How do you use Whisper for speech to text in Home Assistant?

You use Whisper for speech to text in Home Assistant by installing its speech app next to Piper, letting the Wyoming integration find both, and then selecting Whisper under Speech-to-text in a voice assistant. Whisper is the open-ended engine in Home Assistant's local setup. It tries to transcribe everything you say, which lets it handle a shopping list or a named timer, but that flexibility costs processing time and hardware.

This answer follows the project's local setup guide and its hardware advice, checked on 17 October 2026. The steps assume you run the Home Assistant Operating System, which the project says you need to run the whole local voice setup together.

Set up Whisper with Piper in five steps

The setup has the same shape as the one for Speech-to-Phrase. Install the Whisper app and the Piper app, then:

1. Start both apps. Under Settings > Devices & services, the Wyoming integration should discover both services. Select Add for each ¹.
2. Go to Settings > Voice assistants and select Add assistant.
3. Under Conversation agent, select Home Assistant.
4. Under Speech-to-text, select Whisper, and under Text-to-speech, select Piper. Choose the language for each. Depending on your language, you may be able to pick a regional variant.
5. Expose your devices to Assist, or the assistant cannot control them by voice.

Check how Whisper handles your language before you set it up. The project's own example is French, which may work well enough with the largest model, and that model needs a GPU ². Our question on Whisper languages covers the language list, and the question on which Whisper model is best covers the model sizes.

Whisper needs a faster computer than Speech-to-Phrase

The main cost of Whisper is time. The project's current documentation says that on a Raspberry Pi 4, Whisper takes around 8 seconds to process incoming voice commands ³. On the same machine, Speech-to-Phrase transcribes in under one second ⁴.

For that reason the project recommends a faster machine. Its advice for running Whisper locally is at least an Intel N100 or an equivalent processor, which runs the Base model at a reasonable speed for English and Spanish ⁵. The project also says Whisper is only a great option if you have powerful hardware at home and plan to go beyond simple home control, for example by pairing the assistant with a language model ⁶.

Pick Whisper only if you need open speech

Whisper is the right engine when you want to speak freely, dictate a shopping list, name a timer or pass the request to a language model. Speech-to-Phrase is the better engine when you only want fixed home commands, such as lights, locks and timers, on a small machine. The table compares the two using the project's figures.

Language matters for the hardware choice too. The project says that French may work well enough with the largest Whisper model, but that model needs a GPU, and it is unusable on a Raspberry Pi or on an N100-class PC ². If your language needs a large model, plan for a GPU before you buy.

Know the limits of open speech

Open speech is harder to transcribe than a fixed phrase list. Our guide to word error rate explains how transcription accuracy is measured, and why one percentage cannot be compared across different test sets.

Whisper can also produce text the speaker never said. The Whisper authors list hallucination, where the model outputs a transcript that is entirely unrelated to the audio, among its known failure modes ⁷. Check a command before you rely on it for something that matters. The glossary entry for word error rate defines the measure.

If your language has no good local model

If you cannot get a good result locally in your language, the project says that for some languages it believes Home Assistant Cloud provides the best experience, because the local speech-to-text models do not support them ⁸. The cloud route sends your audio to a service outside your home, so decide that trade-off before you switch. Our question on whether Home Assistant voice works offline sets out what stays local and what does not.

For the full pipeline, from the wake word to the spoken reply, read our guide to Home Assistant voice. For how the voice commands themselves are set up, see how to set up voice commands in Home Assistant. For the engine underneath, our guide to running Whisper locally explains the open-source model in more depth.