Smart Voice Control

Voice Local Privacy

With Home Assistant Assist you can control your smart home with your voice - 100% locally without cloud! No data leaves your home, and it works even when the internet is down.

Recommendations 2025

Voice Satellites (hardware):

• M5Stack ATOM Echo - Cheapest (~$13), small, good sound
• ESP32-S3-BOX-3 - Best display, touchscreen (~$50)
• Home Assistant Voice PE - Official HA hardware (~$59)

Software Stack:

• Whisper - Speech-to-text (OpenAI, local)
• Piper - Text-to-speech (local, fast)
• openWakeWord - Wake word detection

Recommendation: Start with ATOM Echo + Whisper/Piper!

---

🎤 Voice Control Overview

Assist Pipeline Components

Component	Function	Local Options
Wake Word	Listens for activation word	openWakeWord, microWakeWord
STT	Speech to text	Whisper, Speech-to-Phrase
Intent	Understands commands	HA Conversation, LLM
TTS	Text to speech	Piper, Home Assistant Cloud
Satellite	Microphone/speaker	ESP32, Voice PE

Hardware Comparison

Device	Price	Display	Microphone	Speaker	Wake Word
ATOM Echo	~$13	❌	✅	✅ (small)	On-device
S3-BOX-3	~$50	✅ Touch	✅✅	✅	On-device
Voice PE	~$59	✅	✅✅	✅✅	On-device
CoreS3SE	~$70	✅ Touch	✅✅	✅	On-device

---

🔧 Local Voice Setup

1. Install Add-ons

Whisper (Speech-to-Text)

# Settings → Add-ons → Add-on Store
# Search "Whisper" → Install

# Configuration (Settings tab):
model: small  # tiny, base, small, medium, large
language: en  # Your language

# Start add-on
# Wait for model download (can take time)

Note

Model sizes:

• tiny - Fastest, lowest quality
• small - Good balance (recommended)
• medium - Better quality, slower
• large - Best quality, requires lots of RAM

Piper (Text-to-Speech)

# Settings → Add-ons → Add-on Store
# Search "Piper" → Install

# Configuration:
voice: en_US-lessac-medium  # English voice

# Start add-on

openWakeWord (Wake Word)

# Settings → Add-ons → Add-on Store
# Search "openWakeWord" → Install → Start

# Supported wake words:
# - "Ok Nabu"
# - "Hey Jarvis"
# - "Alexa"
# - "Hey Mycroft"

2. Wyoming Integration

Add Integrations

# Settings → Devices & Services → Add Integration

# 1. Search "Wyoming Protocol"
# - Whisper should auto-discover
# - Click "Configure" → "Submit"

# 2. Piper should also auto-discover
# - Click "Configure" → "Submit"

# 3. openWakeWord likewise
# - Click "Configure" → "Submit"

Tip

If they don’t auto-discover, add manually:

• Host: core-whisper / core-piper / core-openwakeword
• Port: 10300 / 10200 / 10400

3. Assist Pipeline

Create Voice Assistant

# Settings → Voice Assistants → Add Assistant

# Name: "Local Assistant"
# Language: English

# Conversation agent: Home Assistant
# Speech-to-text: Whisper
# Text-to-speech: Piper
# Wake word: openWakeWord (choose wake word)

# Save

Test Pipeline

# Click "Try pipeline" button
# Say: "Turn on the living room light"

# Check:
# - Was speech recognized correctly?
# - Was command understood?
# - Did you hear the response?

---

📱 Voice Satellites

M5Stack ATOM Echo

Price: ~$13

The cheapest voice satellite:

• ✅ ESP32 based
• ✅ Built-in microphone + speaker
• ✅ On-device wake word (microWakeWord)
• ✅ LED status indicator
• ✅ Easy web setup
• ⚠️ Small speaker (low volume)
• ❌ No display

Installation

# 1. Go to: https://www.home-assistant.io/voice_control/thirteen-usd-voice-remote/
# 2. Click "Connect" in Chrome/Edge
# 3. Select COM port
# 4. Click "Install Voice Assistant"
# 5. Enter WiFi credentials
# 6. Device appears in HA

ESPHome Config (advanced)

substitutions:
  name: living-room-voice
  friendly_name: "Living Room Voice Assistant"
  micro_wake_word_model: hey_jarvis

packages:
  m5stack.atom-echo:
    url: https://github.com/esphome/firmware
    files:
      - voice-assistant/m5stack-atom-echo.yaml
    refresh: 0s

esphome:
  name: ${name}
  friendly_name: ${friendly_name}

wifi:
  ssid: !secret wifi_ssid
  password: !secret wifi_password

Buy: M5Stack, AliExpress

ESP32-S3-BOX-3

Price: ~$50

Best all-round voice satellite:

• ✅ 2.4” touchscreen display
• ✅ Dual microphones (better recognition)
• ✅ Good speaker
• ✅ On-device wake word
• ✅ Official ESPHome support
• ✅ Includes dock/stand
• ⚠️ Requires USB-C power

Installation

# 1. ESPHome → New Device
# 2. Name: "bedroom-voice"
# 3. Device: ESP32-S3-BOX-3
# 4. Add to YAML:

packages:
  esphome.voice-assistant:
    url: https://github.com/esphome/firmware
    files:
      - voice-assistant/esp32-s3-box-3.yaml
    refresh: 0s

substitutions:
  micro_wake_word_model: okay_nabu

esphome:
  name: bedroom-voice
  friendly_name: "Bedroom Voice"

wifi:
  ssid: !secret wifi_ssid
  password: !secret wifi_password

api:
  encryption:
    key: !secret api_key

Buy: Espressif, AliExpress, Amazon

Home Assistant Voice PE

Price: ~$59

Official Home Assistant hardware:

• ✅ Designed specifically for HA
• ✅ Premium microphones
• ✅ Good speaker
• ✅ E-ink display
• ✅ Physical mute button
• ✅ Pre-installed firmware
• ✅ Official support

Setup

# 1. Connect USB-C power
# 2. Follow on-screen instructions
# 3. Scan QR code with HA app
# 4. Select WiFi network
# 5. Assign to room
# 6. Done!

Best User Experience

Voice PE is the easiest solution - plug and play!

Buy: Home Assistant Store

---

🗣️ Wake Words

Available Wake Words

Wake Word	Language	Model
”Ok Nabu”	Multi	openWakeWord, microWakeWord
”Hey Jarvis”	English	microWakeWord
”Alexa”	Multi	microWakeWord
”Hey Mycroft”	English	microWakeWord

Create Custom Wake Word

# 1. Go to: https://www.home-assistant.io/voice_control/create_wake_word/

# 2. Choose a unique word (3-4 syllables)
# - Avoid common words
# - Only English supported currently

# 3. Generate training data with Piper
# 4. Train model (may take several attempts)
# 5. Download and install

---

🤖 Automations with Voice

automation:
  # Broadcast to all speakers
  - alias: "Voice - Good Morning Broadcast"
    trigger:
      - platform: time
        at: "07:00:00"
    condition:
      - condition: state
        entity_id: binary_sensor.workday
        state: "on"
    action:
      - service: tts.speak
        target:
          entity_id: tts.piper
        data:
          media_player_entity_id:
            - media_player.living_room_speaker
            - media_player.kitchen_speaker
          message: >
            Good morning! It's 7 o'clock.
            The temperature outside is {{ states('sensor.outdoor_temperature') }} degrees.
            {% if states('sensor.rain_probability') | int > 50 %}
            Remember your umbrella - there's a chance of rain.
            {% endif %}

  # Voice reminder
  - alias: "Voice - Washing Machine Done"
    trigger:
      - platform: state
        entity_id: binary_sensor.washing_machine_running
        to: "off"
    action:
      - service: tts.speak
        target:
          entity_id: tts.piper
        data:
          media_player_entity_id: media_player.kitchen_speaker
          message: "The washing machine is done. Don't forget to empty it."

  # Welcome home
  - alias: "Voice - Welcome Home"
    trigger:
      - platform: state
        entity_id: person.brian
        to: "home"
    action:
      - delay: "00:00:30"
      - service: tts.speak
        target:
          entity_id: tts.piper
        data:
          media_player_entity_id: media_player.hallway_speaker
          message: >
            Welcome home!
            It's {{ states('sensor.indoor_temperature') }} degrees inside.

---

🎯 Voice Commands

Supported Commands

# Lights
"Turn on the living room light"
"Turn off all lights"
"Set the kitchen brightness to 50 percent"
"Change the bedroom color to blue"

# Climate
"What's the temperature?"
"Set the thermostat to 72 degrees"
"Turn on the bathroom heater"

# Devices
"Turn on the TV"
"Start the vacuum"
"Lock the front door"

# Information
"What's the weather today?"
"When does the sun set?"
"Is anyone home?"

# Scenes
"Activate movie night"
"Good night"
"I'm leaving home"

Customize Sentences

# configuration.yaml or via UI

intent_script:
  CustomWelcome:
    speech:
      text: "Welcome! What can I help with?"

conversation:
  intents:
    CustomWelcome:
      - "hey [assistant]"
      - "hello"
      - "what can you do"

---

📊 Dashboard

type: vertical-stack
cards:
  # Voice status
  - type: entities
    title: "🎤 Voice Control"
    entities:
      - entity: assist_satellite.living_room_voice
        name: "Living Room Satellite"
      - entity: assist_satellite.bedroom_voice
        name: "Bedroom Satellite"
      - entity: binary_sensor.whisper_running
        name: "Whisper Status"
      - entity: binary_sensor.piper_running
        name: "Piper Status"

  # Test button
  - type: button
    name: "Test Voice"
    tap_action:
      action: call-service
      service: tts.speak
      target:
        entity_id: tts.piper
      data:
        media_player_entity_id: media_player.living_room_speaker
        message: "Voice control is working!"

---

❓ Frequently Asked Questions

Ofte stillede spørgsmål

How much RAM does local voice require?

Whisper 'small' model uses about 2GB RAM. 'Medium' uses 5GB. Raspberry Pi 4 with 4GB can run 'small'. For better performance a mini-PC or dedicated server is recommended.

Can I use multiple wake words?

Yes! Since HA 2025.10 you can have up to 2 wake words per satellite. You can use 'Ok Nabu' for one language and 'Hey Jarvis' for another pipeline.

Why is the response slow?

Usually the Whisper model is too large for your hardware. Try 'tiny' or 'small' model. Also check that you're running on a fast storage device (SSD, not SD card).

Can I use this with LLMs like ChatGPT?

Yes! You can configure your Assist pipeline to use OpenAI, Ollama (local), or other LLMs as the conversation agent for more natural interactions.

What's the difference between Whisper and Speech-to-Phrase?

Whisper is a general speech-to-text model that understands everything. Speech-to-Phrase is optimized for smart home commands and is faster/lighter, but more limited.

---

📚 Next Steps

ESP32 Projects

Build more ESPHome devices.

See guide →

Automations

Advanced automations.

See guide →

---

Last updated: December 2025