Text-to-speech (voice output)
Set up local Piper voices or online services, install and test voices and manage the MuPiBox speech cache.
The box speaks: titles and categories when tapped, the time, the battery level, messages from the admin interface or Home Assistant. Under Text-to-speech you set up the voice.
The basic rule: one box = one language = one active voice. After a change the box creates its speech files again in the background.
Configuration
| Choice | Description |
|---|---|
| Local (Piper) | Voices run directly on the box, without internet and free of charge. Recommended. |
| Microsoft Azure | Official service with neural voices; your own Speech key is needed, a free F0 tier exists. The text is sent to Azure. |
| Google Cloud TTS | Official service; your own service account JSON and enabled billing are needed. |
| Google Translate (experimental) | Free and without credentials, uses an unofficial endpoint that can change at any time. |
Keys and JSON files are only stored locally and never shown again after saving.
- Background CPU: limits how much processing power preparing speech in the background may use, so playback stays smooth.
- Pre-rendering: prepares new or changed texts automatically. Off by default for online services because it uses up the free quota.
Voices
The voice catalogue lists available Piper voices per language. Many can be tried first with a short sample; the full model is only stored with "Install".
Test a voice
Enter a test text and listen – through the box's current audio output.
Cache
Spoken texts are stored as audio files and played immediately.
- Create missing: collects all visible content and only creates missing files.
- New generation: rebuilds the cache for the current language/voice (e.g. after changing the voice).
- Clean up: deletes outdated generations immediately.