Home Assistant6 min read

Home Assistant Multi-Wake Word Voice Control: Mastering Local AI & Custom Triggers

Configure multiple wake words in Home Assistant using local AI, set up custom voice commands, and keep all voice processing on your own hardware.

Daniele Messi

· updated · Geneva

Italian version →
In this article 7
  1. Key Takeaways
  2. The Evolution of Home Assistant Local Wake Word in 2026
  3. Why Multi-Wake Word Smart Home Control is Essential
  4. Setting Up Local AI: openWakeWord, Whisper, and Piper
  5. Configuring Multi-Wake Words and HA Custom Voice Commands
  6. Conclusion
  7. Related Articles

Key Takeaways

  • Home Assistant local wake word technology in 2026 enables unparalleled privacy and responsiveness for smart home voice control.
  • Implementing multi-wake word smart home setups allows for personalized and context-aware voice interactions for different users or scenarios.
  • Local AI, including models like Whisper and Piper, processes voice commands entirely on-device, eliminating cloud dependency and enhancing data security.
  • Configuring HA custom voice commands through the Home Assistant voice assistant pipeline offers deep integration and flexible automation triggers.

In 2026, smart home voice control has shifted toward local processing, with better privacy, speed, and customization as a result. For Home Assistant enthusiasts, mastering Home Assistant local wake word capabilities is no longer a niche pursuit but a cornerstone of a truly intelligent and secure smart home. This guide covers setting up multi-wake word voice control, using local AI, and writing HA custom voice commands that run entirely on your own hardware.

The Evolution of Home Assistant Local Wake Word in 2026

The shift to local AI for voice processing represents a significant leap forward for smart homes. Unlike traditional cloud-based assistants that send your voice data to external servers for processing, Home Assistant local wake word systems perform all recognition and intent parsing on your local network. This means your commands are processed almost instantaneously, and your privacy is fully preserved, as no audio leaves your home. Deploying Home Assistant local wake word solutions can reduce cloud latency by over 90%, making your smart home feel more responsive and intuitive.

Modern advancements in on-device machine learning models, like optimized versions of Whisper for speech-to-text and Piper for text-to-speech, have made this local processing highly accurate and efficient. This local-first approach matters for anyone who cares about data sovereignty and wants to avoid the privacy implications of always-on cloud microphones. For more on privacy-focused voice solutions, see our article on Home Assistant Local Whisper & TTS: The Ultimate Privacy Voice Assistant.

Why Multi-Wake Word Smart Home Control is Essential

Imagine a smart home that responds differently based on who is speaking or the context of the command. This is the power of a multi-wake word smart home. Instead of a single, generic trigger phrase, you can configure several wake words, each designed for specific purposes or users. For instance, you might use “Hey Home” for general control, “Hey Kitchen” to target only kitchen devices, or a name-based phrase so children get a restricted command set while adults get full access. Each wake word routes to a different voice pipeline in Home Assistant, so the intent parsing and the actions that follow can be entirely separate.

This separation keeps things predictable. A phrase tied to a restricted pipeline can only trigger automations you explicitly allow, which matters when you have guests or young children using the same hardware. It also lets you tune detection sensitivity per phrase: a wake word you use constantly can have a tighter threshold than one reserved for rarely used scenes.

Setting Up Local AI: openWakeWord, Whisper, and Piper

The three components you need for a fully local voice pipeline are a wake word engine, a speech-to-text model, and a text-to-speech model. The standard stack on Home Assistant in 2026 is openWakeWord for detection, Wyoming Faster-Whisper for transcription, and Wyoming Piper for spoken responses.

Install all three as Home Assistant add-ons from the add-on store. Each exposes a Wyoming protocol endpoint that the voice pipeline configuration points to. They register automatically once running, so no configuration.yaml edits are needed. Confirm they appear under Settings > Voice Assistants > Add Assistant.

For openWakeWord, the add-on ships with several prebuilt models. To add a custom phrase, train a small model using the openWakeWord training notebook and drop the resulting .tflite file into the add-on’s model directory, typically /share/openwakeword/. Restart the add-on and the new phrase appears in the wake word selector.

Whisper model size is the main tuning knob for accuracy versus CPU load. The small model handles most accents well on a mid-range x86 machine or a capable single-board computer. Step up to medium only if transcription accuracy is genuinely a problem, because RAM usage and processing time increase noticeably on constrained hardware.

Piper voice quality has improved considerably. Pick a voice from the available model list and set speaker speed in the add-on options. Most configurations work well at a value between 1.0 and 1.1 for natural-sounding output.

Configuring Multi-Wake Words and HA Custom Voice Commands

Each voice assistant pipeline in Home Assistant has its own wake word assignment. Go to Settings > Voice Assistants, create a new pipeline, and select a different wake word from the openWakeWord dropdown. Point each pipeline to the same Whisper and Piper instances, or to separate ones if you want different voices per pipeline.

Satellite devices, such as ESP32-S3 boxes running the ESPHome voice satellite firmware, let you assign a pipeline per device. A bedroom satellite can listen for one phrase and route to a pipeline configured for low-light scene control, while a kitchen satellite uses a different phrase focused on timers and shopping lists.

Custom voice commands live in the conversation integration’s intent configuration. Create a file at config/custom_sentences/en/custom_intents.yaml and define your intents there:

language: "en"
intents:
  TurnOnWorkMode:
    data:
      - sentences:
          - "work mode on"
          - "start work mode"
          - "enable focus mode"

Wire each intent to an automation using the conversation_command trigger. The intent name becomes the trigger identifier, matched exactly in an automation condition. For slot-based commands where you want to pass a variable such as a room name or brightness level, use the list-based slot syntax in the sentence template and reference the slot value in the action. The Home Assistant documentation for custom sentences covers the full slot syntax and is worth reading before writing anything complex.

Conclusion

A local wake word setup in Home Assistant gives you voice control that is fast, private, and entirely under your control. The combination of openWakeWord, Faster-Whisper, and Piper covers all three stages of the pipeline without sending audio outside your network. Multi-wake word configuration lets you give different users or rooms distinct trigger phrases and separate command sets, which pays off quickly in households with mixed access needs. Custom intents move you beyond the built-in command vocabulary and let you trigger any automation you can write. Start with a single pipeline and one satellite device, get the configuration stable, and expand from there.

The end. Thanks for reading.

Keep reading.

Unleashing Local AI with Home Assistant: Ollama Integration 29 Apr · 8 min Home Assistant & LoRaWAN: Building Long-Range Sensor Networks 12 Oct · 12 min