What is the best speech to text for Home Assistant?

For most homes, the best speech to text for Home Assistant depends on one question: do you only need fixed commands, or do you want to speak freely? If you only need commands such as turning on lights, locking doors or setting a timer, Speech-to-Phrase is the best choice. It runs on your own Home Assistant system, and the project reports that it transcribes in under one second on a Raspberry Pi 4 ¹. If you want to speak freely, choose Whisper, which handles open speech but takes longer per command. For some languages with no good local model, the project believes Home Assistant Cloud gives the best experience.

This answer follows the project's own documentation for Assist, the voice assistant built into Home Assistant. Our full guide to Home Assistant voice walks through the whole setup, and the figures below are as published on 17 October 2026.

Speech-to-Phrase is fastest because it only expects known phrases

Speech-to-Phrase is fast because it is built for a narrow job. It recognizes the commands it was built from, so it does not have to transcribe every possible sentence. The narrow design also sets its limits. The project says that shopping lists, naming a timer and broadcasts are not usable out of the box ². If your daily use includes any of those, Speech-to-Phrase will disappoint you, and Whisper is the better match.

A command that uses a device name Home Assistant does not know will not match either, so the names in your setup matter as much as the engine.

Whisper handles open speech, but each command takes longer

Whisper is the open-ended option. The project describes it as an open-ended speech model that will try to transcribe everything ³. That open-ended design is why it can transcribe a shopping list or a named timer.

The speed figure is where the sources disagree, so read both. The current documentation says Whisper takes around 8 seconds per command on a Raspberry Pi 4 ⁴. The February 2025 blog post gave a lower figure of at least 5 seconds on a Raspberry Pi 4 ⁵. We could not find a reconciliation in the sources we checked. On either figure, Whisper is slower than Speech-to-Phrase on the same hardware.

For that reason the project recommends a faster machine. Its advice is at least an Intel N100 or an equivalent processor, which runs the Whisper Base model reasonably fast for English and Spanish ⁶.

The choice depends on hardware, speech and language

Home Assistant Cloud is optional. The project says it supplies the speech engines used by Assist, and that Home Assistant works fully without it ⁷. It is the fallback, not the default. The project also says that for some languages it believes Home Assistant Cloud gives the best experience, because the local speech-to-text models do not support them ⁸. Check your language before you commit to a local engine.

Cloud speech sends your audio off your network, so it is a privacy decision as well as a speed and accuracy one. Our guide to speech to text privacy explains what to check before you choose a cloud service.

What to do

1. Check your language on the project's list of supported languages. If your language has no good local model, plan for Home Assistant Cloud.
2. Pick Speech-to-Phrase if your home only needs fixed commands. A Raspberry Pi 4 is enough, and it is the fastest option in this comparison.
3. Pick Whisper if you want to speak freely or pair Assist with a language model. Use at least an Intel N100 or an equivalent processor.
4. Test your device names. Say each command as you would say it at home, and rename a device or add an alias if Assist does not match it.

If you want to know which commands Assist understands, read our question on Home Assistant voice commands. For the language model side, read whether Home Assistant can use ChatGPT or a local LLM for voice. For the hardware that listens for your wake word, read what the Voice Preview Edition is.

The four stages of a request are a wake word, speech to text, an intent match or language model, and text to speech. Speech-to-Phrase and Whisper both replace only the speech to text stage, so the rest of the setup stays the same. For the wider field, see speech to text AI and voice agents.