Home Assistant Voice Preview Edition: What Assist Can Do Today

I cover everything you can do with Home Assistant's Assist voice system today, from basic commands to timers, room context, and AI integration. The Voice Preview Edition hardware accelerates development and language support thanks to community contributions. Local speech-to-text remains the biggest limitation for many languages, but cloud options and Piper TTS deliver excellent results.

Origins of Assist

Voice control in Home Assistant started as far back as 2017. The big push came in 2023 with the Year of the Voice that added Assist, improved intents, added languages, created wake words, and built solid local and cloud options. Since then we've added integrated AI, timers, and better on-device wake words. The Voice Preview Edition continues that momentum.

Commands and Timers

Assist turns spoken commands into Home Assistant actions. You can control almost every part of your smart home with intents for on, off, play, pause, open, close and more. It also answers questions about time, weather, and temperature.

Timers were one of the most requested features. You can set, pause, adjust, or cancel them. You can attach actions like turning off the TV in 15 minutes. On the Voice Preview Edition the LED ring counts down and flashes when done. Saying 'stop' silences the alarm without a wake word.

Device Exposure, Aliases, and Room Context

Unlike other voice assistants, you decide exactly which devices are exposed. You can hide a door lock but expose the open sensor instead.

Aliases let you give devices multiple names so you can speak naturally. Room context is excellent: tell the device which room it's in and organize devices by area, then commands like 'turn off the lights' or 'pause music' affect only that room.

Wake Words and Speech Processing

microWakeWord runs on the device itself for fast response. Community members donate voice samples to improve accuracy. Available on-device wake words include 'Okay Nabu', 'Hey Jarvis', and 'Hey Mycroft'.

Speech processing happens in an Assist pipeline. It is too heavy for the Voice Preview hardware, so you can use Home Assistant Cloud for the most accurate and private option. Local alternatives exist but depend heavily on community language support.

Language Support, TTS, and STT

We support over 25 languages for intents and over 30 for Piper text-to-speech. Piper is fast, local, and runs well on a Pi 4. Voices come from our community.

Local speech-to-text using Whisper is the main bottleneck. Only about 15 languages work well on reasonable hardware like an Intel N100. We're building simpler sentence recognition that could run on a Pi 4 but needs predefined phrases. Cloud usually gives better results. You can mix both and switch automatically when internet is down.

AI Integration

We combine local conversation agents with AI for natural language understanding. You can use cloud providers like ChatGPT, Gemini, or Claude, or run models locally with Ollama if you have a capable GPU.

AI handles slight name variations and questions outside built-in intents. It can combine sensor data for useful answers like air quality reports and suggestions. Results are still experimental and not always consistent, but the choice is yours.

Who it's for

Get it if: Home Assistant users who want open, private voice control and are willing to contribute to language support or run local AI.

Skip it if: People needing perfect local performance in unsupported languages or who want fully consistent AI home control right now.