questions	Requires an optional cloud or local LLM add-on
Multi-room audio & music	Improving, less mature than Sonos/Alexa

Running a Local LLM for Free-Form Conversation

The “requires an optional cloud or local LLM add-on” line in the table above is worth unpacking, because it’s the single biggest gap between Home Assistant Voice and something like Alexa out of the box. Home Assistant’s built-in intent system is deliberately narrow - it matches phrases like “turn on the kitchen lights” against known devices and areas, and anything outside that pattern (open-ended questions, general conversation) simply isn’t handled without extra setup. Bridging that gap means installing a local LLM add-on (commonly Ollama, running a small model like Llama 3.2 or a similarly sized model), which needs meaningfully more hardware than the Raspberry Pi or NUC most people run Home Assistant on - realistically a machine with a dedicated GPU or a recent Intel/AMD chip with strong integrated graphics, since a bare Raspberry Pi will process a conversational query slowly enough to be genuinely annoying to use daily. The alternative is routing free-form queries to a cloud LLM (OpenAI, Anthropic or similar via an integration), which keeps the hardware bar low but reintroduces exactly the “leaves the building” trade-off that local voice was meant to avoid for that specific query type. Most households land on a middle ground: fully local for device control and simple queries, and either skip conversational AI entirely or accept the cloud round-trip for the rare occasion it’s actually wanted.

Why Bother, If Alexa Already Works?

The honest answer: for pure convenience, Alexa and Google Assistant are still slicker for general use. Home Assistant Voice is for a specific kind of household - one that wants voice control of the home without any recordings leaving the building, and one that’s already invested in Home Assistant for automations. It pairs naturally with the advanced automations and ESPHome side of Home Assistant, since you can trigger the same automations by voice that you’ve already built with sensors and scripts.

Getting Started

• Install the Assist pipeline from Home Assistant’s settings (built in, no extra add-on required for basic use)
• Add a Voice PE or a DIY ESP32-S3-based satellite as a Wyoming satellite device
• Train custom commands and areas so Home Assistant knows which room a voice command came from

Choosing Wake Word and Microphone Placement

The Voice PE’s default wake word works reliably in most rooms, but placement matters more than people expect - putting it too close to a TV, extractor fan, or another speaker introduces background noise that measurably hurts recognition accuracy. A shelf or worktop at roughly seated head-height, away from major noise sources, gives noticeably better results than a corner or a spot right next to a kitchen extractor. If you’re running more than one satellite, Home Assistant automatically figures out which one heard a command most clearly rather than both responding at once, so there’s no real downside to adding a second unit in a large or oddly-shaped room.

Troubleshooting False Triggers and Missed Commands

• The satellite wakes up on its own or from unrelated speech - background TV or radio dialogue occasionally contains sounds close enough to the wake word to trigger a false positive; moving the satellite further from a TV speaker or switching to a less common custom wake word usually resolves this
• Commands work in one room but not another - each satellite’s microphone sensitivity and room acoustics differ, so a command that’s recognised reliably in a quiet bedroom can fail in an echoey kitchen with hard surfaces; Home Assistant’s per-satellite volume and sensitivity settings are worth tuning individually rather than assuming one configuration suits every room
• Recognition accuracy drops with regional accents - the default speech-to-text model handles most UK English reliably but can misfire on strong regional accents or unusual device/area names; renaming devices to simpler, more phonetically distinct names (avoiding near-homophones like “living room lamp” and “living room lamps”) measurably improves recognition
• A command is heard correctly but nothing happens - this is usually an area-assignment mismatch rather than a voice-recognition failure; check that the device is actually assigned to the area you referenced in the command, since Home Assistant matches “the kitchen lights” against area assignments, not physical location

Frequently Asked Questions

Does Home Assistant Voice work without an internet connection?

Yes, for the core capabilities (device control, simple sensor queries) - since speech-to-text and intent recognition run locally, these keep working even if your internet is down. Free-form conversational queries that specifically route to a cloud or hosted LLM add-on are the exception, since those genuinely need an internet connection.

Can I build a DIY voice satellite instead of buying the Voice PE?

Yes - an ESP32-S3 board with a microphone and speaker can run the Wyoming satellite firmware, at a lower hardware cost than the official Voice PE if you’re comfortable with the DIY setup. The Voice PE exists specifically to remove that DIY step for anyone who’d rather not troubleshoot hardware.

How much does a full multi-room Home Assistant Voice setup typically cost?

Budgeting one Voice PE per commonly-used room (kitchen, living room, main bedroom) at roughly £50-60 each puts a three-room setup around £150-180 total, with no ongoing subscription - comparable to or cheaper than an equivalent number of Echo devices, without the recurring cloud dependency.

For comparison against the mainstream ecosystems, our Google Home vs Amazon Alexa guide covers where cloud assistants still have the edge.