Hi Adam,
Hope you don't mind, but I just got Claude to write the reply to this as it's a lengthy explanation and easier that way.
This should do it for you, but let me know if you need further assistance:
Sophia is only the conversation agent. It handles the "understand what was said and decide what to do" step. It isn't STT (speech to text) or TTS (text to speech), so Home Assistant's Voice Assistant pipeline needs separate components for those. Home Assistant Cloud is just the only option listed because you haven't installed any local ones yet. You can run everything locally with Whisper (STT) and Piper (TTS).
The pipeline looks like this:
Mic → Whisper (STT) → Sophia (conversation agent) → Piper (TTS) → Speaker
Option 1: Home Assistant OS or Supervised (add-ons)
- Go to Settings → Add-ons → Add-on Store.
- Install Whisper and start it. Pick a model on its Configuration tab.
tiny-int8 or base-int8 is fine for low-power hardware, and small-int8 is more accurate if your CPU can handle it. Set your language too.
- Install Piper and start it. Choose a voice on its Configuration tab, e.g.
en_US-lessac-medium.
- Go to Settings → Devices & Services. Both should be auto-discovered under the Wyoming Protocol integration. Click Configure and Submit on each.
Option 2: Home Assistant Container or Core (no add-ons)
Run the Wyoming containers yourself:
docker run -d --name wyoming-whisper -p 10300:10300 -v /path/to/whisper-data:/data \
rhasspy/wyoming-whisper --model small-int8 --language en
docker run -d --name wyoming-piper -p 10200:10200 -v /path/to/piper-data:/data \
rhasspy/wyoming-piper --voice en_US-lessac-medium
Then go to Settings → Devices & Services → Add Integration → Wyoming Protocol and add each one using your server's IP, with port 10300 for Whisper and 10200 for Piper.
Then set up your assistant
Go to Settings → Voice Assistants and edit (or add) your assistant:
- Conversation agent: Sophia NLU
- Speech-to-text: faster-whisper
- Text-to-speech: piper (pick your voice)
That should also fix your "no response from Sophia" issue. With TTS set to None, Sophia is still processing the command, but nothing gets spoken back.
A note on "Prefer handling commands locally"
That setting only controls whether Home Assistant's built-in intent handler gets the first shot at a command before it goes to your conversation agent. It has nothing to do with where STT and TTS run. Once Whisper and Piper are set up, the whole chain stays on your own network.
Let me know how it goes!