Voice Assistant & Smart Display¶

Overview¶

Self-hosted Echo Show replacement using a Samsung Galaxy Tab A8 as an always-on smart display with fully local voice assistant, Spotify control, and calling.

Architecture:

Tab A8 (Fully Kiosk Browser + HA Companion App)
  |
  | wake word detected on-device (microWakeWord)
  | audio sent to the server after wake word
  v
server (<YOUR_SERVER_IP>)
  ├── Home Assistant (host network, port 8123)
  ├── Wyoming Whisper (STT, port 10300)
  ├── Wyoming Piper (TTS, port 10200)
  └── Wyoming openWakeWord (port 10400, for future ESP32 satellites)

The HA Companion App (2026.3+) handles wake word detection on-device using microWakeWord. No audio leaves the tablet until after the wake word is detected. The openWakeWord container is for future ESP32 voice satellites only.

Phase 1: Voice Pipeline Containers¶

Directory Setup¶

sudo mkdir -p /opt/docker/wyoming/{whisper-data,piper-data,openwakeword-data,openwakeword-custom}
sudo chown -R $USER:$USER /opt/docker/wyoming

Compose File¶

/opt/docker/wyoming/compose.yaml

services:
  # Speech-to-Text (faster-whisper)
  whisper:
    container_name: whisper
    image: rhasspy/wyoming-whisper:3.8.1
    restart: unless-stopped
    ports:
      - "10300:10300"
    volumes:
      - ./whisper-data:/data
    command:
      - "--model"
      - "small-int8"   # ~1 s per command on NUC-class hardware; noticeably more accurate than tiny
      - "--language"
      - "en"
    dns:
      - <YOUR_DNS_SERVER_IP>

  # Text-to-Speech (piper)
  piper:
    container_name: piper
    image: rhasspy/wyoming-piper:2.5.2
    restart: unless-stopped
    ports:
      - "10200:10200"
    volumes:
      - ./piper-data:/data
    command:
      - "--voice"
      - "en_GB-alan-medium"   # more voices: https://rhasspy.github.io/piper-samples/
    dns:
      - <YOUR_DNS_SERVER_IP>

  # Wake word detection (for ESP32 satellites; the tablet detects its wake word on-device)
  openwakeword:
    container_name: openwakeword
    image: rhasspy/wyoming-openwakeword:2.1.0
    restart: unless-stopped
    ports:
      - "10400:10400"
    volumes:
      - ./openwakeword-data:/data
      - ./openwakeword-custom:/custom
    command:
      - "--preload-model"
      - "hey_jarvis"
      - "--custom-model-dir"
      - "/custom"
    dns:
      - <YOUR_DNS_SERVER_IP>
    # The Wyoming port answering is the useful signal; the image has bash but no curl/nc.
    healthcheck:
      test: ["CMD", "bash", "-c", "</dev/tcp/127.0.0.1/10400"]
      interval: 60s
      timeout: 5s
      retries: 3
      start_period: 60s

Deploy¶

cd /opt/docker/wyoming
docker compose up -d
docker compose logs -f

Models download on first boot (1-3 minutes). Wait for Ready in each container's logs.

Model Notes¶

• Whisper small-int8: ~1 second processing on a NUC-class Intel CPU, ~500MB RAM. Drop to tiny-int8 if too slow, bump to medium-int8 if accuracy is poor.
• Piper en_GB-alan-medium: the voice I use; en_US-lessac-medium is a good general US English voice. Preview others at rhasspy.github.io/piper-samples.
• openWakeWord: Built-in words include ok_nabu, hey_jarvis, hey_mycroft, alexa. Repeat --preload-model to preload several. Only used by ESP32 satellites, not the tablet.

To change a model, edit the command in compose.yaml and run docker compose up -d. The new model downloads automatically.

Phase 2: Connect to Home Assistant¶

Add Wyoming Integrations¶

Add three separate Wyoming integrations, one for each container. Use the server's LAN/VLAN IP (not localhost).

1. Settings > Devices & Services > + Add Integration > Wyoming Protocol
2. Host: <YOUR_SERVER_IP>, Port: 10300 (detected as Whisper STT)
3. Repeat: Host: <YOUR_SERVER_IP>, Port: 10200 (detected as Piper TTS)
4. Repeat: Host: <YOUR_SERVER_IP>, Port: 10400 (detected as openWakeWord)

Create Assist Pipeline¶

1. Settings > Voice Assistants > + Add Assistant
2. Name: Home (any name)
3. Speech-to-text: faster-whisper
4. Text-to-speech: piper
5. Wake word: hey_jarvis (or ok_nabu)
6. Conversation agent: Home Assistant (default)

Expose Entities¶

Assist can only control entities explicitly exposed to it.

1. Settings > Voice Assistants > Expose tab
2. Toggle on entities you want voice-controllable
3. Set clear aliases for anything with cryptic names
4. Assign entities to Areas (Settings > Areas & Zones) for room-based commands like "turn off the lights in the kitchen"

Phase 3: Tablet Setup¶

Factory Reset & Debloat¶

Factory reset the Tab A8 first. After initial Google sign-in, debloat via ADB before installing anything else.

From any machine with ADB (e.g., your desktop):

sudo apt install android-tools-adb

Enable Developer Options on tablet (Settings > About tablet > tap Build number 7 times), then enable USB debugging. Connect via USB.

adb devices   # approve prompt on tablet

Remove Samsung bloatware:

adb shell pm uninstall -k --user 0 com.samsung.android.app.spage
adb shell pm uninstall -k --user 0 com.samsung.android.bixby.agent
adb shell pm uninstall -k --user 0 com.samsung.android.bixby.service
adb shell pm uninstall -k --user 0 com.samsung.android.visionintelligence
adb shell pm uninstall -k --user 0 com.samsung.android.game.gamehome
adb shell pm uninstall -k --user 0 com.samsung.android.game.gametools
adb shell pm uninstall -k --user 0 com.samsung.android.app.tips
adb shell pm uninstall -k --user 0 com.samsung.android.mobileservice
adb shell pm uninstall -k --user 0 com.samsung.android.app.watchmanagerstub
adb shell pm uninstall -k --user 0 com.samsung.android.ardrawing
adb shell pm uninstall -k --user 0 com.samsung.android.aremoji
adb shell pm uninstall -k --user 0 com.samsung.android.arzone
adb shell pm uninstall -k --user 0 com.samsung.android.app.dressroom
adb shell pm uninstall -k --user 0 com.samsung.android.spay
adb shell pm uninstall -k --user 0 com.samsung.android.spayfw
adb shell pm uninstall -k --user 0 com.samsung.android.app.sbrowseredge
adb shell pm uninstall -k --user 0 com.samsung.android.samsungpass
adb shell pm uninstall -k --user 0 com.samsung.android.samsungpassautofill
adb shell pm uninstall -k --user 0 com.sec.android.app.samsungapps
adb shell pm uninstall -k --user 0 com.samsung.android.themestore
adb shell pm uninstall -k --user 0 com.samsung.android.app.reminder
adb shell pm uninstall -k --user 0 com.samsung.android.calendar
adb shell pm uninstall -k --user 0 com.samsung.android.email.provider
adb shell pm uninstall -k --user 0 com.samsung.android.app.notes
adb shell pm uninstall -k --user 0 com.microsoft.office.officehubrow
adb shell pm uninstall -k --user 0 com.microsoft.skydrive
adb shell pm uninstall -k --user 0 com.microsoft.appmanager
adb shell pm uninstall -k --user 0 com.linkedin.android
adb shell pm uninstall -k --user 0 com.facebook.katana
adb shell pm uninstall -k --user 0 com.facebook.appmanager
adb shell pm uninstall -k --user 0 com.facebook.system
adb shell pm uninstall -k --user 0 com.netflix.mediaclient
adb shell pm uninstall -k --user 0 com.samsung.android.forest
adb shell pm uninstall -k --user 0 com.samsung.android.app.social

To find anything else: adb shell pm list packages -f | grep samsung

To restore if needed: adb shell cmd package install-existing <package.name>

Install HA Companion App¶

1. Install from Play Store
2. Open, enter HA URL: http://homeassistant.<YOUR_LOCAL_DOMAIN>:8123 (or direct IP)
3. Grant all permissions (microphone, camera, location, notifications)
4. Configure wake word: Settings > Companion app > Assist for Android
5. Tap Set as default, set Home Assistant as default digital assistant
6. Enable Wake word detection, select "Hey Jarvis"
7. Select the Assist pipeline you created

Install Fully Kiosk Browser¶

1. Install from Play Store (~$9 for Plus license to remove watermark)
2. Open, enter start URL: http://homeassistant.<YOUR_LOCAL_DOMAIN>:8123/tablet/0
3. Swipe from left edge to open settings menu
4. Configure:
   • Web Content Settings > Autoplay Videos: ON
   • Web Content Settings > Enable Microphone Access: OFF (critical -- prevents mic conflict with Companion App)
   • Device Management > Keep Screen On: ON
   • Motion Detection > Enable: ON (use camera, not microphone)
   • Motion Detection > Turn Screen On on Motion: ON
   • Remote Administration > Enable: ON, set a password
   • Kiosk Mode > Enable Kiosk Mode: ON
   • Kiosk Mode > Kiosk Exit Gesture: set a gesture you'll remember
   • Kiosk Mode > App Whitelist: add one per line:

     io.homeassistant.companion.android
     com.google.android.apps.googlevoice
     com.facebook.orca
     

   • Kiosk Mode > Applications to Run on Start in Background:

     io.homeassistant.companion.android
     

Add Fully Kiosk Integration to HA¶

1. Settings > Devices & Services > + Add Integration > Fully Kiosk Browser
2. Enter the tablet's IP and Remote Admin password
3. Uncheck "Verify SSL"

Phase 4: Dashboard¶

Create a dedicated tablet dashboard: Settings > Dashboards > + Add Dashboard, name "Tablet", URL tablet.

Greeting Card¶

type: markdown
content: >
  {% set hour = now().hour %}
  {% if hour < 12 %}
  <h1 style="font-size: 2.5em; font-weight: 300; margin: 0;">Good morning</h1>
  {% elif hour < 17 %}
  <h1 style="font-size: 2.5em; font-weight: 300; margin: 0;">Good afternoon</h1>
  {% else %}
  <h1 style="font-size: 2.5em; font-weight: 300; margin: 0;">Good evening</h1>
  {% endif %}
  <h3 style="font-size: 1.2em; font-weight: 300; opacity: 0.6; margin: 0;">{{ now().strftime('%A, %B %-d') }}</h3>

Weather¶

Add a Weather Forecast card. If no weather integration exists: Settings > Integrations > + Add > Met.no (no API key needed).

Timer¶

Create a timer helper: Settings > Devices & Services > Helpers > + Create Helper > Timer. Name it, set a default duration.

Add to dashboard:

type: tile
entity: timer.timer

Tap the card to start. Voice command "Hey Jarvis, set a timer for 5 minutes" works natively via Assist (separate from this entity).

Media / Spotify¶

See Phase 5. Once configured, add:

type: media-control
entity: media_player.spotify_YOUR_USERNAME

Call Buttons¶

See Phase 6. Add Button cards for each contact.

Phase 5: Spotify¶

Official Integration¶

1. Go to developer.spotify.com/dashboard, log in with Spotify Premium
2. Create App:
   • Redirect URI: https://my.home-assistant.io/redirect/oauth (only if you disabled My Home Assistant: http://homeassistant.<YOUR_LOCAL_DOMAIN>:8123/auth/external/callback)
   • Select "Web API"
3. Copy Client ID and Client Secret
4. In HA: Settings > Integrations > + Add > Spotify
5. Enter credentials, authenticate

Spotcast (for voice-triggered playback)¶

The official integration cannot start playback on idle devices. Spotcast works around this.

1. Install HACS if not already installed (hacs.xyz)
2. HACS > Integrations > search "Spotcast" > install
3. Get Spotify cookies:
   • Open incognito window, go to open.spotify.com, log in
   • DevTools (F12) > Application > Cookies > open.spotify.com
   • Copy sp_dc and sp_key values
   • Close window WITHOUT logging out
4. Add to configuration.yaml:

spotcast:
  sp_dc: "your_sp_dc_value"
  sp_key: "your_sp_key_value"

1. Restart HA

Tablet as Spotify Connect Device¶

Install the Spotify app on the Tab A8 and log in. The tablet appears as a Spotify Connect target for playback through its speakers.

Phase 6: Calling¶

Apps on Tablet¶

• Google Voice: PSTN calls (real phone numbers). Set up at voice.google.com. In app settings, set calls to "Prefer Wi-Fi and mobile data".
• Facebook Messenger: Video calls to contacts on Messenger.

Both apps are whitelisted in Fully Kiosk (done in Phase 3). Incoming calls overlay the kiosk, and Fully Kiosk returns to the dashboard after the call ends.

HA Scripts¶

Create scripts in Settings > Automations & Scenes > Scripts > + Add Script.

Phone calls go through the Companion App's command_activity notification command rather than Fully Kiosk. It fires Android's ACTION_CALL intent, which dials the number straight away in the default dialer (Google Voice) instead of just opening the dialer with the number filled in. Messenger and the dialer app itself are opened with Fully Kiosk's start_application.

Call Mom (Google Voice):

alias: "Call Mom"
sequence:
  - action: notify.mobile_app_<TABLET>
    data:
      message: "command_activity"
      data:
        intent_action: "android.intent.action.CALL"
        intent_uri: "tel:+15551234567"

Call Dad (Google Voice): the same with tel:+15559876543.

Call Grandpa (Messenger):

alias: "Call Grandpa"
sequence:
  - action: fully_kiosk.start_application
    target:
      device_id: YOUR_TABLET_DEVICE_ID
    data:
      application: "com.facebook.orca"

notify.mobile_app_<TABLET> is the tablet's Companion App notify service (Developer Tools > Actions, search notify.mobile_app). Find YOUR_TABLET_DEVICE_ID by creating the script via the visual editor and selecting the tablet as the target device.

Dashboard Buttons¶

type: button
name: "Call Mom"
icon: mdi:phone
icon_height: 80px
tap_action:
  action: perform-action
  perform_action: script.call_mom
show_state: false

type: button
name: "Call Grandpa"
icon: mdi:video
icon_height: 80px
tap_action:
  action: perform-action
  perform_action: script.call_grandpa
show_state: false

Use mdi:phone for voice calls, mdi:video for Messenger video calls.

Voice-Triggered Calling¶

Create config/custom_sentences/en/call_contacts.yaml:

language: "en"
intents:
  CallMom:
    data:
      - sentences:
          - "call mom"
          - "call mommy"
          - "phone mom"
  CallDad:
    data:
      - sentences:
          - "call dad"
          - "call daddy"
          - "phone dad"
  CallGrandpa:
    data:
      - sentences:
          - "call grandpa"
          - "phone grandpa"
  VideoCallGrandpa:
    data:
      - sentences:
          - "video call grandpa"
  OpenPhone:
    data:
      - sentences:
          - "open the phone"
          - "open dialer"
          - "open google voice"

Add to configuration.yaml:

intent_script:
  CallMom:
    speech:
      text: "Calling mom"
    action:
      - action: notify.mobile_app_<TABLET>
        data:
          message: "command_activity"
          data:
            intent_action: "android.intent.action.CALL"
            intent_uri: "tel:+15551234567"
  CallDad:
    speech:
      text: "Calling dad"
    action:
      - action: notify.mobile_app_<TABLET>
        data:
          message: "command_activity"
          data:
            intent_action: "android.intent.action.CALL"
            intent_uri: "tel:+15559876543"
  VideoCallGrandpa:
    speech:
      text: "Opening messenger for grandpa"
    action:
      - action: fully_kiosk.start_application
        target:
          device_id: YOUR_TABLET_DEVICE_ID
        data:
          application: "com.facebook.orca"
  OpenPhone:
    speech:
      text: "Opening the phone"
    action:
      - action: fully_kiosk.start_application
        target:
          device_id: YOUR_TABLET_DEVICE_ID
        data:
          application: "com.google.android.apps.googlevoice"
  # CallGrandpa: same pattern as CallMom with grandpa's number

Restart HA. "Hey Jarvis, call mom" now works.

HACS Recommendations¶

• card-mod: Custom CSS styling on any card (remove backgrounds, etc.)
• Mushroom Cards: Clean, modern, touch-friendly cards
• Mini Media Player: Better Spotify/media card than default
• Browser Mod: HA control of the tablet's browser (pop-up camera feeds, navigate views)
• Spotcast: Voice-triggered Spotify playback (covered in Phase 5)

Future: ESP32 Voice Satellites¶

Adding an M5Stack ATOM Echo (~$13) or ESP32-S3-BOX as a dedicated voice satellite:

• Satellite streams audio to HA
• openWakeWord on the server handles wake word detection
• Whisper transcribes, Piper responds
• Audio plays through the satellite speaker

The Wyoming infrastructure deployed here supports this with no changes.

Smart Switches¶

For integrating existing lights with HA/voice, replace dumb switches with Zigbee smart switches.

Recommended: Inovelli Blue Series (VZM31-SN)

• Zigbee 3.0, full Zigbee2MQTT support
• Works with and without neutral wire (important for older homes)
• Dimmer + on/off in one device
• Multi-tap scene control for HA automations
• LED notification bar
• 3-way compatible
• ~$40/switch

Budget alternative: Sonoff ZBMINIL2 (~$10) -- no-neutral relay that fits behind existing switches. Keeps your current switch appearance.

Sources & further reading¶