[Integration] ElevenLabs: a natural voice for Gladys announcements

Development
1

Hello everyone,

Gladys 5.3 now allows external integrations to provide speech synthesis. I took advantage of this to create an ElevenLabs integration. Your scene announcements (« The washing machine has finished », « Someone is at the door »…) are read aloud on your speakers with the ElevenLabs voice of your choice, without a Gladys Plus subscription.

What it does

  • All the voices in your ElevenLabs account, by name or ID, including those added from the voice library. A « List my voices » button displays them all, and the voice used is checked.
  • Model choice: Flash v2.5 (fast, half the credits), Multilingual v2 (most natural) or v3 (most expressive). The speed and stability of the voice are adjustable.
  • Credit-efficient: a phrase already spoken with the same voice is replayed from memory, and an announcement sent to multiple speakers is synthesized only once.
  • Clear connection status: verified key and voice, remaining credits and renewal date. For example: « Connected to ElevenLabs, voice « George ». 8,500 credits remaining out of 10,000, renewed on November 12 ».
  • Compatible with Sonos, Google Cast and AirPlay speakers, via the scene action « Speak on a speaker ».

image
image800×534 53.1 KB

Installation

  1. You need Gladys 5.3.0 or newer, installed in host network mode (standard installation).
  2. Create an API key on elevenlabs.io. The free plan is enough to try (10,000 credits per month).
  3. In Gladys, install ElevenLabs from the integrations store, then paste the key and choose a voice.
  4. On the Artificial Intelligence page, choose ElevenLabs in the Voice card.

After that, any scene with the Speak on a speaker action uses the new voice.

Good to know

  • Each announcement consumes approximately one ElevenLabs credit per character (half with Flash).
  • The text of the announcements is sent to ElevenLabs :cloud:. The key is stored as a secret by Gladys and is never displayed again.
  • Limitation: in the browser or app, the voice assistant responds in text. Gladys serves the audio in http:// on the local network, and an https:// page refuses to play it. Speakers are not affected.

The code is open source (Apache-2.0): GitHub - cicoub13/gladys-elevenlabs · GitHub. The complete documentation, including a troubleshooting section, is displayed in the integrations catalog.

Your feedback is welcome: the voices you prefer, bugs, ideas for improvement. Here or in the GitHub issues!

2

Great integration, their v4 + v4 turbo models were recently released. Do you think we could also have access to them in the integration?

3

Oh yes, surprised that Claude didn’t put them. I’ll check

4

Disponible en v1.1.0