Voice Assist Setup
Selora AI configures Home Assistant Voice Assist end-to-end — engine, pipeline, and per-device tuning — so homeowners get working voice control without learning STT and TTS.
Summary
Home Assistant’s Voice Assist lets you talk to your home, but getting it working properly means choosing speech-to-text, text-to-speech, and wake-word engines and wiring them into an Assist pipeline. For a homeowner who just wants to say “Hey Home Assistant, turn off the lights,” that is a lot of machinery. Selora AI already knows your home — so it should be the one that configures voice, prescriptively, with convention over configuration.
The problem
- Home Assistant asks users to understand TTS and STT. Voice Assist has real moving parts — speech-to-text (STT), text-to-speech (TTS), and a wake-word engine — and the setup flow hands newcomers a list of engines with no guidance on which combination will actually run well on their hardware.
- The defaults are generic. Stock pipelines pick engines and options for a generic setup, not for the specific speakers, microphones, and rooms in a given home, so recognition quality is left to luck.
- No local-first story. Newcomers end up flipping between cloud STT and TTS services they must sign up for, even when a fully local stack fits their hardware perfectly.
- Troubleshooting is on the user. When a phrase is not understood, nothing tells the homeowner whether the microphone, the engine choice, or the network is the problem.
What’s included
- A Selora AI guided flow that sets up Voice Assist end-to-end for the home — no engine-picker screens, no STT or TTS vocabulary required.
- Engine selection by convention: Selora chooses the STT, TTS, and wake-word stack from the installation manifest — audio hardware, network setup, and local-versus-cloud preference — and writes the Assist pipeline to match.
- Prescriptive defaults tuned per home: microphone and speaker roles, device placement, and pipeline options sized to the home’s layout and hardware instead of generic settings.
- A “say it, it works” verification step: after setup, Selora runs a sample phrase through the new pipeline and reports what it understood, so the homeowner gets immediate confirmation that voice is working.
- Sensible fallbacks: if the hardware cannot run a fully local stack, Selora explains the trade-off and configures the best cloud option rather than failing silently.
Why it matters
Voice is the most natural interface to a home, and Home Assistant is good at it — but the gap between “installed” and “working” is what kills adoption. Every homeowner with Selora AI should get working voice control in a few minutes, the same way they get the rest of their setup: because Selora knows the home, Selora configures it.
The complexity
- Matching engine capability to hardware. Local speech-to-text and text-to-speech have real CPU and memory footprints. Selora must score the home’s hardware and pick a stack that runs well on it, not the stack that wins on a spec sheet.
- Tuning across a heterogeneous device fleet. Microphone quality, speaker placement, and room acoustics vary widely between homes. Prescriptive defaults need per-device profiles, not one global configuration.
- Keeping the configuration maintainable. Voice Assist configuration lives inside Home Assistant. Selora must own the conventions it writes so that future updates, recipes, and the AI agent itself do not step on them.
Out of scope (for now)
- Custom wake words.
- Multi-language and multilingual pipelines.
- Audio hardware itself — microphones and speakers remain a homeowner or installer choice; Selora configures what is present.
- Replacing Home Assistant’s Voice Assist — Selora configures it, it does not fork it.
Last modified November 17, 2026: docs(roadmap): add Voice Assist Setup (c8c8638)