🎙️ Home Assistant Voice PE

Ambient, localized voice control for your home ecosystem. Connect the Home Assistant Voice Preview Edition (nabu casa PE) to your Companion Hub for fully local voice commands — entirely independent of big-tech listening servers.

Overview

The CI-Home-Assistant-Voice-PE integration bridges the Home Assistant Voice Preview Edition hardware  with your Companion Hub’s local AI stack. Voice processing happens entirely on your Hub using Whisper (STT) and Piper (TTS) — no voice data is sent to Amazon, Google, or any cloud provider.

The integration runs as a companion service to your Hub’s Home Assistant instance, providing:

• Local wake-word detection on the PE hardware
• Whisper-powered on-premise speech-to-text
• Companion Agent integration for general knowledge questions beyond smart home commands
• Piper-powered local text-to-speech for natural-sounding responses

Key Features

• 100% local voice processing — wake word, STT, intent recognition, and TTS all run on your Hub
• Wake word customization — use the default “Hey Jarvis” / “OK Nabu” or train a custom wake word
• Dual-mode responses — Home Assistant handles smart home commands; Companion Agent handles general queries
• Piper TTS — natural, locally rendered voice responses (multiple voices available)
• Multi-room support — use multiple PE units with independent wake words
• Digital Memory integration — conversational queries and answers are optionally logged to memory

Use Cases

• “Turn off all the lights in the kitchen” — Home Assistant handles it locally
• “What’s on my agenda tomorrow?” — routed to your Companion Agent
• “Play jazz in the living room” — sent to your Hub’s Navidrome or Snapcast
• “Lock the front door” — triggers a Home Assistant automation
• “Remind me to call the dentist at 3pm” — creates a task via your Hub’s task manager

Architecture

HA Voice PE Hardware
├── Microphone array
├── Wake word processor (on-device)
└── Audio stream
          │  detected audio
          ▼
CI-HA-Voice Hub App
├── Whisper STT     ──▶ transcribed text
│                            │
│          ┌─────────────────┴──────────────────┐
│          ▼                                    ▼
│   Home Assistant                    Companion Agent
│   (smart home intents)              (general queries)
│          │                                    │
│          └────────────────┬───────────────────┘
│                           ▼
└── Piper TTS ◀── response text
          │
          ▼
HA Voice PE Speaker

Setup

Usage

Issuing a Smart Home Command

Say the wake word followed by a command: “Hey Companion, turn on the living room lights.”

Asking the Agent

Say the wake word followed by a general question: “Hey Companion, how much RAM does my server have?” — this is routed to your Companion Agent.

Changing the Wake Word

Navigate to http://ha-voice.ci.localhost → Wake Word and select from the available options or upload a custom model.

Adjusting Volume

The PE hardware has physical volume buttons. You can also set media player volume via Home Assistant: “Set living room volume to 40%.”

Troubleshooting

Wake word not triggering Ensure the microphone is not muted (check the physical mute button on the PE). Test microphone sensitivity in Diagnostics → Microphone Test.

STT transcription is inaccurate Try a larger Whisper model in Settings → Transcription. Ensure your Hub has sufficient CPU/GPU resources. Reduce background noise near the device.

TTS response is robotic or wrong voice Select a different Piper voice model in Settings → Text-to-Speech. High-quality voices (marked HD) sound significantly more natural.

Agent routing not working Check the agent endpoint is accessible: curl http://open-webui.ci.localhost/api/health. Verify the routing is enabled in CI-HA-Voice settings.

Links

• GitHub Repository 
• Home Assistant Voice PE Hardware 
• Whisper (local STT) 
• Piper TTS