Clean TTS for Sonoff iHost & Smart Home
Studio-Grade Neural Voice Announcements for Your Home Automations
Give your Sonoff iHost and smart home a natural, human-like voice. Clean TTS brings lifelike neural text-to-speech directly to the iHost’s built-in speaker and to any smart media player on your local network.
What Clean TTS Does
Natural Neural Speech: Sounds like a real person talking in your room, with natural intonation, breathing rhythm, and clear pronunciation across multiple languages (English, Romanian, German, French, Italian, Spanish, Catalan, etc.).- Speaks on iHost Internal Speaker: Plays announcements directly out of your Sonoff iHost built-in speaker — zero extra cables, amplifiers, or audio accessories needed.
Never Cuts Off Announcements: If multiple sensors trigger at the exact same moment (e.g. front door opens while a motion alert sounds), Clean TTS speaks every announcement clearly one after another instead of cutting words off or playing noisy overlapping sound.
Pleasant Wake-Up Chime: Softly alerts household members that an announcement is coming before speaking.
Smart Pronunciation Dictionary: Correctly pronounces smart home brand names and units instead of awkwardly spelling them out (e.g., “Zigbee” sounds like “zig-bi”, “Shelly” sounds like “she-lee”, “kWh” is read as “kilowatt hours”). Includes a live 
Play button to test words before saving.
Live Web Studio: Friendly dashboard at http://ihost.local:8123to test voices, adjust defaults, and test speech right from your browser.
AI Assistant Ready (MCP): Connects natively with AI assistants (Claude Desktop, Hermes Agent, Antigravity) so your AI agents can speak out loud on your iHost.
Quick 1-Minute Setup on Sonoff iHost
- In the eWeLink CUBE web interface, navigate to Docker → Images → Search:
tmalex/clean-tts - Download the
latestimage. - Run the container with:
- Network Mode: Select
host(do NOT usebridgemode; bridge causes connection errors with the internal speaker API). Inhostmode, the port is fixed to8123. - Volume Mount: Create a dedicated partition / folder specifically for Clean TTS (e.g.
clean-tts) and map it to/data. Do NOT mount it to another container’s directory (like Node-RED). This dedicated folder keeps your custom words, generated voices, and settings permanently safe across container updates.
- Open your browser at:
http://ihost.local:8123 - On your first visit, click “Request Access Token” and confirm Allow on your iHost screen to authorize physical speaker playback.
How Parameters Work (Override on the Fly)
In Clean TTS, you can configure your favorite default voice, volume, and speed in the web dashboard.
Whenever you send an announcement from Node-RED, Home Assistant, or an HTTP request, you can simply send the text, or you can optionally override any setting for that specific announcement:
| Parameter | Type | Example | What it does |
|---|---|---|---|
text |
string | "Front door opened" |
Required. The message to speak. |
lang |
string | "en", "ro", "de", "fr", "ca" |
Language code. Automatically picks the best native voice. |
voice |
string | "en-US-JennyNeural" |
Select a specific male or female voice. |
volume |
string | "+20%", "-15%" |
Volume boost or reduction for this specific alert (e.g. louder for alarms, quieter at night). |
rate |
string | "+10%", "-10%" |
Speech speed (faster or slower). |
pitch |
string | "+5Hz", "-5Hz" |
Voice pitch adjustment. |
bell |
boolean | true / false |
Enable or disable the notification chime before speaking. |
nocache |
boolean | true / false |
Set true for spontaneous one-off alerts (current time, live sensor readings) to skip disk cache. |
Tip: Any parameter you omit automatically falls back to your saved settings from the web dashboard!
Caches vs. No-Cache: Fixed alerts (“Front door opened”) are cached on disk for instant (0ms) replay. Spontaneous alerts (“It is 12:45 PM, 21.5°C”) can be sent with "nocache": true so 0 bytes are wasted on your storage drive. You can also run the container with -e NOCACHE=true to enforce streaming mode by default.
Node-RED Integration (Step-by-Step)
Method A: Speak on the Sonoff iHost Speaker (Recommended)
To make your Sonoff iHost speak an announcement out loud:
- Add an
http requestnode to your flow. - Configure the node:
- Method:
POST - URL:
http://ihost.local:8123/api/tts - Return:
a parsed JSON object
- In a
changeorfunctionnode just before the request, setmsg.payload:
msg.payload = {
"text": "Motion detected in the backyard garden.",
"bell": true,
"volume": "+20%"
};
return msg;
When triggered, your iHost speaker chimes and speaks the message clearly!
Method B: Play Night-Time Gentle Alert (Node-RED Example)
Speak quietly without a wake-up chime:
msg.payload = {
"text": "Living room lights turned off. Good night!",
"bell": false,
"volume": "-20%",
"rate": "-5%"
};
return msg;
Method C: Stream to External Speakers or Home Assistant
If you want to play announcements on external DLNA speakers, Sonos, or through Home Assistant’s media_player.play_media service, use a simple GET request:
http://ihost.local:8123/api/tts?text=Welcome+home&lang=en&bell=true
In Home Assistant:
service: media_player.play_media
target:
entity_id: media_player.living_room_speaker
data:
media_content_id: "http://ihost.local:8123/api/tts?text=Front+door+opened&bell=true"
media_content_type: "music"
Home Assistant Integration (Plug-and-Play MaryTTS)
Clean TTS emulates the native Home Assistant MaryTTS protocol out of the box. No HACS or custom integrations required!
1. Add to configuration.yaml:
tts:
- platform: marytts
host: "<clean-tts-host>" # IP or hostname of your Clean TTS server or Sonoff iHost
port: 8123
voice: "en-US-JennyNeural" # Alt: en-US-GuyNeural, en-US-AriaNeural, en-GB-SoniaNeural, en-GB-RyanNeural
codec: "WAVE_FILE" # Or "MP3"
cache: false # Optional: set false to prevent Home Assistant from caching dynamic speech locally
Restart Home Assistant to apply the configuration.
2. Use in Automations or Developer Tools:
action: tts.speak
target:
entity_id: tts.marytts
data:
media_player_entity_id: media_player.living_room_speaker
message: "Living room temperature is 22 degrees."
Or using the legacy tts.marytts_say service:
action: tts.marytts_say
target:
entity_id: media_player.living_room_speaker
data:
message: "Attention, the front door has been opened."
Clean TTS applies your phonetic dictionary corrections, loudness normalization, and chime bell automatically!
Quick cURL Examples for Testing
Speak on iHost Speaker:
curl -X POST http://ihost.local:8123/api/tts \
-H "Content-Type: application/json" \
-d '{
"text": "Washing machine cycle finished.",
"bell": true,
"volume": "+10%"
}'
Urgent Alert (Louder & Faster):
curl -X POST http://ihost.local:8123/api/tts \
-H "Content-Type: application/json" \
-d '{
"text": "Warning! Water leak detected in the bathroom.",
"bell": true,
"volume": "+40%",
"rate": "+15%"
}'
Web Dashboard Overview
Access http://ihost.local:8123 from any browser on your network:
- Studio (
/): Test any text, preview voices, and adjust global volume and speed sliders. - Phonetic Dictionary (
/dict): Teach Clean TTS how to pronounce tricky local words or smart home brands with instant audio preview. 
Backup & Restore (/backup): Download a full ZIP backup of your dictionaries and settings with 1 click.
Languages (/languages): Switch between English, Romanian, French, German, Spanish, Italian, and Catalan.- AI Agent Hub (
/mcp): Connect AI assistants like Claude Desktop to control speech hands-free.
Răspuns în Italiană:
