Audio Assistant: Build a Local Voice Assistant with Home Assistant and Whisper
An audio assistant you control: local voice input, Wake Word detection, and LLM answers. We show the Home Assistant + Whisper + Ollama stack.
📜 Table of Contents
Big-tech voice assistants listen, log, and serve ads. A local audio assistant does the same job with none of that: wake word on-device, speech-to-text with Whisper, answers from a local LLM, and your data never leaves the house.
The Stack
Home Assistant (open source) is the brain: it integrates wake-word detection, media players, and automations. For speech-to-text, Wyoming protocol integrations wrap faster-whisper (24,776 stars) or whisper.cpp (52,615 stars) - both run on a Raspberry Pi 5 for basic use. For answers, point the assistant at Ollama (177,902 stars) running a 7-8B model; add a text-to-speech engine (piper or coqui TTS, 45,859 stars) for spoken replies.
Assembly: HAOS on a Pi or mini PC, install Wyoming faster-whisper and piper add-ons, configure the wake word (openWakeWord), then connect the conversation agent to Ollama. Total hardware cost: a Pi 5 plus a microphone array - under 100 dollars, fully owned.
Comparison
| Component | Role | Tool |
|---|---|---|
| Brain | Automation hub | Home Assistant |
| STT | Speech to text | faster-whisper 24,776 |
| LLM | Answers | Ollama 177,902 |
| TTS | Spoken replies | piper / coqui 45,859 |
FAQ
Q: Do I need a GPU?
A: No - whisper.cpp and 7-8B models run on CPU, just slower. A Pi 5 handles basic commands; a mini PC with 16GB RAM is snappier.
Q: Can it control smart home devices?
A: Yes - that is Home Assistant main job. Say turn off the lights and the assistant triggers the automation.
Written by our editorial team; tools listed here are tested or verified against public sources. Links point to official sites or GitHub repos for reference only — no paid placements.